You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/db/db_impl/db_impl.cc

5243 lines
185 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "db/db_impl/db_impl.h"
#include <stdint.h>
#ifdef OS_SOLARIS
#include <alloca.h>
#endif
#include <algorithm>
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
#include <cinttypes>
#include <cstdio>
#include <map>
#include <set>
#include <sstream>
#include <stdexcept>
#include <string>
#include <unordered_map>
#include <utility>
#include <vector>
#include "db/arena_wrapped_db_iter.h"
#include "db/builder.h"
#include "db/compaction/compaction_job.h"
#include "db/db_info_dumper.h"
#include "db/db_iter.h"
#include "db/dbformat.h"
#include "db/error_handler.h"
#include "db/event_helpers.h"
#include "db/external_sst_file_ingestion_job.h"
#include "db/flush_job.h"
#include "db/forward_iterator.h"
#include "db/import_column_family_job.h"
#include "db/job_context.h"
#include "db/log_reader.h"
#include "db/log_writer.h"
db: avoid `#include`ing malloc and jemalloc simultaneously Summary: This fixes a compilation failure on Linux when the system libc is not glibc. jemalloc's configure script incorrectly assumes that glibc is always used on Linux systems, producing glibc-style signatures; when the system libc is e.g. musl, the following error is observed: ``` [ 0%] Building CXX object CMakeFiles/rocksdb.dir/db/db_impl.cc.o In file included from /go/src/github.com/cockroachdb/cockroach/c-deps/rocksdb.src/table/block.h:19:0, from /go/src/github.com/cockroachdb/cockroach/c-deps/rocksdb.src/db/db_impl.cc:77: /x-tools/x86_64-unknown-linux-musl/x86_64-unknown-linux-musl/sysroot/usr/include/malloc.h:19:8: error: declaration of 'size_t malloc_usable_size(void*)' has a different exception specifier size_t malloc_usable_size(void *); ^~~~~~~~~~~~~~~~~~ In file included from /go/src/github.com/cockroachdb/cockroach/c-deps/rocksdb.src/db/db_impl.cc:20:0: /go/native/x86_64-unknown-linux-musl/jemalloc/include/jemalloc/jemalloc.h:78:33: note: from previous declaration 'size_t malloc_usable_size(void*) throw ()' # define je_malloc_usable_size malloc_usable_size ^ /go/native/x86_64-unknown-linux-musl/jemalloc/include/jemalloc/jemalloc.h:239:41: note: in expansion of macro 'je_malloc_usable_size' JEMALLOC_EXPORT size_t JEMALLOC_NOTHROW je_malloc_usable_size( ^~~~~~~~~~~~~~~~~~~~~ CMakeFiles/rocksdb.dir/build.make:350: recipe for target 'CMakeFiles/rocksdb.dir/db/db_impl.cc.o' failed ``` This works around the issue by rearranging the sources such that jemalloc's headers are never in the same scope as the system's malloc header. The jemalloc issue has been reported as well, see: https://github.com/jemalloc/jemalloc/issues/778. cc tschottdorf Closes https://github.com/facebook/rocksdb/pull/2188 Differential Revision: D5163048 Pulled By: siying fbshipit-source-id: c553125458892def175c1be5682b0330d80b2a0d
8 years ago
#include "db/malloc_stats.h"
#include "db/memtable.h"
#include "db/memtable_list.h"
#include "db/merge_context.h"
#include "db/merge_helper.h"
#include "db/periodic_work_scheduler.h"
#include "db/range_tombstone_fragmenter.h"
#include "db/table_cache.h"
#include "db/table_properties_collector.h"
#include "db/transaction_log_impl.h"
#include "db/version_set.h"
#include "db/write_batch_internal.h"
#include "db/write_callback.h"
Experimental support for SST unique IDs (#8990) Summary: * New public header unique_id.h and function GetUniqueIdFromTableProperties which computes a universally unique identifier based on table properties of table files from recent RocksDB versions. * Generation of DB session IDs is refactored so that they are guaranteed unique in the lifetime of a process running RocksDB. (SemiStructuredUniqueIdGen, new test included.) Along with file numbers, this enables SST unique IDs to be guaranteed unique among SSTs generated in a single process, and "better than random" between processes. See https://github.com/pdillinger/unique_id * In addition to public API producing 'external' unique IDs, there is a function for producing 'internal' unique IDs, with functions for converting between the two. In short, the external ID is "safe" for things people might do with it, and the internal ID enables more "power user" features for the future. Specifically, the external ID goes through a hashing layer so that any subset of bits in the external ID can be used as a hash of the full ID, while also preserving uniqueness guarantees in the first 128 bits (bijective both on first 128 bits and on full 192 bits). Intended follow-up: * Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into the third 64-bit value of the unique ID.) * Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968) Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990 Test Plan: Unit tests added, and checking of unique ids in stress test. NOTE in stress test we do not generate nearly enough files to thoroughly stress uniqueness, but the test trims off pieces of the ID to check for uniqueness so that we can infer (with some assumptions) stronger properties in the aggregate. Reviewed By: zhichao-cao, mrambacher Differential Revision: D31582865 Pulled By: pdillinger fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
#include "env/unique_id_gen.h"
#include "file/file_util.h"
#include "file/filename.h"
#include "file/random_access_file_reader.h"
#include "file/sst_file_manager_impl.h"
#include "logging/auto_roll_logger.h"
#include "logging/log_buffer.h"
#include "logging/logging.h"
#include "monitoring/in_memory_stats_history.h"
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
#include "monitoring/instrumented_mutex.h"
#include "monitoring/iostats_context_imp.h"
#include "monitoring/perf_context_imp.h"
#include "monitoring/persistent_stats_history.h"
#include "monitoring/thread_status_updater.h"
#include "monitoring/thread_status_util.h"
#include "options/cf_options.h"
#include "options/options_helper.h"
#include "options/options_parser.h"
#include "port/port.h"
#include "rocksdb/cache.h"
#include "rocksdb/compaction_filter.h"
#include "rocksdb/convenience.h"
#include "rocksdb/db.h"
#include "rocksdb/env.h"
#include "rocksdb/merge_operator.h"
#include "rocksdb/statistics.h"
#include "rocksdb/stats_history.h"
#include "rocksdb/status.h"
#include "rocksdb/table.h"
#include "rocksdb/version.h"
#include "rocksdb/write_buffer_manager.h"
#include "table/block_based/block.h"
#include "table/block_based/block_based_table_factory.h"
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
#include "table/get_context.h"
#include "table/merging_iterator.h"
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
#include "table/multiget_context.h"
#include "table/sst_file_dumper.h"
#include "table/table_builder.h"
#include "table/two_level_iterator.h"
Experimental support for SST unique IDs (#8990) Summary: * New public header unique_id.h and function GetUniqueIdFromTableProperties which computes a universally unique identifier based on table properties of table files from recent RocksDB versions. * Generation of DB session IDs is refactored so that they are guaranteed unique in the lifetime of a process running RocksDB. (SemiStructuredUniqueIdGen, new test included.) Along with file numbers, this enables SST unique IDs to be guaranteed unique among SSTs generated in a single process, and "better than random" between processes. See https://github.com/pdillinger/unique_id * In addition to public API producing 'external' unique IDs, there is a function for producing 'internal' unique IDs, with functions for converting between the two. In short, the external ID is "safe" for things people might do with it, and the internal ID enables more "power user" features for the future. Specifically, the external ID goes through a hashing layer so that any subset of bits in the external ID can be used as a hash of the full ID, while also preserving uniqueness guarantees in the first 128 bits (bijective both on first 128 bits and on full 192 bits). Intended follow-up: * Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into the third 64-bit value of the unique ID.) * Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968) Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990 Test Plan: Unit tests added, and checking of unique ids in stress test. NOTE in stress test we do not generate nearly enough files to thoroughly stress uniqueness, but the test trims off pieces of the ID to check for uniqueness so that we can infer (with some assumptions) stronger properties in the aggregate. Reviewed By: zhichao-cao, mrambacher Differential Revision: D31582865 Pulled By: pdillinger fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
#include "table/unique_id_impl.h"
#include "test_util/sync_point.h"
#include "trace_replay/trace_replay.h"
#include "util/autovector.h"
#include "util/cast_util.h"
#include "util/coding.h"
#include "util/compression.h"
#include "util/crc32c.h"
#include "util/defer.h"
#include "util/mutexlock.h"
#include "util/stop_watch.h"
#include "util/string_util.h"
#include "utilities/trace/replayer_impl.h"
namespace ROCKSDB_NAMESPACE {
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
const std::string kDefaultColumnFamilyName("default");
const std::string kPersistentStatsColumnFamilyName(
"___rocksdb_stats_history___");
void DumpRocksDBBuildVersion(Logger* log);
CompressionType GetCompressionFlush(
const ImmutableCFOptions& ioptions,
const MutableCFOptions& mutable_cf_options) {
// Compressing memtable flushes might not help unless the sequential load
// optimization is used for leveled compaction. Otherwise the CPU and
// latency overhead is not offset by saving much space.
if (ioptions.compaction_style == kCompactionStyleUniversal) {
if (mutable_cf_options.compaction_options_universal
.compression_size_percent < 0) {
return mutable_cf_options.compression;
} else {
return kNoCompression;
}
} else if (!ioptions.compression_per_level.empty()) {
// For leveled compress when min_level_to_compress != 0.
return ioptions.compression_per_level[0];
} else {
return mutable_cf_options.compression;
}
}
namespace {
void DumpSupportInfo(Logger* logger) {
ROCKS_LOG_HEADER(logger, "Compression algorithms supported:");
for (auto& compression : OptionsHelper::compression_type_string_map) {
if (compression.second != kNoCompression &&
compression.second != kDisableCompressionOption) {
ROCKS_LOG_HEADER(logger, "\t%s supported: %d", compression.first.c_str(),
CompressionTypeSupported(compression.second));
}
}
ROCKS_LOG_HEADER(logger, "Fast CRC32 supported: %s",
crc32c::IsFastCrc32Supported().c_str());
}
} // namespace
DBImpl::DBImpl(const DBOptions& options, const std::string& dbname,
Make backups openable as read-only DBs (#8142) Summary: A current limitation of backups is that you don't know the exact database state of when the backup was taken. With this new feature, you can at least inspect the backup's DB state without restoring it by opening it as a read-only DB. Rather than add something like OpenAsReadOnlyDB to the BackupEngine API, which would inhibit opening stackable DB implementations read-only (if/when their APIs support it), we instead provide a DB name and Env that can be used to open as a read-only DB. Possible follow-up work: * Add a version of GetBackupInfo for a single backup. * Let CreateNewBackup return the BackupID of the newly-created backup. Implementation details: Refactored ChrootFileSystem to split off new base class RemapFileSystem, which allows more general remapping of files. We use this base class to implement BackupEngineImpl::RemapSharedFileSystem. To minimize API impact, I decided to just add these fields `name_for_open` and `env_for_open` to those set by GetBackupInfo when include_file_details=true. Creating the RemapSharedFileSystem adds a bit to the memory consumption, perhaps unnecessarily in some cases, but this has been mitigated by (a) only initialize the RemapSharedFileSystem lazily when GetBackupInfo with include_file_details=true is called, and (b) using the existing `shared_ptr<FileInfo>` objects to hold most of the mapping data. To enhance API safety, RemapSharedFileSystem is wrapped by new ReadOnlyFileSystem which rejects any attempts to write. This uncovered a couple of places in which DB::OpenForReadOnly would write to the filesystem, so I fixed these. Added a release note because this affects logging. Additional minor refactoring in backupable_db.cc to support the new functionality. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8142 Test Plan: new test (run with ASAN and UBSAN), added to stress test and ran it for a while with amplified backup_one_in Reviewed By: ajkr Differential Revision: D27535408 Pulled By: pdillinger fbshipit-source-id: 04666d310aa0261ef6b2385c43ca793ce1dfd148
4 years ago
const bool seq_per_batch, const bool batch_per_txn,
bool read_only)
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
: dbname_(dbname),
own_info_log_(options.info_log == nullptr),
Make backups openable as read-only DBs (#8142) Summary: A current limitation of backups is that you don't know the exact database state of when the backup was taken. With this new feature, you can at least inspect the backup's DB state without restoring it by opening it as a read-only DB. Rather than add something like OpenAsReadOnlyDB to the BackupEngine API, which would inhibit opening stackable DB implementations read-only (if/when their APIs support it), we instead provide a DB name and Env that can be used to open as a read-only DB. Possible follow-up work: * Add a version of GetBackupInfo for a single backup. * Let CreateNewBackup return the BackupID of the newly-created backup. Implementation details: Refactored ChrootFileSystem to split off new base class RemapFileSystem, which allows more general remapping of files. We use this base class to implement BackupEngineImpl::RemapSharedFileSystem. To minimize API impact, I decided to just add these fields `name_for_open` and `env_for_open` to those set by GetBackupInfo when include_file_details=true. Creating the RemapSharedFileSystem adds a bit to the memory consumption, perhaps unnecessarily in some cases, but this has been mitigated by (a) only initialize the RemapSharedFileSystem lazily when GetBackupInfo with include_file_details=true is called, and (b) using the existing `shared_ptr<FileInfo>` objects to hold most of the mapping data. To enhance API safety, RemapSharedFileSystem is wrapped by new ReadOnlyFileSystem which rejects any attempts to write. This uncovered a couple of places in which DB::OpenForReadOnly would write to the filesystem, so I fixed these. Added a release note because this affects logging. Additional minor refactoring in backupable_db.cc to support the new functionality. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8142 Test Plan: new test (run with ASAN and UBSAN), added to stress test and ran it for a while with amplified backup_one_in Reviewed By: ajkr Differential Revision: D27535408 Pulled By: pdillinger fbshipit-source-id: 04666d310aa0261ef6b2385c43ca793ce1dfd148
4 years ago
initial_db_options_(SanitizeOptions(dbname, options, read_only)),
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
env_(initial_db_options_.env),
io_tracer_(std::make_shared<IOTracer>()),
immutable_db_options_(initial_db_options_),
fs_(immutable_db_options_.fs, io_tracer_),
mutable_db_options_(initial_db_options_),
stats_(immutable_db_options_.stats),
mutex_(stats_, immutable_db_options_.clock, DB_MUTEX_WAIT_MICROS,
immutable_db_options_.use_adaptive_mutex),
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
default_cf_handle_(nullptr),
error_handler_(this, immutable_db_options_, &mutex_),
event_logger_(immutable_db_options_.info_log.get()),
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
6 years ago
max_total_in_memory_state_(0),
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
file_options_(BuildDBOptions(immutable_db_options_, mutable_db_options_)),
file_options_for_compaction_(fs_->OptimizeForCompactionTableWrite(
file_options_, immutable_db_options_)),
seq_per_batch_(seq_per_batch),
batch_per_txn_(batch_per_txn),
next_job_id_(1),
shutting_down_(false),
db_lock_(nullptr),
manual_compaction_paused_(false),
bg_cv_(&mutex_),
logfile_number_(0),
log_dir_synced_(false),
log_empty_(true),
persist_stats_cf_handle_(nullptr),
log_sync_cv_(&mutex_),
total_log_size_(0),
is_snapshot_supported_(true),
write_buffer_manager_(immutable_db_options_.write_buffer_manager.get()),
write_thread_(immutable_db_options_),
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
nonmem_write_thread_(immutable_db_options_),
write_controller_(mutable_db_options_.delayed_write_rate),
last_batch_group_size_(0),
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
unscheduled_flushes_(0),
unscheduled_compactions_(0),
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
bg_bottom_compaction_scheduled_(0),
bg_compaction_scheduled_(0),
num_running_compactions_(0),
bg_flush_scheduled_(0),
num_running_flushes_(0),
bg_purge_scheduled_(0),
disable_delete_obsolete_files_(0),
pending_purge_obsolete_files_(0),
delete_obsolete_files_last_run_(immutable_db_options_.clock->NowMicros()),
last_stats_dump_time_microsec_(0),
has_unpersisted_data_(false),
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
7 years ago
unable_to_release_oldest_log_(false),
num_running_ingest_file_(0),
#ifndef ROCKSDB_LITE
wal_manager_(immutable_db_options_, file_options_, io_tracer_,
seq_per_batch),
#endif // ROCKSDB_LITE
bg_work_paused_(0),
bg_compaction_paused_(0),
refitting_level_(false),
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
opened_successfully_(false),
#ifndef ROCKSDB_LITE
periodic_work_scheduler_(nullptr),
#endif // ROCKSDB_LITE
two_write_queues_(options.two_write_queues),
manual_wal_flush_(options.manual_wal_flush),
// last_sequencee_ is always maintained by the main queue that also writes
// to the memtable. When two_write_queues_ is disabled last seq in
// memtable is the same as last seq published to the readers. When it is
// enabled but seq_per_batch_ is disabled, last seq in memtable still
// indicates last published seq since wal-only writes that go to the 2nd
// queue do not consume a sequence number. Otherwise writes performed by
// the 2nd queue could change what is visible to the readers. In this
// cases, last_seq_same_as_publish_seq_==false, the 2nd queue maintains a
// separate variable to indicate the last published sequence.
last_seq_same_as_publish_seq_(
!(seq_per_batch && options.two_write_queues)),
// Since seq_per_batch_ is currently set only by WritePreparedTxn which
// requires a custom gc for compaction, we use that to set use_custom_gc_
// as well.
use_custom_gc_(seq_per_batch),
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
shutdown_initiated_(false),
own_sfm_(options.sst_file_manager == nullptr),
preserve_deletes_(options.preserve_deletes),
closed_(false),
atomic_flush_install_cv_(&mutex_),
blob_callback_(immutable_db_options_.sst_file_manager.get(), &mutex_,
&error_handler_, &event_logger_,
immutable_db_options_.listeners, dbname_) {
// !batch_per_trx_ implies seq_per_batch_ because it is only unset for
// WriteUnprepared, which should use seq_per_batch_.
assert(batch_per_txn_ || seq_per_batch_);
// TODO: Check for an error here
env_->GetAbsolutePath(dbname, &db_absolute_path_).PermitUncheckedError();
// Reserve ten files or so for other uses and give the rest to TableCache.
// Give a large number for setting of "infinite" open files.
const int table_cache_size = (mutable_db_options_.max_open_files == -1)
? TableCache::kInfiniteCapacity
: mutable_db_options_.max_open_files - 10;
LRUCacheOptions co;
co.capacity = table_cache_size;
co.num_shard_bits = immutable_db_options_.table_cache_numshardbits;
co.metadata_charge_policy = kDontChargeCacheMetadata;
table_cache_ = NewLRUCache(co);
SetDbSessionId();
assert(!db_session_id_.empty());
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
versions_.reset(new VersionSet(dbname_, &immutable_db_options_, file_options_,
table_cache_.get(), write_buffer_manager_,
&write_controller_, &block_cache_tracer_,
io_tracer_, db_session_id_));
support for concurrent adds to memtable Summary: This diff adds support for concurrent adds to the skiplist memtable implementations. Memory allocation is made thread-safe by the addition of a spinlock, with small per-core buffers to avoid contention. Concurrent memtable writes are made via an additional method and don't impose a performance overhead on the non-concurrent case, so parallelism can be selected on a per-batch basis. Write thread synchronization is an increasing bottleneck for higher levels of concurrency, so this diff adds --enable_write_thread_adaptive_yield (default off). This feature causes threads joining a write batch group to spin for a short time (default 100 usec) using sched_yield, rather than going to sleep on a mutex. If the timing of the yield calls indicates that another thread has actually run during the yield then spinning is avoided. This option improves performance for concurrent situations even without parallel adds, although it has the potential to increase CPU usage (and the heuristic adaptation is not yet mature). Parallel writes are not currently compatible with inplace updates, update callbacks, or delete filtering. Enable it with --allow_concurrent_memtable_write (and --enable_write_thread_adaptive_yield). Parallel memtable writes are performance neutral when there is no actual parallelism, and in my experiments (SSD server-class Linux and varying contention and key sizes for fillrandom) they are always a performance win when there is more than one thread. Statistics are updated earlier in the write path, dropping the number of DB mutex acquisitions from 2 to 1 for almost all cases. This diff was motivated and inspired by Yahoo's cLSM work. It is more conservative than cLSM: RocksDB's write batch group leader role is preserved (along with all of the existing flush and write throttling logic) and concurrent writers are blocked until all memtable insertions have completed and the sequence number has been advanced, to preserve linearizability. My test config is "db_bench -benchmarks=fillrandom -threads=$T -batch_size=1 -memtablerep=skip_list -value_size=100 --num=1000000/$T -level0_slowdown_writes_trigger=9999 -level0_stop_writes_trigger=9999 -disable_auto_compactions --max_write_buffer_number=8 -max_background_flushes=8 --disable_wal --write_buffer_size=160000000 --block_size=16384 --allow_concurrent_memtable_write" on a two-socket Xeon E5-2660 @ 2.2Ghz with lots of memory and an SSD hard drive. With 1 thread I get ~440Kops/sec. Peak performance for 1 socket (numactl -N1) is slightly more than 1Mops/sec, at 16 threads. Peak performance across both sockets happens at 30 threads, and is ~900Kops/sec, although with fewer threads there is less performance loss when the system has background work. Test Plan: 1. concurrent stress tests for InlineSkipList and DynamicBloom 2. make clean; make check 3. make clean; DISABLE_JEMALLOC=1 make valgrind_check; valgrind db_bench 4. make clean; COMPILE_WITH_TSAN=1 make all check; db_bench 5. make clean; COMPILE_WITH_ASAN=1 make all check; db_bench 6. make clean; OPT=-DROCKSDB_LITE make check 7. verify no perf regressions when disabled Reviewers: igor, sdong Reviewed By: sdong Subscribers: MarkCallaghan, IslamAbdelRahman, anthony, yhchiang, rven, sdong, guyg8, kradhakrishnan, dhruba Differential Revision: https://reviews.facebook.net/D50589
9 years ago
column_family_memtables_.reset(
new ColumnFamilyMemTablesImpl(versions_->GetColumnFamilySet()));
DumpRocksDBBuildVersion(immutable_db_options_.info_log.get());
DumpDBFileSummary(immutable_db_options_, dbname_, db_session_id_);
immutable_db_options_.Dump(immutable_db_options_.info_log.get());
mutable_db_options_.Dump(immutable_db_options_.info_log.get());
DumpSupportInfo(immutable_db_options_.info_log.get());
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
7 years ago
// always open the DB with 0 here, which means if preserve_deletes_==true
// we won't drop any deletion markers until SetPreserveDeletesSequenceNumber()
// is called by client and this seqnum is advanced.
preserve_deletes_seqnum_.store(0);
Stall writes in WriteBufferManager when memory_usage exceeds buffer_size (#7898) Summary: When WriteBufferManager is shared across DBs and column families to maintain memory usage under a limit, OOMs have been observed when flush cannot finish but writes continuously insert to memtables. In order to avoid OOMs, when memory usage goes beyond buffer_limit_ and DBs tries to write, this change will stall incoming writers until flush is completed and memory_usage drops. Design: Stall condition: When total memory usage exceeds WriteBufferManager::buffer_size_ (memory_usage() >= buffer_size_) WriterBufferManager::ShouldStall() returns true. DBImpl first block incoming/future writers by calling write_thread_.BeginWriteStall() (which adds dummy stall object to the writer's queue). Then DB is blocked on a state State::Blocked (current write doesn't go through). WBStallInterface object maintained by every DB instance is added to the queue of WriteBufferManager. If multiple DBs tries to write during this stall, they will also be blocked when check WriteBufferManager::ShouldStall() returns true. End Stall condition: When flush is finished and memory usage goes down, stall will end only if memory waiting to be flushed is less than buffer_size/2. This lower limit will give time for flush to complete and avoid continous stalling if memory usage remains close to buffer_size. WriterBufferManager::EndWriteStall() is called, which removes all instances from its queue and signal them to continue. Their state is changed to State::Running and they are unblocked. DBImpl then signal all incoming writers of that DB to continue by calling write_thread_.EndWriteStall() (which removes dummy stall object from the queue). DB instance creates WBMStallInterface which is an interface to block and signal DBs during stall. When DB needs to be blocked or signalled by WriteBufferManager, state_for_wbm_ state is changed accordingly (RUNNING or BLOCKED). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7898 Test Plan: Added a new test db/db_write_buffer_manager_test.cc Reviewed By: anand1976 Differential Revision: D26093227 Pulled By: akankshamahajan15 fbshipit-source-id: 2bbd982a3fb7033f6de6153aa92a221249861aae
4 years ago
if (write_buffer_manager_) {
wbm_stall_.reset(new WBMStallInterface());
}
}
Status DBImpl::Resume() {
ROCKS_LOG_INFO(immutable_db_options_.info_log, "Resuming DB");
InstrumentedMutexLock db_mutex(&mutex_);
if (!error_handler_.IsDBStopped() && !error_handler_.IsBGWorkStopped()) {
// Nothing to do
return Status::OK();
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
if (error_handler_.IsRecoveryInProgress()) {
// Don't allow a mix of manual and automatic recovery
return Status::Busy();
}
mutex_.Unlock();
Status s = error_handler_.RecoverFromBGError(true);
mutex_.Lock();
return s;
}
// This function implements the guts of recovery from a background error. It
// is eventually called for both manual as well as automatic recovery. It does
// the following -
// 1. Wait for currently scheduled background flush/compaction to exit, in
// order to inadvertently causing an error and thinking recovery failed
// 2. Flush memtables if there's any data for all the CFs. This may result
// another error, which will be saved by error_handler_ and reported later
// as the recovery status
// 3. Find and delete any obsolete files
// 4. Schedule compactions if needed for all the CFs. This is needed as the
// flush in the prior step might have been a no-op for some CFs, which
// means a new super version wouldn't have been installed
Status DBImpl::ResumeImpl(DBRecoverContext context) {
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
mutex_.AssertHeld();
WaitForBackgroundWork();
Status s;
if (shutdown_initiated_) {
// Returning shutdown status to SFM during auto recovery will cause it
// to abort the recovery and allow the shutdown to progress
s = Status::ShutdownInProgress();
}
if (s.ok()) {
Status bg_error = error_handler_.GetBGError();
if (bg_error.severity() > Status::Severity::kHardError) {
ROCKS_LOG_INFO(
immutable_db_options_.info_log,
"DB resume requested but failed due to Fatal/Unrecoverable error");
s = bg_error;
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
}
// Make sure the IO Status stored in version set is set to OK.
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
5 years ago
bool file_deletion_disabled = !IsFileDeletionsEnabled();
if (s.ok()) {
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
5 years ago
IOStatus io_s = versions_->io_status();
if (io_s.IsIOError()) {
// If resuming from IOError resulted from MANIFEST write, then assert
// that we must have already set the MANIFEST writer to nullptr during
// clean-up phase MANIFEST writing. We must have also disabled file
// deletions.
assert(!versions_->descriptor_log_);
assert(file_deletion_disabled);
// Since we are trying to recover from MANIFEST write error, we need to
// switch to a new MANIFEST anyway. The old MANIFEST can be corrupted.
// Therefore, force writing a dummy version edit because we do not know
// whether there are flush jobs with non-empty data to flush, triggering
// appends to MANIFEST.
VersionEdit edit;
auto cfh =
static_cast_with_check<ColumnFamilyHandleImpl>(default_cf_handle_);
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
5 years ago
assert(cfh);
ColumnFamilyData* cfd = cfh->cfd();
const MutableCFOptions& cf_opts = *cfd->GetLatestMutableCFOptions();
s = versions_->LogAndApply(cfd, cf_opts, &edit, &mutex_,
directories_.GetDbDir());
if (!s.ok()) {
io_s = versions_->io_status();
if (!io_s.ok()) {
s = error_handler_.SetBGError(io_s,
BackgroundErrorReason::kManifestWrite);
}
}
}
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// We cannot guarantee consistency of the WAL. So force flush Memtables of
// all the column families
if (s.ok()) {
FlushOptions flush_opts;
// We allow flush to stall write since we are trying to resume from error.
flush_opts.allow_write_stall = true;
if (immutable_db_options_.atomic_flush) {
autovector<ColumnFamilyData*> cfds;
SelectColumnFamiliesForAtomicFlush(&cfds);
mutex_.Unlock();
s = AtomicFlushMemTables(cfds, flush_opts, context.flush_reason);
mutex_.Lock();
} else {
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->IsDropped()) {
continue;
}
cfd->Ref();
mutex_.Unlock();
s = FlushMemTable(cfd, flush_opts, context.flush_reason);
mutex_.Lock();
cfd->UnrefAndTryDelete();
if (!s.ok()) {
break;
}
}
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
if (!s.ok()) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"DB resume requested but failed due to Flush failure [%s]",
s.ToString().c_str());
}
}
JobContext job_context(0);
FindObsoleteFiles(&job_context, true);
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
if (s.ok()) {
s = error_handler_.ClearBGError();
} else {
// NOTE: this is needed to pass ASSERT_STATUS_CHECKED
// in the DBSSTTest.DBWithMaxSpaceAllowedRandomized test.
// See https://github.com/facebook/rocksdb/pull/7715#issuecomment-754947952
error_handler_.GetRecoveryError().PermitUncheckedError();
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
}
mutex_.Unlock();
job_context.manifest_file_number = 1;
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
job_context.Clean();
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
if (s.ok()) {
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
5 years ago
assert(versions_->io_status().ok());
// If we reach here, we should re-enable file deletions if it was disabled
// during previous error handling.
if (file_deletion_disabled) {
// Always return ok
s = EnableFileDeletions(/*force=*/true);
if (!s.ok()) {
ROCKS_LOG_INFO(
immutable_db_options_.info_log,
"DB resume requested but could not enable file deletions [%s]",
s.ToString().c_str());
}
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
5 years ago
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
ROCKS_LOG_INFO(immutable_db_options_.info_log, "Successfully resumed DB");
}
mutex_.Lock();
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// Check for shutdown again before scheduling further compactions,
// since we released and re-acquired the lock above
if (shutdown_initiated_) {
s = Status::ShutdownInProgress();
}
if (s.ok()) {
for (auto cfd : *versions_->GetColumnFamilySet()) {
SchedulePendingCompaction(cfd);
}
MaybeScheduleFlushOrCompaction();
}
// Wake up any waiters - in this case, it could be the shutdown thread
bg_cv_.SignalAll();
// No need to check BGError again. If something happened, event listener would
// be notified and the operation causing it would have failed
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
return s;
}
void DBImpl::WaitForBackgroundWork() {
// Wait for background work to finish
while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
bg_flush_scheduled_) {
bg_cv_.Wait();
}
}
// Will lock the mutex_, will wait for completion if wait is true
void DBImpl::CancelAllBackgroundWork(bool wait) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Shutdown: canceling all background work");
#ifndef ROCKSDB_LITE
if (periodic_work_scheduler_ != nullptr) {
periodic_work_scheduler_->Unregister(this);
}
#endif // !ROCKSDB_LITE
InstrumentedMutexLock l(&mutex_);
if (!shutting_down_.load(std::memory_order_acquire) &&
has_unpersisted_data_.load(std::memory_order_relaxed) &&
!mutable_db_options_.avoid_flush_during_shutdown) {
if (immutable_db_options_.atomic_flush) {
autovector<ColumnFamilyData*> cfds;
SelectColumnFamiliesForAtomicFlush(&cfds);
mutex_.Unlock();
Status s =
AtomicFlushMemTables(cfds, FlushOptions(), FlushReason::kShutDown);
s.PermitUncheckedError(); //**TODO: What to do on error?
mutex_.Lock();
} else {
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (!cfd->IsDropped() && cfd->initialized() && !cfd->mem()->IsEmpty()) {
cfd->Ref();
mutex_.Unlock();
Status s = FlushMemTable(cfd, FlushOptions(), FlushReason::kShutDown);
s.PermitUncheckedError(); //**TODO: What to do on error?
mutex_.Lock();
cfd->UnrefAndTryDelete();
}
}
}
versions_->GetColumnFamilySet()->FreeDeadColumnFamilies();
}
Persist data during user initiated shutdown Summary: Move the manual memtable flush for databases containing data that has bypassed the WAL from DBImpl's destructor to CancleAllBackgroundWork(). CancelAllBackgroundWork() is a publicly exposed API which allows async operations performed by background threads to be disabled on a database. In effect, this places the database into a "shutdown" state in advance of calling the database object's destructor. No compactions or flushing of SST files can occur once a call to this API completes. When writes are issued to a database with WriteOptions::disableWAL set to true, DBImpl::has_unpersisted_data_ is set so that memtables can be flushed when the database object is destroyed. If CancelAllBackgroundWork() has been called prior to DBImpl's destructor, this flush operation is not possible and is skipped, causing unnecessary loss of data. Since CancelAllBackgroundWork() is already invoked by DBImpl's destructor in order to perform the thread join portion of its cleanup processing, moving the manual memtable flush to CancelAllBackgroundWork() ensures data is persisted regardless of client behavior. Test Plan: Write an amount of data that will not cause a memtable flush to a rocksdb database with all writes marked with WriteOptions::disableWAL. Properly "close" the database. Reopen database and verify that the data was persisted. Reviewers: IslamAbdelRahman, yiwu, yoshinorim, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D62277
8 years ago
shutting_down_.store(true, std::memory_order_release);
bg_cv_.SignalAll();
if (!wait) {
return;
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
WaitForBackgroundWork();
Persist data during user initiated shutdown Summary: Move the manual memtable flush for databases containing data that has bypassed the WAL from DBImpl's destructor to CancleAllBackgroundWork(). CancelAllBackgroundWork() is a publicly exposed API which allows async operations performed by background threads to be disabled on a database. In effect, this places the database into a "shutdown" state in advance of calling the database object's destructor. No compactions or flushing of SST files can occur once a call to this API completes. When writes are issued to a database with WriteOptions::disableWAL set to true, DBImpl::has_unpersisted_data_ is set so that memtables can be flushed when the database object is destroyed. If CancelAllBackgroundWork() has been called prior to DBImpl's destructor, this flush operation is not possible and is skipped, causing unnecessary loss of data. Since CancelAllBackgroundWork() is already invoked by DBImpl's destructor in order to perform the thread join portion of its cleanup processing, moving the manual memtable flush to CancelAllBackgroundWork() ensures data is persisted regardless of client behavior. Test Plan: Write an amount of data that will not cause a memtable flush to a rocksdb database with all writes marked with WriteOptions::disableWAL. Properly "close" the database. Reopen database and verify that the data was persisted. Reviewers: IslamAbdelRahman, yiwu, yoshinorim, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D62277
8 years ago
}
Status DBImpl::CloseHelper() {
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// Guarantee that there is no background error recovery in progress before
// continuing with the shutdown
mutex_.Lock();
shutdown_initiated_ = true;
error_handler_.CancelErrorRecovery();
while (error_handler_.IsRecoveryInProgress()) {
bg_cv_.Wait();
}
mutex_.Unlock();
// Below check is added as recovery_error_ is not checked and it causes crash
// in DBSSTTest.DBWithMaxSpaceAllowedWithBlobFiles when space limit is
// reached.
error_handler_.GetRecoveryError().PermitUncheckedError();
// CancelAllBackgroundWork called with false means we just set the shutdown
// marker. After this we do a variant of the waiting and unschedule work
// (to consider: moving all the waiting into CancelAllBackgroundWork(true))
CancelAllBackgroundWork(false);
mutex_.Lock();
env_->UnSchedule(this, Env::Priority::BOTTOM);
env_->UnSchedule(this, Env::Priority::LOW);
env_->UnSchedule(this, Env::Priority::HIGH);
Status ret = Status::OK();
// Wait for background work to finish
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
while (bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
bg_flush_scheduled_ || bg_purge_scheduled_ ||
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
pending_purge_obsolete_files_ ||
error_handler_.IsRecoveryInProgress()) {
TEST_SYNC_POINT("DBImpl::~DBImpl:WaitJob");
bg_cv_.Wait();
}
Fix race condition causing double deletion of ssts Summary: Possible interleaved execution of background compaction thread calling `FindObsoleteFiles (no full scan) / PurgeObsoleteFiles` and user thread calling `FindObsoleteFiles (full scan) / PurgeObsoleteFiles` can lead to race condition on which RocksDB attempts to delete a file twice. The second attempt will fail and return `IO error`. This may occur to other files, but this PR targets sst. Also add a unit test to verify that this PR fixes the issue. The newly added unit test `obsolete_files_test` has a test case for this scenario, implemented in `ObsoleteFilesTest#RaceForObsoleteFileDeletion`. `TestSyncPoint`s are used to coordinate the interleaving the `user_thread` and background compaction thread. They execute as follows ``` timeline user_thread background_compaction thread t1 | FindObsoleteFiles(full_scan=false) t2 | FindObsoleteFiles(full_scan=true) t3 | PurgeObsoleteFiles t4 | PurgeObsoleteFiles V ``` When `user_thread` invokes `FindObsoleteFiles` with full scan, it collects ALL files in RocksDB directory, including the ones that background compaction thread have collected in its job context. Then `user_thread` will see an IO error when trying to delete these files in `PurgeObsoleteFiles` because background compaction thread has already deleted the file in `PurgeObsoleteFiles`. To fix this, we make RocksDB remember which (SST) files have been found by threads after calling `FindObsoleteFiles` (see `DBImpl#files_grabbed_for_purge_`). Therefore, when another thread calls `FindObsoleteFiles` with full scan, it will not collect such files. ajkr could you take a look and comment? Thanks! Closes https://github.com/facebook/rocksdb/pull/3638 Differential Revision: D7384372 Pulled By: riversand963 fbshipit-source-id: 01489516d60012e722ee65a80e1449e589ce26d3
7 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::CloseHelper:PendingPurgeFinished",
&files_grabbed_for_purge_);
EraseThreadStatusDbInfo();
flush_scheduler_.Clear();
Refactor trimming logic for immutable memtables (#5022) Summary: MyRocks currently sets `max_write_buffer_number_to_maintain` in order to maintain enough history for transaction conflict checking. The effectiveness of this approach depends on the size of memtables. When memtables are small, it may not keep enough history; when memtables are large, this may consume too much memory. We are proposing a new way to configure memtable list history: by limiting the memory usage of immutable memtables. The new option is `max_write_buffer_size_to_maintain` and it will take precedence over the old `max_write_buffer_number_to_maintain` if they are both set to non-zero values. The new option accounts for the total memory usage of flushed immutable memtables and mutable memtable. When the total usage exceeds the limit, RocksDB may start dropping immutable memtables (which is also called trimming history), starting from the oldest one. The semantics of the old option actually works both as an upper bound and lower bound. History trimming will start if number of immutable memtables exceeds the limit, but it will never go below (limit-1) due to history trimming. In order the mimic the behavior with the new option, history trimming will stop if dropping the next immutable memtable causes the total memory usage go below the size limit. For example, assuming the size limit is set to 64MB, and there are 3 immutable memtables with sizes of 20, 30, 30. Although the total memory usage is 80MB > 64MB, dropping the oldest memtable will reduce the memory usage to 60MB < 64MB, so in this case no memtable will be dropped. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5022 Differential Revision: D14394062 Pulled By: miasantreble fbshipit-source-id: 60457a509c6af89d0993f988c9b5c2aa9e45f5c5
5 years ago
trim_history_scheduler_.Clear();
Make mempurge a background process (equivalent to in-memory compaction). (#8505) Summary: In https://github.com/facebook/rocksdb/issues/8454, I introduced a new process baptized `MemPurge` (memtable garbage collection). This new PR is built upon this past mempurge prototype. In this PR, I made the `mempurge` process a background task, which provides superior performance since the mempurge process does not cling on the db_mutex anymore, and addresses severe restrictions from the past iteration (including a scenario where the past mempurge was failling, when a memtable was mempurged but was still referred to by an iterator/snapshot/...). Now the mempurge process ressembles an in-memory compaction process: the stack of immutable memtables is filtered out, and the useful payload is used to populate an output memtable. If the output memtable is filled at more than 60% capacity (arbitrary heuristic) the mempurge process is aborted and a regular flush process takes place, else the output memtable is kept in the immutable memtable stack. Note that adding this output memtable to the `imm()` memtable stack does not trigger another flush process, so that the flush thread can go to sleep at the end of a successful mempurge. MemPurge is activated by making the `experimental_allow_mempurge` flag `true`. When activated, the `MemPurge` process will always happen when the flush reason is `kWriteBufferFull`. The 3 unit tests confirm that this process supports `Put`, `Get`, `Delete`, `DeleteRange` operators and is compatible with `Iterators` and `CompactionFilters`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8505 Reviewed By: pdillinger Differential Revision: D29619283 Pulled By: bjlemaire fbshipit-source-id: 8a99bee76b63a8211bff1a00e0ae32360aaece95
4 years ago
// For now, simply trigger a manual flush at close time
// on all the column families.
// TODO(bjlemaire): Check if this is needed. Also, in the
// future we can contemplate doing a more fine-grained
// flushing by first checking if there is a need for
// flushing (but need to implement something
// else than imm()->IsFlushPending() because the output
// memtables added to imm() dont trigger flushes).
Memtable sampling for mempurge heuristic. (#8628) Summary: Changes the API of the MemPurge process: the `bool experimental_allow_mempurge` and `experimental_mempurge_policy` flags have been replaced by a `double experimental_mempurge_threshold` option. This change of API reflects another major change introduced in this PR: the MemPurgeDecider() function now works by sampling the memtables being flushed to estimate the overall amount of useful payload (payload minus the garbage), and then compare this useful payload estimate with the `double experimental_mempurge_threshold` value. Therefore, when the value of this flag is `0.0` (default value), mempurge is simply deactivated. On the other hand, a value of `DBL_MAX` would be equivalent to always going through a mempurge regardless of the garbage ratio estimate. At the moment, a `double experimental_mempurge_threshold` value else than 0.0 or `DBL_MAX` is opnly supported`with the `SkipList` memtable representation. Regarding the sampling, this PR includes the introduction of a `MemTable::UniqueRandomSample` function that collects (approximately) random entries from the memtable by using the new `SkipList::Iterator::RandomSeek()` under the hood, or by iterating through each memtable entry, depending on the target sample size and the total number of entries. The unit tests have been readapted to support this new API. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8628 Reviewed By: pdillinger Differential Revision: D30149315 Pulled By: bjlemaire fbshipit-source-id: 1feef5390c95db6f4480ab4434716533d3947f27
3 years ago
if (immutable_db_options_.experimental_mempurge_threshold > 0.0) {
Make mempurge a background process (equivalent to in-memory compaction). (#8505) Summary: In https://github.com/facebook/rocksdb/issues/8454, I introduced a new process baptized `MemPurge` (memtable garbage collection). This new PR is built upon this past mempurge prototype. In this PR, I made the `mempurge` process a background task, which provides superior performance since the mempurge process does not cling on the db_mutex anymore, and addresses severe restrictions from the past iteration (including a scenario where the past mempurge was failling, when a memtable was mempurged but was still referred to by an iterator/snapshot/...). Now the mempurge process ressembles an in-memory compaction process: the stack of immutable memtables is filtered out, and the useful payload is used to populate an output memtable. If the output memtable is filled at more than 60% capacity (arbitrary heuristic) the mempurge process is aborted and a regular flush process takes place, else the output memtable is kept in the immutable memtable stack. Note that adding this output memtable to the `imm()` memtable stack does not trigger another flush process, so that the flush thread can go to sleep at the end of a successful mempurge. MemPurge is activated by making the `experimental_allow_mempurge` flag `true`. When activated, the `MemPurge` process will always happen when the flush reason is `kWriteBufferFull`. The 3 unit tests confirm that this process supports `Put`, `Get`, `Delete`, `DeleteRange` operators and is compatible with `Iterators` and `CompactionFilters`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8505 Reviewed By: pdillinger Differential Revision: D29619283 Pulled By: bjlemaire fbshipit-source-id: 8a99bee76b63a8211bff1a00e0ae32360aaece95
4 years ago
Status flush_ret;
mutex_.Unlock();
for (ColumnFamilyData* cf : *versions_->GetColumnFamilySet()) {
if (immutable_db_options_.atomic_flush) {
flush_ret = AtomicFlushMemTables({cf}, FlushOptions(),
FlushReason::kManualFlush);
if (!flush_ret.ok()) {
ROCKS_LOG_INFO(
immutable_db_options_.info_log,
"Atomic flush memtables failed upon closing (mempurge).");
}
} else {
flush_ret =
FlushMemTable(cf, FlushOptions(), FlushReason::kManualFlush);
if (!flush_ret.ok()) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Flush memtables failed upon closing (mempurge).");
}
}
}
mutex_.Lock();
}
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
while (!flush_queue_.empty()) {
const FlushRequest& flush_req = PopFirstFromFlushQueue();
for (const auto& iter : flush_req) {
iter.first->UnrefAndTryDelete();
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
}
}
Make mempurge a background process (equivalent to in-memory compaction). (#8505) Summary: In https://github.com/facebook/rocksdb/issues/8454, I introduced a new process baptized `MemPurge` (memtable garbage collection). This new PR is built upon this past mempurge prototype. In this PR, I made the `mempurge` process a background task, which provides superior performance since the mempurge process does not cling on the db_mutex anymore, and addresses severe restrictions from the past iteration (including a scenario where the past mempurge was failling, when a memtable was mempurged but was still referred to by an iterator/snapshot/...). Now the mempurge process ressembles an in-memory compaction process: the stack of immutable memtables is filtered out, and the useful payload is used to populate an output memtable. If the output memtable is filled at more than 60% capacity (arbitrary heuristic) the mempurge process is aborted and a regular flush process takes place, else the output memtable is kept in the immutable memtable stack. Note that adding this output memtable to the `imm()` memtable stack does not trigger another flush process, so that the flush thread can go to sleep at the end of a successful mempurge. MemPurge is activated by making the `experimental_allow_mempurge` flag `true`. When activated, the `MemPurge` process will always happen when the flush reason is `kWriteBufferFull`. The 3 unit tests confirm that this process supports `Put`, `Get`, `Delete`, `DeleteRange` operators and is compatible with `Iterators` and `CompactionFilters`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8505 Reviewed By: pdillinger Differential Revision: D29619283 Pulled By: bjlemaire fbshipit-source-id: 8a99bee76b63a8211bff1a00e0ae32360aaece95
4 years ago
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
while (!compaction_queue_.empty()) {
auto cfd = PopFirstFromCompactionQueue();
cfd->UnrefAndTryDelete();
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
}
if (default_cf_handle_ != nullptr || persist_stats_cf_handle_ != nullptr) {
// we need to delete handle outside of lock because it does its own locking
mutex_.Unlock();
if (default_cf_handle_) {
delete default_cf_handle_;
default_cf_handle_ = nullptr;
}
if (persist_stats_cf_handle_) {
delete persist_stats_cf_handle_;
persist_stats_cf_handle_ = nullptr;
}
mutex_.Lock();
}
// Clean up obsolete files due to SuperVersion release.
// (1) Need to delete to obsolete files before closing because RepairDB()
// scans all existing files in the file system and builds manifest file.
// Keeping obsolete files confuses the repair process.
// (2) Need to check if we Open()/Recover() the DB successfully before
// deleting because if VersionSet recover fails (may be due to corrupted
// manifest file), it is not able to identify live files correctly. As a
// result, all "live" files can get deleted by accident. However, corrupted
// manifest is recoverable by RepairDB().
if (opened_successfully_) {
JobContext job_context(next_job_id_.fetch_add(1));
FindObsoleteFiles(&job_context, true);
mutex_.Unlock();
// manifest number starting from 2
job_context.manifest_file_number = 1;
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
job_context.Clean();
mutex_.Lock();
}
for (auto l : logs_to_free_) {
delete l;
}
for (auto& log : logs_) {
uint64_t log_number = log.writer->get_log_number();
Status s = log.ClearWriter();
if (!s.ok()) {
ROCKS_LOG_WARN(
immutable_db_options_.info_log,
"Unable to Sync WAL file %s with error -- %s",
LogFileName(immutable_db_options_.GetWalDir(), log_number).c_str(),
s.ToString().c_str());
// Retain the first error
if (ret.ok()) {
ret = s;
}
}
}
logs_.clear();
// Table cache may have table handles holding blocks from the block cache.
// We need to release them before the block cache is destroyed. The block
// cache may be destroyed inside versions_.reset(), when column family data
// list is destroyed, so leaving handles in table cache after
// versions_.reset() may cause issues.
// Here we clean all unreferenced handles in table cache.
// Now we assume all user queries have finished, so only version set itself
// can possibly hold the blocks from block cache. After releasing unreferenced
// handles here, only handles held by version set left and inside
// versions_.reset(), we will release them. There, we need to make sure every
// time a handle is released, we erase it from the cache too. By doing that,
// we can guarantee that after versions_.reset(), table cache is empty
// so the cache can be safely destroyed.
table_cache_->EraseUnRefEntries();
for (auto& txn_entry : recovered_transactions_) {
delete txn_entry.second;
}
// versions need to be destroyed before table_cache since it can hold
// references to table_cache.
versions_.reset();
mutex_.Unlock();
if (db_lock_ != nullptr) {
// TODO: Check for unlock error
env_->UnlockFile(db_lock_).PermitUncheckedError();
}
ROCKS_LOG_INFO(immutable_db_options_.info_log, "Shutdown complete");
LogFlush(immutable_db_options_.info_log);
#ifndef ROCKSDB_LITE
// If the sst_file_manager was allocated by us during DB::Open(), ccall
// Close() on it before closing the info_log. Otherwise, background thread
// in SstFileManagerImpl might try to log something
if (immutable_db_options_.sst_file_manager && own_sfm_) {
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
sfm->Close();
}
#endif // ROCKSDB_LITE
if (immutable_db_options_.info_log && own_info_log_) {
Status s = immutable_db_options_.info_log->Close();
Fix many tests to run with MEM_ENV and ENCRYPTED_ENV; Introduce a MemoryFileSystem class (#7566) Summary: This PR does a few things: 1. The MockFileSystem class was split out from the MockEnv. This change would theoretically allow a MockFileSystem to be used by other Environments as well (if we created a means of constructing one). The MockFileSystem implements a FileSystem in its entirety and does not rely on any Wrapper implementation. 2. Make the RocksDB test suite work when MOCK_ENV=1 and ENCRYPTED_ENV=1 are set. To accomplish this, a few things were needed: - The tests that tried to use the "wrong" environment (Env::Default() instead of env_) were updated - The MockFileSystem was changed to support the features it was missing or mishandled (such as recursively deleting files in a directory or supporting renaming of a directory). 3. Updated the test framework to have a ROCKSDB_GTEST_SKIP macro. This can be used to flag tests that are skipped. Currently, this defaults to doing nothing (marks the test as SUCCESS) but will mark the tests as SKIPPED when RocksDB is upgraded to a version of gtest that supports this (gtest-1.10). I have run a full "make check" with MEM_ENV, ENCRYPTED_ENV, both, and neither under both MacOS and RedHat. A few tests were disabled/skipped for the MEM/ENCRYPTED cases. The error_handler_fs_test fails/hangs for MEM_ENV (presumably a timing problem) and I will introduce another PR/issue to track that problem. (I will also push a change to disable those tests soon). There is one more test in DBTest2 that also fails which I need to investigate or skip before this PR is merged. Theoretically, this PR should also allow the test suite to run against an Env loaded from the registry, though I do not have one to try it with currently. Finally, once this is accepted, it would be nice if there was a CircleCI job to run these tests on a checkin so this effort does not become stale. I do not know how to do that, so if someone could write that job, it would be appreciated :) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7566 Reviewed By: zhichao-cao Differential Revision: D24408980 Pulled By: jay-zhuang fbshipit-source-id: 911b1554a4d0da06fd51feca0c090a4abdcb4a5f
4 years ago
if (!s.ok() && !s.IsNotSupported() && ret.ok()) {
ret = s;
}
}
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
Stall writes in WriteBufferManager when memory_usage exceeds buffer_size (#7898) Summary: When WriteBufferManager is shared across DBs and column families to maintain memory usage under a limit, OOMs have been observed when flush cannot finish but writes continuously insert to memtables. In order to avoid OOMs, when memory usage goes beyond buffer_limit_ and DBs tries to write, this change will stall incoming writers until flush is completed and memory_usage drops. Design: Stall condition: When total memory usage exceeds WriteBufferManager::buffer_size_ (memory_usage() >= buffer_size_) WriterBufferManager::ShouldStall() returns true. DBImpl first block incoming/future writers by calling write_thread_.BeginWriteStall() (which adds dummy stall object to the writer's queue). Then DB is blocked on a state State::Blocked (current write doesn't go through). WBStallInterface object maintained by every DB instance is added to the queue of WriteBufferManager. If multiple DBs tries to write during this stall, they will also be blocked when check WriteBufferManager::ShouldStall() returns true. End Stall condition: When flush is finished and memory usage goes down, stall will end only if memory waiting to be flushed is less than buffer_size/2. This lower limit will give time for flush to complete and avoid continous stalling if memory usage remains close to buffer_size. WriterBufferManager::EndWriteStall() is called, which removes all instances from its queue and signal them to continue. Their state is changed to State::Running and they are unblocked. DBImpl then signal all incoming writers of that DB to continue by calling write_thread_.EndWriteStall() (which removes dummy stall object from the queue). DB instance creates WBMStallInterface which is an interface to block and signal DBs during stall. When DB needs to be blocked or signalled by WriteBufferManager, state_for_wbm_ state is changed accordingly (RUNNING or BLOCKED). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7898 Test Plan: Added a new test db/db_write_buffer_manager_test.cc Reviewed By: anand1976 Differential Revision: D26093227 Pulled By: akankshamahajan15 fbshipit-source-id: 2bbd982a3fb7033f6de6153aa92a221249861aae
4 years ago
if (write_buffer_manager_ && wbm_stall_) {
write_buffer_manager_->RemoveDBFromQueue(wbm_stall_.get());
}
if (ret.IsAborted()) {
// Reserve IsAborted() error for those where users didn't release
// certain resource and they can release them and come back and
// retry. In this case, we wrap this exception to something else.
return Status::Incomplete(ret.ToString());
}
return ret;
}
Status DBImpl::CloseImpl() { return CloseHelper(); }
DBImpl::~DBImpl() {
InstrumentedMutexLock closing_lock_guard(&closing_mutex_);
if (!closed_) {
closed_ = true;
closing_status_ = CloseHelper();
closing_status_.PermitUncheckedError();
}
}
void DBImpl::MaybeIgnoreError(Status* s) const {
if (s->ok() || immutable_db_options_.paranoid_checks) {
// No change needed
} else {
ROCKS_LOG_WARN(immutable_db_options_.info_log, "Ignoring error %s",
s->ToString().c_str());
*s = Status::OK();
}
}
const Status DBImpl::CreateArchivalDirectory() {
if (immutable_db_options_.WAL_ttl_seconds > 0 ||
immutable_db_options_.WAL_size_limit_MB > 0) {
std::string archivalPath =
ArchivalDirectory(immutable_db_options_.GetWalDir());
return env_->CreateDirIfMissing(archivalPath);
}
return Status::OK();
}
void DBImpl::PrintStatistics() {
auto dbstats = immutable_db_options_.stats;
if (dbstats) {
ROCKS_LOG_INFO(immutable_db_options_.info_log, "STATISTICS:\n %s",
dbstats->ToString().c_str());
}
}
void DBImpl::StartPeriodicWorkScheduler() {
#ifndef ROCKSDB_LITE
#ifndef NDEBUG
// It only used by test to disable scheduler
bool disable_scheduler = false;
TEST_SYNC_POINT_CALLBACK(
"DBImpl::StartPeriodicWorkScheduler:DisableScheduler",
&disable_scheduler);
if (disable_scheduler) {
return;
}
#endif // !NDEBUG
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
{
InstrumentedMutexLock l(&mutex_);
periodic_work_scheduler_ = PeriodicWorkScheduler::Default();
TEST_SYNC_POINT_CALLBACK("DBImpl::StartPeriodicWorkScheduler:Init",
&periodic_work_scheduler_);
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
}
periodic_work_scheduler_->Register(
this, mutable_db_options_.stats_dump_period_sec,
mutable_db_options_.stats_persist_period_sec);
#endif // !ROCKSDB_LITE
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
}
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
11 years ago
// esitmate the total size of stats_history_
size_t DBImpl::EstimateInMemoryStatsHistorySize() const {
size_t size_total =
sizeof(std::map<uint64_t, std::map<std::string, uint64_t>>);
if (stats_history_.size() == 0) return size_total;
size_t size_per_slice =
sizeof(uint64_t) + sizeof(std::map<std::string, uint64_t>);
// non-empty map, stats_history_.begin() guaranteed to exist
std::map<std::string, uint64_t> sample_slice(stats_history_.begin()->second);
for (const auto& pairs : sample_slice) {
size_per_slice +=
pairs.first.capacity() + sizeof(pairs.first) + sizeof(pairs.second);
}
size_total = size_per_slice * stats_history_.size();
return size_total;
}
void DBImpl::PersistStats() {
TEST_SYNC_POINT("DBImpl::PersistStats:Entry");
#ifndef ROCKSDB_LITE
if (shutdown_initiated_) {
return;
}
TEST_SYNC_POINT("DBImpl::PersistStats:StartRunning");
uint64_t now_seconds =
immutable_db_options_.clock->NowMicros() / kMicrosInSecond;
Statistics* statistics = immutable_db_options_.stats;
if (!statistics) {
return;
}
size_t stats_history_size_limit = 0;
{
InstrumentedMutexLock l(&mutex_);
stats_history_size_limit = mutable_db_options_.stats_history_buffer_size;
}
std::map<std::string, uint64_t> stats_map;
if (!statistics->getTickerMap(&stats_map)) {
return;
}
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"------- PERSISTING STATS -------");
if (immutable_db_options_.persist_stats_to_disk) {
WriteBatch batch;
Status s = Status::OK();
if (stats_slice_initialized_) {
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Reading %" ROCKSDB_PRIszt " stats from statistics\n",
stats_slice_.size());
for (const auto& stat : stats_map) {
if (s.ok()) {
char key[100];
int length =
EncodePersistentStatsKey(now_seconds, stat.first, 100, key);
// calculate the delta from last time
if (stats_slice_.find(stat.first) != stats_slice_.end()) {
uint64_t delta = stat.second - stats_slice_[stat.first];
s = batch.Put(persist_stats_cf_handle_,
Slice(key, std::min(100, length)), ToString(delta));
}
}
}
}
stats_slice_initialized_ = true;
std::swap(stats_slice_, stats_map);
if (s.ok()) {
WriteOptions wo;
wo.low_pri = true;
wo.no_slowdown = true;
wo.sync = false;
s = Write(wo, &batch);
}
if (!s.ok()) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
"Writing to persistent stats CF failed -- %s",
s.ToString().c_str());
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
} else {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Writing %" ROCKSDB_PRIszt " stats with timestamp %" PRIu64
" to persistent stats CF succeeded",
stats_slice_.size(), now_seconds);
}
// TODO(Zhongyi): add purging for persisted data
} else {
InstrumentedMutexLock l(&stats_history_mutex_);
// calculate the delta from last time
if (stats_slice_initialized_) {
std::map<std::string, uint64_t> stats_delta;
for (const auto& stat : stats_map) {
if (stats_slice_.find(stat.first) != stats_slice_.end()) {
stats_delta[stat.first] = stat.second - stats_slice_[stat.first];
}
}
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Storing %" ROCKSDB_PRIszt " stats with timestamp %" PRIu64
" to in-memory stats history",
stats_slice_.size(), now_seconds);
stats_history_[now_seconds] = stats_delta;
}
stats_slice_initialized_ = true;
std::swap(stats_slice_, stats_map);
TEST_SYNC_POINT("DBImpl::PersistStats:StatsCopied");
// delete older stats snapshots to control memory consumption
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
size_t stats_history_size = EstimateInMemoryStatsHistorySize();
bool purge_needed = stats_history_size > stats_history_size_limit;
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[Pre-GC] In-memory stats history size: %" ROCKSDB_PRIszt
" bytes, slice count: %" ROCKSDB_PRIszt,
stats_history_size, stats_history_.size());
while (purge_needed && !stats_history_.empty()) {
stats_history_.erase(stats_history_.begin());
purge_needed =
EstimateInMemoryStatsHistorySize() > stats_history_size_limit;
}
add more tracing for stats history (#5566) Summary: Sample info log output from db_bench: In-memory: ``` 2019/07/12-21:39:19.478490 7fa01b3f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:39:19.478633 7fa01b3f5700 [_impl/db_impl.cc:753] Storing 145 stats with timestamp 1562992759 to in-memory stats history 2019/07/12-21:39:19.478670 7fa01b3f5700 [_impl/db_impl.cc:766] [Pre-GC] In-memory stats history size: 1051218 bytes, slice count: 103 2019/07/12-21:39:19.478704 7fa01b3f5700 [_impl/db_impl.cc:775] [Post-GC] In-memory stats history size: 1051218 bytes, slice count: 102 ``` On-disk: ``` 2019/07/12-21:48:53.862548 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:53.862553 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:53.862852 7f24943f5700 [_impl/db_impl.cc:737] Writing 145 stats with timestamp 1562993333 to persistent stats CF succeeded ``` ``` 2019/07/12-21:48:51.861711 7f24943f5700 [_impl/db_impl.cc:702] ------- PERSISTING STATS ------- 2019/07/12-21:48:51.861729 7f24943f5700 [_impl/db_impl.cc:709] Reading 145 stats from statistics 2019/07/12-21:48:51.861921 7f24943f5700 [_impl/db_impl.cc:732] Writing to persistent stats CF failed -- Result incomplete: Write stall ... 2019/07/12-21:48:51.873032 7f2494bf6700 [WARN] [lumn_family.cc:749] [default] Stopping writes because we have 2 immutable memtables (waiting for flush), max_write_buffer_number is set to 2 ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5566 Differential Revision: D16258187 Pulled By: miasantreble fbshipit-source-id: 292497099b941418590ed4312411bee36e244dc5
6 years ago
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[Post-GC] In-memory stats history size: %" ROCKSDB_PRIszt
" bytes, slice count: %" ROCKSDB_PRIszt,
stats_history_size, stats_history_.size());
}
TEST_SYNC_POINT("DBImpl::PersistStats:End");
#endif // !ROCKSDB_LITE
}
bool DBImpl::FindStatsByTime(uint64_t start_time, uint64_t end_time,
uint64_t* new_time,
std::map<std::string, uint64_t>* stats_map) {
assert(new_time);
assert(stats_map);
if (!new_time || !stats_map) return false;
// lock when search for start_time
{
InstrumentedMutexLock l(&stats_history_mutex_);
auto it = stats_history_.lower_bound(start_time);
if (it != stats_history_.end() && it->first < end_time) {
// make a copy for timestamp and stats_map
*new_time = it->first;
*stats_map = it->second;
return true;
} else {
return false;
}
}
}
Status DBImpl::GetStatsHistory(
uint64_t start_time, uint64_t end_time,
std::unique_ptr<StatsHistoryIterator>* stats_iterator) {
if (!stats_iterator) {
return Status::InvalidArgument("stats_iterator not preallocated.");
}
if (immutable_db_options_.persist_stats_to_disk) {
stats_iterator->reset(
new PersistentStatsHistoryIterator(start_time, end_time, this));
} else {
stats_iterator->reset(
new InMemoryStatsHistoryIterator(start_time, end_time, this));
}
return (*stats_iterator)->status();
}
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
void DBImpl::DumpStats() {
TEST_SYNC_POINT("DBImpl::DumpStats:1");
#ifndef ROCKSDB_LITE
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
std::string stats;
if (shutdown_initiated_) {
return;
}
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
TEST_SYNC_POINT("DBImpl::DumpStats:StartRunning");
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
{
InstrumentedMutexLock l(&mutex_);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->initialized()) {
// Release DB mutex for gathering cache entry stats. Pass over all
// column families for this first so that other stats are dumped
// near-atomically.
// Get a ref before unlocking
cfd->Ref();
{
InstrumentedMutexUnlock u(&mutex_);
cfd->internal_stats()->CollectCacheEntryStats(/*foreground=*/false);
}
cfd->UnrefAndTryDelete();
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
}
}
const std::string* property = &DB::Properties::kDBStats;
const DBPropertyInfo* property_info = GetPropertyInfo(*property);
assert(property_info != nullptr);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
assert(!property_info->need_out_of_mutex);
default_cf_internal_stats_->GetStringProperty(*property_info, *property,
&stats);
property = &DB::Properties::kCFStatsNoFileHistogram;
property_info = GetPropertyInfo(*property);
assert(property_info != nullptr);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
assert(!property_info->need_out_of_mutex);
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->initialized()) {
cfd->internal_stats()->GetStringProperty(*property_info, *property,
&stats);
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
11 years ago
}
}
property = &DB::Properties::kCFFileHistogram;
property_info = GetPropertyInfo(*property);
assert(property_info != nullptr);
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
assert(!property_info->need_out_of_mutex);
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->initialized()) {
cfd->internal_stats()->GetStringProperty(*property_info, *property,
&stats);
}
}
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
}
TEST_SYNC_POINT("DBImpl::DumpStats:2");
ROCKS_LOG_INFO(immutable_db_options_.info_log,
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
"------- DUMPING STATS -------");
ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s", stats.c_str());
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
if (immutable_db_options_.dump_malloc_stats) {
stats.clear();
DumpMallocStats(&stats);
if (!stats.empty()) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
"------- Malloc STATS -------");
ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s", stats.c_str());
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
}
}
#endif // !ROCKSDB_LITE
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
11 years ago
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
PrintStatistics();
}
void DBImpl::FlushInfoLog() {
if (shutdown_initiated_) {
return;
}
TEST_SYNC_POINT("DBImpl::FlushInfoLog:StartRunning");
LogFlush(immutable_db_options_.info_log);
}
Status DBImpl::TablesRangeTombstoneSummary(ColumnFamilyHandle* column_family,
int max_entries_to_print,
std::string* out_str) {
auto* cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
ColumnFamilyData* cfd = cfh->cfd();
SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
Version* version = super_version->current;
Status s =
version->TablesRangeTombstoneSummary(max_entries_to_print, out_str);
CleanupSuperVersion(super_version);
return s;
}
void DBImpl::ScheduleBgLogWriterClose(JobContext* job_context) {
if (!job_context->logs_to_free.empty()) {
for (auto l : job_context->logs_to_free) {
AddToLogsToFreeQueue(l);
}
job_context->logs_to_free.clear();
}
}
FSDirectory* DBImpl::GetDataDir(ColumnFamilyData* cfd, size_t path_id) const {
assert(cfd);
FSDirectory* ret_dir = cfd->GetDataDir(path_id);
if (ret_dir == nullptr) {
return directories_.GetDataDir(path_id);
}
return ret_dir;
}
Status DBImpl::SetOptions(
ColumnFamilyHandle* column_family,
const std::unordered_map<std::string, std::string>& options_map) {
#ifdef ROCKSDB_LITE
(void)column_family;
(void)options_map;
return Status::NotSupported("Not supported in ROCKSDB LITE");
#else
auto* cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
if (options_map.empty()) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"SetOptions() on column family [%s], empty input",
cfd->GetName().c_str());
return Status::InvalidArgument("empty input");
}
MutableCFOptions new_options;
Status s;
Status persist_options_status;
persist_options_status.PermitUncheckedError(); // Allow uninitialized access
SuperVersionContext sv_context(/* create_superversion */ true);
{
auto db_options = GetDBOptions();
InstrumentedMutexLock l(&mutex_);
s = cfd->SetOptions(db_options, options_map);
if (s.ok()) {
new_options = *cfd->GetLatestMutableCFOptions();
// Append new version to recompute compaction score.
VersionEdit dummy_edit;
s = versions_->LogAndApply(cfd, new_options, &dummy_edit, &mutex_,
directories_.GetDbDir());
// Trigger possible flush/compactions. This has to be before we persist
// options to file, otherwise there will be a deadlock with writer
// thread.
InstallSuperVersionAndScheduleWork(cfd, &sv_context, new_options);
persist_options_status = WriteOptionsFile(
false /*need_mutex_lock*/, true /*need_enter_write_thread*/);
bg_cv_.SignalAll();
}
}
sv_context.Clean();
ROCKS_LOG_INFO(
immutable_db_options_.info_log,
"SetOptions() on column family [%s], inputs:", cfd->GetName().c_str());
for (const auto& o : options_map) {
ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s: %s\n", o.first.c_str(),
o.second.c_str());
}
if (s.ok()) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[%s] SetOptions() succeeded", cfd->GetName().c_str());
new_options.Dump(immutable_db_options_.info_log.get());
if (!persist_options_status.ok()) {
s = persist_options_status;
}
} else {
ROCKS_LOG_WARN(immutable_db_options_.info_log, "[%s] SetOptions() failed",
cfd->GetName().c_str());
}
LogFlush(immutable_db_options_.info_log);
return s;
#endif // ROCKSDB_LITE
}
Status DBImpl::SetDBOptions(
const std::unordered_map<std::string, std::string>& options_map) {
#ifdef ROCKSDB_LITE
(void)options_map;
return Status::NotSupported("Not supported in ROCKSDB LITE");
#else
if (options_map.empty()) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"SetDBOptions(), empty input.");
return Status::InvalidArgument("empty input");
}
MutableDBOptions new_options;
Status s;
Status persist_options_status = Status::OK();
bool wal_changed = false;
WriteContext write_context;
{
InstrumentedMutexLock l(&mutex_);
s = GetMutableDBOptionsFromStrings(mutable_db_options_, options_map,
&new_options);
if (new_options.bytes_per_sync == 0) {
new_options.bytes_per_sync = 1024 * 1024;
}
if (MutableDBOptionsAreEqual(mutable_db_options_, new_options)) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"SetDBOptions(), input option value is not changed, "
"skipping updating.");
persist_options_status.PermitUncheckedError();
return s;
}
DBOptions new_db_options =
BuildDBOptions(immutable_db_options_, new_options);
if (s.ok()) {
s = ValidateOptions(new_db_options);
}
if (s.ok()) {
for (auto c : *versions_->GetColumnFamilySet()) {
if (!c->IsDropped()) {
auto cf_options = c->GetLatestCFOptions();
s = ColumnFamilyData::ValidateOptions(new_db_options, cf_options);
if (!s.ok()) {
break;
}
}
}
}
if (s.ok()) {
const BGJobLimits current_bg_job_limits =
GetBGJobLimits(mutable_db_options_.max_background_flushes,
mutable_db_options_.max_background_compactions,
mutable_db_options_.max_background_jobs,
/* parallelize_compactions */ true);
const BGJobLimits new_bg_job_limits = GetBGJobLimits(
new_options.max_background_flushes,
new_options.max_background_compactions,
new_options.max_background_jobs, /* parallelize_compactions */ true);
const bool max_flushes_increased =
new_bg_job_limits.max_flushes > current_bg_job_limits.max_flushes;
const bool max_compactions_increased =
new_bg_job_limits.max_compactions >
current_bg_job_limits.max_compactions;
if (max_flushes_increased || max_compactions_increased) {
if (max_flushes_increased) {
env_->IncBackgroundThreadsIfNeeded(new_bg_job_limits.max_flushes,
Env::Priority::HIGH);
}
if (max_compactions_increased) {
env_->IncBackgroundThreadsIfNeeded(new_bg_job_limits.max_compactions,
Env::Priority::LOW);
}
MaybeScheduleFlushOrCompaction();
}
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
6 years ago
if (new_options.stats_dump_period_sec !=
mutable_db_options_.stats_dump_period_sec ||
new_options.stats_persist_period_sec !=
mutable_db_options_.stats_persist_period_sec) {
mutex_.Unlock();
periodic_work_scheduler_->Unregister(this);
periodic_work_scheduler_->Register(
this, new_options.stats_dump_period_sec,
new_options.stats_persist_period_sec);
mutex_.Lock();
}
write_controller_.set_max_delayed_write_rate(
new_options.delayed_write_rate);
table_cache_.get()->SetCapacity(new_options.max_open_files == -1
? TableCache::kInfiniteCapacity
: new_options.max_open_files - 10);
wal_changed = mutable_db_options_.wal_bytes_per_sync !=
new_options.wal_bytes_per_sync;
mutable_db_options_ = new_options;
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
file_options_for_compaction_ = FileOptions(new_db_options);
file_options_for_compaction_ = fs_->OptimizeForCompactionTableWrite(
file_options_for_compaction_, immutable_db_options_);
versions_->ChangeFileOptions(mutable_db_options_);
//TODO(xiez): clarify why apply optimize for read to write options
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
file_options_for_compaction_ = fs_->OptimizeForCompactionTableRead(
file_options_for_compaction_, immutable_db_options_);
file_options_for_compaction_.compaction_readahead_size =
mutable_db_options_.compaction_readahead_size;
WriteThread::Writer w;
write_thread_.EnterUnbatched(&w, &mutex_);
if (total_log_size_ > GetMaxTotalWalSize() || wal_changed) {
Status purge_wal_status = SwitchWAL(&write_context);
if (!purge_wal_status.ok()) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"Unable to purge WAL files in SetDBOptions() -- %s",
purge_wal_status.ToString().c_str());
}
}
persist_options_status = WriteOptionsFile(
false /*need_mutex_lock*/, false /*need_enter_write_thread*/);
write_thread_.ExitUnbatched(&w);
} else {
// To get here, we must have had invalid options and will not attempt to
// persist the options, which means the status is "OK/Uninitialized.
persist_options_status.PermitUncheckedError();
}
}
ROCKS_LOG_INFO(immutable_db_options_.info_log, "SetDBOptions(), inputs:");
for (const auto& o : options_map) {
ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s: %s\n", o.first.c_str(),
o.second.c_str());
}
if (s.ok()) {
ROCKS_LOG_INFO(immutable_db_options_.info_log, "SetDBOptions() succeeded");
new_options.Dump(immutable_db_options_.info_log.get());
if (!persist_options_status.ok()) {
if (immutable_db_options_.fail_if_options_file_error) {
s = Status::IOError(
"SetDBOptions() succeeded, but unable to persist options",
persist_options_status.ToString());
}
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"Unable to persist options in SetDBOptions() -- %s",
persist_options_status.ToString().c_str());
}
} else {
ROCKS_LOG_WARN(immutable_db_options_.info_log, "SetDBOptions failed");
}
LogFlush(immutable_db_options_.info_log);
return s;
#endif // ROCKSDB_LITE
}
// return the same level if it cannot be moved
int DBImpl::FindMinimumEmptyLevelFitting(
ColumnFamilyData* cfd, const MutableCFOptions& /*mutable_cf_options*/,
int level) {
mutex_.AssertHeld();
const auto* vstorage = cfd->current()->storage_info();
int minimum_level = level;
for (int i = level - 1; i > 0; --i) {
// stop if level i is not empty
if (vstorage->NumLevelFiles(i) > 0) break;
// stop if level i is too small (cannot fit the level files)
if (vstorage->MaxBytesForLevel(i) < vstorage->NumLevelBytes(level)) {
break;
}
minimum_level = i;
}
return minimum_level;
}
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
Status DBImpl::FlushWAL(bool sync) {
if (manual_wal_flush_) {
IOStatus io_s;
{
// We need to lock log_write_mutex_ since logs_ might change concurrently
InstrumentedMutexLock wl(&log_write_mutex_);
log::Writer* cur_log_writer = logs_.back().writer;
io_s = cur_log_writer->WriteBuffer();
}
if (!io_s.ok()) {
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
ROCKS_LOG_ERROR(immutable_db_options_.info_log, "WAL flush error %s",
io_s.ToString().c_str());
// In case there is a fs error we should set it globally to prevent the
// future writes
IOStatusCheck(io_s);
// whether sync or not, we should abort the rest of function upon error
return std::move(io_s);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
}
if (!sync) {
ROCKS_LOG_DEBUG(immutable_db_options_.info_log, "FlushWAL sync=false");
return std::move(io_s);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
}
}
if (!sync) {
return Status::OK();
}
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
// sync = true
ROCKS_LOG_DEBUG(immutable_db_options_.info_log, "FlushWAL sync=true");
return SyncWAL();
}
Status DBImpl::SyncWAL() {
autovector<log::Writer*, 1> logs_to_sync;
bool need_log_dir_sync;
uint64_t current_log_number;
{
InstrumentedMutexLock l(&mutex_);
assert(!logs_.empty());
// This SyncWAL() call only cares about logs up to this number.
current_log_number = logfile_number_;
while (logs_.front().number <= current_log_number &&
logs_.front().getting_synced) {
log_sync_cv_.Wait();
}
// First check that logs are safe to sync in background.
for (auto it = logs_.begin();
it != logs_.end() && it->number <= current_log_number; ++it) {
if (!it->writer->file()->writable_file()->IsSyncThreadSafe()) {
return Status::NotSupported(
"SyncWAL() is not supported for this implementation of WAL file",
immutable_db_options_.allow_mmap_writes
? "try setting Options::allow_mmap_writes to false"
: Slice());
}
}
for (auto it = logs_.begin();
it != logs_.end() && it->number <= current_log_number; ++it) {
auto& log = *it;
assert(!log.getting_synced);
log.getting_synced = true;
logs_to_sync.push_back(log.writer);
}
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
need_log_dir_sync = !log_dir_synced_;
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
10 years ago
}
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
TEST_SYNC_POINT("DBWALTest::SyncWALNotWaitWrite:1");
RecordTick(stats_, WAL_FILE_SYNCED);
Status status;
IOStatus io_s;
for (log::Writer* log : logs_to_sync) {
io_s = log->file()->SyncWithoutFlush(immutable_db_options_.use_fsync);
if (!io_s.ok()) {
status = io_s;
break;
}
}
if (!io_s.ok()) {
ROCKS_LOG_ERROR(immutable_db_options_.info_log, "WAL Sync error %s",
io_s.ToString().c_str());
// In case there is a fs error we should set it globally to prevent the
// future writes
IOStatusCheck(io_s);
}
if (status.ok() && need_log_dir_sync) {
status = directories_.GetWalDir()->FsyncWithDirOptions(
IOOptions(), nullptr,
DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
}
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
TEST_SYNC_POINT("DBWALTest::SyncWALNotWaitWrite:2");
TEST_SYNC_POINT("DBImpl::SyncWAL:BeforeMarkLogsSynced:1");
{
InstrumentedMutexLock l(&mutex_);
if (status.ok()) {
status = MarkLogsSynced(current_log_number, need_log_dir_sync);
} else {
MarkLogsNotSynced(current_log_number);
}
}
TEST_SYNC_POINT("DBImpl::SyncWAL:BeforeMarkLogsSynced:2");
return status;
}
Expose DB methods to lock and unlock the WAL (#5146) Summary: Expose DB methods to lock and unlock the WAL. These methods are intended to use by MyRocks in order to obtain WAL coordinates in consistent way. Usage scenario is following: MySQL has performance_schema.log_status which provides information that enables a backup tool to copy the required log files without locking for the duration of copy. To populate this table MySQL does following: 1. Lock the binary log. Transactions are not allowed to commit now 2. Save the binary log coordinates 3. Walk through the storage engines and lock writes on each engine. For InnoDB, redo log is locked. For MyRocks, WAL should be locked. 4. Ask storage engines for their coordinates. InnoDB reports its current LSN and checkpoint LSN. MyRocks should report active WAL files names and sizes. 5. Release storage engine's locks 6. Unlock binary log Backup tool will then use this information to copy InnoDB, RocksDB and MySQL binary logs up to specified positions to end up with consistent DB state after restore. Currently, RocksDB allows to obtain the list of WAL files. Only missing bit is the method to lock the writes to WAL files. LockWAL method must flush the WAL in order for the reported size to be accurate (GetSortedWALFiles is using file system stat call to return the file size), also, since backup tool is going to copy the WAL, it is better to be flushed. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5146 Differential Revision: D14815447 Pulled By: maysamyabandeh fbshipit-source-id: eec9535a6025229ed471119f19fe7b3d8ae888a3
6 years ago
Status DBImpl::LockWAL() {
log_write_mutex_.Lock();
auto cur_log_writer = logs_.back().writer;
auto status = cur_log_writer->WriteBuffer();
if (!status.ok()) {
ROCKS_LOG_ERROR(immutable_db_options_.info_log, "WAL flush error %s",
status.ToString().c_str());
// In case there is a fs error we should set it globally to prevent the
// future writes
WriteStatusCheck(status);
}
return std::move(status);
Expose DB methods to lock and unlock the WAL (#5146) Summary: Expose DB methods to lock and unlock the WAL. These methods are intended to use by MyRocks in order to obtain WAL coordinates in consistent way. Usage scenario is following: MySQL has performance_schema.log_status which provides information that enables a backup tool to copy the required log files without locking for the duration of copy. To populate this table MySQL does following: 1. Lock the binary log. Transactions are not allowed to commit now 2. Save the binary log coordinates 3. Walk through the storage engines and lock writes on each engine. For InnoDB, redo log is locked. For MyRocks, WAL should be locked. 4. Ask storage engines for their coordinates. InnoDB reports its current LSN and checkpoint LSN. MyRocks should report active WAL files names and sizes. 5. Release storage engine's locks 6. Unlock binary log Backup tool will then use this information to copy InnoDB, RocksDB and MySQL binary logs up to specified positions to end up with consistent DB state after restore. Currently, RocksDB allows to obtain the list of WAL files. Only missing bit is the method to lock the writes to WAL files. LockWAL method must flush the WAL in order for the reported size to be accurate (GetSortedWALFiles is using file system stat call to return the file size), also, since backup tool is going to copy the WAL, it is better to be flushed. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5146 Differential Revision: D14815447 Pulled By: maysamyabandeh fbshipit-source-id: eec9535a6025229ed471119f19fe7b3d8ae888a3
6 years ago
}
Status DBImpl::UnlockWAL() {
log_write_mutex_.Unlock();
return Status::OK();
}
Status DBImpl::MarkLogsSynced(uint64_t up_to, bool synced_dir) {
mutex_.AssertHeld();
if (synced_dir && logfile_number_ == up_to) {
log_dir_synced_ = true;
}
VersionEdit synced_wals;
for (auto it = logs_.begin(); it != logs_.end() && it->number <= up_to;) {
auto& wal = *it;
assert(wal.getting_synced);
if (logs_.size() > 1) {
if (immutable_db_options_.track_and_verify_wals_in_manifest &&
Always track WAL obsoletion (#7759) Summary: Currently, when a WAL becomes obsolete after flushing, if VersionSet::WalSet does not contain the WAL, we do not track the WAL obsoletion event in MANIFEST. But consider this case: * WAL 10 is synced, a VersionEdit is LogAndApplied to MANIFEST to log this WAL addition event, but the VersionEdit is not applied to WalSet yet since its corresponding ManifestWriter is still pending in the write queue; * Since the above ManifestWriter is blocking, the LogAndApply will block on a conditional variable and release the db mutex, so another LogAndApply can proceed to enqueue other VersionEdits concurrently; * Now flush happens, and WAL 10 becomes obsolete, although WalSet does not contain WAL 10 yet, we should call LogAndApply to enqueue a VersionEdit to indicate the obsoletion of WAL 10; * otherwise, when the queued edit indicating WAL 10 addition is logged to MANIFEST, and DB crashes and reopens, the WAL 10 might have been removed from disk, but it still exists in MANIFEST. This PR changes the behavior to: always `LogAndApply` any WAL addition or obsoletion event, without considering the order issues caused by concurrency, but when applying the edits to `WalSet`, do not add the WALs if they are already obsolete. In this approach, the logical events of WAL addition and obsoletion are always tracked in MANIFEST, so we can inspect the MANIFEST and know all the previous WAL events, but we choose to ignore certain events due to the concurrency issues such as the case above, or the case in https://github.com/facebook/rocksdb/pull/7725. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7759 Test Plan: make check Reviewed By: pdillinger Differential Revision: D25423089 Pulled By: cheng-chang fbshipit-source-id: 9cb9a7fbc1875bf954f2a42f9b6cfd6d49a7b21c
4 years ago
wal.writer->file()->GetFileSize() > 0) {
synced_wals.AddWal(wal.number,
WalMetadata(wal.writer->file()->GetFileSize()));
}
logs_to_free_.push_back(wal.ReleaseWriter());
// To modify logs_ both mutex_ and log_write_mutex_ must be held
InstrumentedMutexLock l(&log_write_mutex_);
it = logs_.erase(it);
} else {
wal.getting_synced = false;
++it;
}
}
assert(logs_.empty() || logs_[0].number > up_to ||
(logs_.size() == 1 && !logs_[0].getting_synced));
Status s;
if (synced_wals.IsWalAddition()) {
// not empty, write to MANIFEST.
s = versions_->LogAndApplyToDefaultColumnFamily(&synced_wals, &mutex_);
if (!s.ok() && versions_->io_status().IsIOError()) {
s = error_handler_.SetBGError(versions_->io_status(),
BackgroundErrorReason::kManifestWrite);
}
}
log_sync_cv_.SignalAll();
return s;
}
void DBImpl::MarkLogsNotSynced(uint64_t up_to) {
mutex_.AssertHeld();
for (auto it = logs_.begin(); it != logs_.end() && it->number <= up_to;
++it) {
auto& wal = *it;
assert(wal.getting_synced);
wal.getting_synced = false;
}
log_sync_cv_.SignalAll();
}
SequenceNumber DBImpl::GetLatestSequenceNumber() const {
return versions_->LastSequence();
}
void DBImpl::SetLastPublishedSequence(SequenceNumber seq) {
versions_->SetLastPublishedSequence(seq);
}
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
7 years ago
bool DBImpl::SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) {
if (seqnum > preserve_deletes_seqnum_.load()) {
preserve_deletes_seqnum_.store(seqnum);
return true;
} else {
return false;
}
}
InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
Arena* arena,
RangeDelAggregator* range_del_agg,
SequenceNumber sequence,
ColumnFamilyHandle* column_family,
bool allow_unprepared_value) {
ColumnFamilyData* cfd;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
cfd = cfh->cfd();
}
mutex_.Lock();
SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
mutex_.Unlock();
return NewInternalIterator(read_options, cfd, super_version, arena,
range_del_agg, sequence, allow_unprepared_value);
}
void DBImpl::SchedulePurge() {
mutex_.AssertHeld();
assert(opened_successfully_);
// Purge operations are put into High priority queue
bg_purge_scheduled_++;
env_->Schedule(&DBImpl::BGWorkPurge, this, Env::Priority::HIGH, nullptr);
}
void DBImpl::BackgroundCallPurge() {
mutex_.Lock();
while (!logs_to_free_queue_.empty()) {
assert(!logs_to_free_queue_.empty());
log::Writer* log_writer = *(logs_to_free_queue_.begin());
logs_to_free_queue_.pop_front();
mutex_.Unlock();
delete log_writer;
mutex_.Lock();
}
while (!superversions_to_free_queue_.empty()) {
assert(!superversions_to_free_queue_.empty());
SuperVersion* sv = superversions_to_free_queue_.front();
superversions_to_free_queue_.pop_front();
mutex_.Unlock();
delete sv;
mutex_.Lock();
}
// Can't use iterator to go over purge_files_ because inside the loop we're
// unlocking the mutex that protects purge_files_.
while (!purge_files_.empty()) {
auto it = purge_files_.begin();
// Need to make a copy of the PurgeFilesInfo before unlocking the mutex.
PurgeFileInfo purge_file = it->second;
const std::string& fname = purge_file.fname;
const std::string& dir_to_sync = purge_file.dir_to_sync;
FileType type = purge_file.type;
uint64_t number = purge_file.number;
int job_id = purge_file.job_id;
purge_files_.erase(it);
mutex_.Unlock();
DeleteObsoleteFileImpl(job_id, fname, dir_to_sync, type, number);
mutex_.Lock();
}
bg_purge_scheduled_--;
bg_cv_.SignalAll();
// IMPORTANT:there should be no code after calling SignalAll. This call may
// signal the DB destructor that it's OK to proceed with destruction. In
// that case, all DB variables will be dealloacated and referencing them
// will cause trouble.
mutex_.Unlock();
}
namespace {
struct IterState {
IterState(DBImpl* _db, InstrumentedMutex* _mu, SuperVersion* _super_version,
bool _background_purge)
: db(_db),
mu(_mu),
super_version(_super_version),
background_purge(_background_purge) {}
DBImpl* db;
InstrumentedMutex* mu;
SuperVersion* super_version;
bool background_purge;
};
static void CleanupIteratorState(void* arg1, void* /*arg2*/) {
IterState* state = reinterpret_cast<IterState*>(arg1);
if (state->super_version->Unref()) {
// Job id == 0 means that this is not our background process, but rather
// user thread
JobContext job_context(0);
state->mu->Lock();
state->super_version->Cleanup();
state->db->FindObsoleteFiles(&job_context, false, true);
if (state->background_purge) {
state->db->ScheduleBgLogWriterClose(&job_context);
state->db->AddSuperVersionsToFreeQueue(state->super_version);
state->db->SchedulePurge();
}
state->mu->Unlock();
if (!state->background_purge) {
delete state->super_version;
}
if (job_context.HaveSomethingToDelete()) {
if (state->background_purge) {
// PurgeObsoleteFiles here does not delete files. Instead, it adds the
// files to be deleted to a job queue, and deletes it in a separate
// background thread.
state->db->PurgeObsoleteFiles(job_context, true /* schedule only */);
state->mu->Lock();
state->db->SchedulePurge();
state->mu->Unlock();
} else {
state->db->PurgeObsoleteFiles(job_context);
}
}
job_context.Clean();
MemTableListVersion Summary: MemTableListVersion is to MemTableList what Version is to VersionSet. I took almost the same ideas to develop MemTableListVersion. The reason is to have copying std::list done in background, while flushing, rather than in foreground (MultiGet() and NewIterator()) under a mutex! Also, whenever we copied MemTableList, we copied also some MemTableList metadata (flush_requested_, commit_in_progress_, etc.), which was wasteful. This diff avoids std::list copy under a mutex in both MultiGet() and NewIterator(). I created a small database with some number of immutable memtables, and creating 100.000 iterators in a single-thread (!) decreased from {188739, 215703, 198028} to {154352, 164035, 159817}. A lot of the savings come from code under a mutex, so we should see much higher savings with multiple threads. Creating new iterator is very important to LogDevice team. I also think this diff will make SuperVersion obsolete for performance reasons. I will try it in the next diff. SuperVersion gave us huge savings on Get() code path, but I think that most of the savings came from copying MemTableList under a mutex. If we had MemTableListVersion, we would never need to copy the entire object (like we still do in NewIterator() and MultiGet()) Test Plan: `make check` works. I will also do `make valgrind_check` before commit Reviewers: dhruba, haobo, kailiu, sdong, emayanke, tnovak Reviewed By: kailiu CC: leveldb Differential Revision: https://reviews.facebook.net/D15255
11 years ago
}
delete state;
}
} // namespace
InternalIterator* DBImpl::NewInternalIterator(const ReadOptions& read_options,
ColumnFamilyData* cfd,
SuperVersion* super_version,
Arena* arena,
RangeDelAggregator* range_del_agg,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
5 years ago
SequenceNumber sequence,
bool allow_unprepared_value) {
InternalIterator* internal_iter;
assert(arena != nullptr);
assert(range_del_agg != nullptr);
// Need to create internal iterator from the arena.
MergeIteratorBuilder merge_iter_builder(
&cfd->internal_comparator(), arena,
!read_options.total_order_seek &&
super_version->mutable_cf_options.prefix_extractor != nullptr);
// Collect iterator for mutable mem
merge_iter_builder.AddIterator(
super_version->mem->NewIterator(read_options, arena));
std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter;
Status s;
if (!read_options.ignore_range_deletions) {
range_del_iter.reset(
super_version->mem->NewRangeTombstoneIterator(read_options, sequence));
range_del_agg->AddTombstones(std::move(range_del_iter));
}
// Collect all needed child iterators for immutable memtables
if (s.ok()) {
super_version->imm->AddIterators(read_options, &merge_iter_builder);
if (!read_options.ignore_range_deletions) {
s = super_version->imm->AddRangeTombstoneIterators(read_options, arena,
range_del_agg);
}
}
TEST_SYNC_POINT_CALLBACK("DBImpl::NewInternalIterator:StatusCallback", &s);
if (s.ok()) {
// Collect iterators for files in L0 - Ln
if (read_options.read_tier != kMemtableTier) {
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
super_version->current->AddIterators(read_options, file_options_,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
5 years ago
&merge_iter_builder, range_del_agg,
allow_unprepared_value);
}
internal_iter = merge_iter_builder.Finish();
IterState* cleanup =
new IterState(this, &mutex_, super_version,
read_options.background_purge_on_iterator_cleanup ||
immutable_db_options_.avoid_unnecessary_blocking_io);
internal_iter->RegisterCleanup(CleanupIteratorState, cleanup, nullptr);
return internal_iter;
} else {
CleanupSuperVersion(super_version);
}
return NewErrorInternalIterator<Slice>(s, arena);
}
ColumnFamilyHandle* DBImpl::DefaultColumnFamily() const {
return default_cf_handle_;
}
ColumnFamilyHandle* DBImpl::PersistentStatsColumnFamily() const {
return persist_stats_cf_handle_;
}
Status DBImpl::Get(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const Slice& key,
PinnableSlice* value) {
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
return Get(read_options, column_family, key, value, /*timestamp=*/nullptr);
}
Status DBImpl::Get(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const Slice& key,
PinnableSlice* value, std::string* timestamp) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
GetImplOptions get_impl_options;
get_impl_options.column_family = column_family;
get_impl_options.value = value;
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
get_impl_options.timestamp = timestamp;
Status s = GetImpl(read_options, key, get_impl_options);
return s;
}
namespace {
class GetWithTimestampReadCallback : public ReadCallback {
public:
explicit GetWithTimestampReadCallback(SequenceNumber seq)
: ReadCallback(seq) {}
bool IsVisibleFullCheck(SequenceNumber seq) override {
return seq <= max_visible_seq_;
}
};
} // namespace
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
Status DBImpl::GetImpl(const ReadOptions& read_options, const Slice& key,
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
GetImplOptions& get_impl_options) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
assert(get_impl_options.value != nullptr ||
get_impl_options.merge_operands != nullptr);
assert(get_impl_options.column_family);
const Comparator* ucmp = get_impl_options.column_family->GetComparator();
assert(ucmp);
size_t ts_sz = ucmp->timestamp_size();
GetWithTimestampReadCallback read_cb(0); // Will call Refresh
#ifndef NDEBUG
if (ts_sz > 0) {
assert(read_options.timestamp);
assert(read_options.timestamp->size() == ts_sz);
} else {
assert(!read_options.timestamp);
}
#endif // NDEBUG
PERF_CPU_TIMER_GUARD(get_cpu_nanos, immutable_db_options_.clock);
StopWatch sw(immutable_db_options_.clock, stats_, DB_GET);
PERF_TIMER_GUARD(get_snapshot_time);
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(
get_impl_options.column_family);
auto cfd = cfh->cfd();
if (tracer_) {
// TODO: This mutex should be removed later, to improve performance when
// tracing is enabled.
InstrumentedMutexLock lock(&trace_mutex_);
if (tracer_) {
// TODO: maybe handle the tracing status?
tracer_->Get(get_impl_options.column_family, key).PermitUncheckedError();
}
}
// Acquire SuperVersion
SuperVersion* sv = GetAndRefSuperVersion(cfd);
TEST_SYNC_POINT("DBImpl::GetImpl:1");
TEST_SYNC_POINT("DBImpl::GetImpl:2");
SequenceNumber snapshot;
if (read_options.snapshot != nullptr) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
if (get_impl_options.callback) {
// Already calculated based on read_options.snapshot
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
snapshot = get_impl_options.callback->max_visible_seq();
} else {
snapshot =
reinterpret_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
}
} else {
// Note that the snapshot is assigned AFTER referencing the super
// version because otherwise a flush happening in between may compact away
// data for the snapshot, so the reader would see neither data that was be
// visible to the snapshot before compaction nor the newer data inserted
// afterwards.
Fix TSAN build and re-enable the tests (#7386) Summary: Resolve TSAN build warnings and re-enable disabled TSAN tests. Not sure if it's a compiler issue or TSAN check issue. Switching from conditional operator to if-else mitigated the problem. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7386 Test Plan: run TSAN check 10 times in circleci. ``` WARNING: ThreadSanitizer: data race (pid=27735) Atomic write of size 8 at 0x7b54000005e8 by thread T32: #0 __tsan_atomic64_store <null> (db_test+0x4cee95) https://github.com/facebook/rocksdb/issues/1 std::__atomic_base<unsigned long>::store(unsigned long, std::memory_order) /usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/bits/atomic_base.h:374:2 (db_test+0x78460e) https://github.com/facebook/rocksdb/issues/2 rocksdb::VersionSet::SetLastSequence(unsigned long) /home/circleci/project/./db/version_set.h:1058:20 (db_test+0x78460e) ... Previous read of size 8 at 0x7b54000005e8 by thread T31: #0 bool rocksdb::DBImpl::MultiCFSnapshot<std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > > >(rocksdb::ReadOptions const&, rocksdb::ReadCallback*, std::function<rocksdb::DBImpl::MultiGetColumnFamilyData* (std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >::iterator&)>&, std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >*, unsigned long*) /home/circleci/project/db/db_impl/db_impl.cc (db_test+0x715087) ``` Reviewed By: ltamasi Differential Revision: D23725226 Pulled By: jay-zhuang fbshipit-source-id: a6d662a5ea68111246cd32ec95f3411a25f76bc6
4 years ago
if (last_seq_same_as_publish_seq_) {
snapshot = versions_->LastSequence();
} else {
snapshot = versions_->LastPublishedSequence();
}
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
if (get_impl_options.callback) {
// The unprep_seqs are not published for write unprepared, so it could be
// that max_visible_seq is larger. Seek to the std::max of the two.
// However, we still want our callback to contain the actual snapshot so
// that it can do the correct visibility filtering.
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
get_impl_options.callback->Refresh(snapshot);
// Internally, WriteUnpreparedTxnReadCallback::Refresh would set
// max_visible_seq = max(max_visible_seq, snapshot)
//
// Currently, the commented out assert is broken by
// InvalidSnapshotReadCallback, but if write unprepared recovery followed
// the regular transaction flow, then this special read callback would not
// be needed.
//
// assert(callback->max_visible_seq() >= snapshot);
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
snapshot = get_impl_options.callback->max_visible_seq();
}
}
// If timestamp is used, we use read callback to ensure <key,t,s> is returned
// only if t <= read_opts.timestamp and s <= snapshot.
// HACK: temporarily overwrite input struct field but restore
SaveAndRestore<ReadCallback*> restore_callback(&get_impl_options.callback);
if (ts_sz > 0) {
assert(!get_impl_options
.callback); // timestamp with callback is not supported
read_cb.Refresh(snapshot);
get_impl_options.callback = &read_cb;
}
TEST_SYNC_POINT("DBImpl::GetImpl:3");
TEST_SYNC_POINT("DBImpl::GetImpl:4");
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
12 years ago
// Prepare to store a list of merge operations if merge occurs.
MergeContext merge_context;
Use only "local" range tombstones during Get (#4449) Summary: Previously, range tombstones were accumulated from every level, which was necessary if a range tombstone in a higher level covered a key in a lower level. However, RangeDelAggregator::AddTombstones's complexity is based on the number of tombstones that are currently stored in it, which is wasteful in the Get case, where we only need to know the highest sequence number of range tombstones that cover the key from higher levels, and compute the highest covering sequence number at the current level. This change introduces this optimization, and removes the use of RangeDelAggregator from the Get path. In the benchmark results, the following command was used to initialize the database: ``` ./db_bench -db=/dev/shm/5k-rts -use_existing_db=false -benchmarks=filluniquerandom -write_buffer_size=1048576 -compression_type=lz4 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -value_size=112 -key_size=16 -block_size=4096 -level_compaction_dynamic_level_bytes=true -num=5000000 -max_background_jobs=12 -benchmark_write_rate_limit=20971520 -range_tombstone_width=100 -writes_per_range_tombstone=100 -max_num_range_tombstones=50000 -bloom_bits=8 ``` ...and the following command was used to measure read throughput: ``` ./db_bench -db=/dev/shm/5k-rts/ -use_existing_db=true -benchmarks=readrandom -disable_auto_compactions=true -num=5000000 -reads=100000 -threads=32 ``` The filluniquerandom command was only run once, and the resulting database was used to measure read performance before and after the PR. Both binaries were compiled with `DEBUG_LEVEL=0`. Readrandom results before PR: ``` readrandom : 4.544 micros/op 220090 ops/sec; 16.9 MB/s (63103 of 100000 found) ``` Readrandom results after PR: ``` readrandom : 11.147 micros/op 89707 ops/sec; 6.9 MB/s (63103 of 100000 found) ``` So it's actually slower right now, but this PR paves the way for future optimizations (see #4493). ---- Pull Request resolved: https://github.com/facebook/rocksdb/pull/4449 Differential Revision: D10370575 Pulled By: abhimadan fbshipit-source-id: 9a2e152be1ef36969055c0e9eb4beb0d96c11f4d
6 years ago
SequenceNumber max_covering_tombstone_seq = 0;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
12 years ago
Status s;
// First look in the memtable, then in the immutable memtable (if any).
// s is both in/out. When in, s could either be OK or MergeInProgress.
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
12 years ago
// merge_operands will contain the sequence of merges in the latter case.
Avoid user key copying for Get/Put/Write with user-timestamp (#5502) Summary: In previous https://github.com/facebook/rocksdb/issues/5079, we added user-specified timestamp to `DB::Get()` and `DB::Put()`. Limitation is that these two functions may cause extra memory allocation and key copy. The reason is that `WriteBatch` does not allocate extra memory for timestamps because it is not aware of timestamp size, and we did not provide an API to assign/update timestamp of each key within a `WriteBatch`. We address these issues in this PR by doing the following. 1. Add a `timestamp_size_` to `WriteBatch` so that `WriteBatch` can take timestamps into account when calling `WriteBatch::Put`, `WriteBatch::Delete`, etc. 2. Add APIs `WriteBatch::AssignTimestamp` and `WriteBatch::AssignTimestamps` so that application can assign/update timestamps for each key in a `WriteBatch`. 3. Avoid key copy in `GetImpl` by adding new constructor to `LookupKey`. Test plan (on devserver): ``` $make clean && COMPILE_WITH_ASAN=1 make -j32 all $./db_basic_test --gtest_filter=Timestamp/DBBasicTestWithTimestampWithParam.PutAndGet/* $make check ``` If the API extension looks good, I will add more unit tests. Some simple benchmark using db_bench. ``` $rm -rf /dev/shm/dbbench/* && TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillseq,readrandom -num=1000000 $rm -rf /dev/shm/dbbench/* && TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=1000000 -disable_wal=true ``` Master is at a78503bd6c80a3c4137df1962a972fe406b4d90b. ``` | | readrandom | fillrandom | | master | 15.53 MB/s | 25.97 MB/s | | PR5502 | 16.70 MB/s | 25.80 MB/s | ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/5502 Differential Revision: D16340894 Pulled By: riversand963 fbshipit-source-id: 51132cf792be07d1efc3ac33f5768c4ee2608bb8
5 years ago
LookupKey lkey(key, snapshot, read_options.timestamp);
PERF_TIMER_STOP(get_snapshot_time);
bool skip_memtable = (read_options.read_tier == kPersistedTier &&
has_unpersisted_data_.load(std::memory_order_relaxed));
bool done = false;
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
std::string* timestamp = ts_sz > 0 ? get_impl_options.timestamp : nullptr;
if (!skip_memtable) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
// Get value associated with key
if (get_impl_options.get_value) {
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
if (sv->mem->Get(lkey, get_impl_options.value->GetSelf(), timestamp, &s,
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
&merge_context, &max_covering_tombstone_seq,
read_options, get_impl_options.callback,
get_impl_options.is_blob_index)) {
done = true;
get_impl_options.value->PinSelf();
RecordTick(stats_, MEMTABLE_HIT);
} else if ((s.ok() || s.IsMergeInProgress()) &&
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
sv->imm->Get(lkey, get_impl_options.value->GetSelf(),
timestamp, &s, &merge_context,
&max_covering_tombstone_seq, read_options,
get_impl_options.callback,
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
get_impl_options.is_blob_index)) {
done = true;
get_impl_options.value->PinSelf();
RecordTick(stats_, MEMTABLE_HIT);
}
} else {
// Get Merge Operands associated with key, Merge Operands should not be
// merged and raw values should be returned to the user.
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
if (sv->mem->Get(lkey, /*value*/ nullptr, /*timestamp=*/nullptr, &s,
&merge_context, &max_covering_tombstone_seq,
read_options, nullptr, nullptr, false)) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
done = true;
RecordTick(stats_, MEMTABLE_HIT);
} else if ((s.ok() || s.IsMergeInProgress()) &&
sv->imm->GetMergeOperands(lkey, &s, &merge_context,
&max_covering_tombstone_seq,
read_options)) {
done = true;
RecordTick(stats_, MEMTABLE_HIT);
}
}
if (!done && !s.ok() && !s.IsMergeInProgress()) {
ReturnAndCleanupSuperVersion(cfd, sv);
return s;
}
}
if (!done) {
PERF_TIMER_GUARD(get_from_output_files_time);
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
sv->current->Get(
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
read_options, lkey, get_impl_options.value, timestamp, &s,
&merge_context, &max_covering_tombstone_seq,
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
get_impl_options.get_value ? get_impl_options.value_found : nullptr,
nullptr, nullptr,
get_impl_options.get_value ? get_impl_options.callback : nullptr,
get_impl_options.get_value ? get_impl_options.is_blob_index : nullptr,
get_impl_options.get_value);
RecordTick(stats_, MEMTABLE_MISS);
}
{
PERF_TIMER_GUARD(get_post_process_time);
ReturnAndCleanupSuperVersion(cfd, sv);
RecordTick(stats_, NUMBER_KEYS_READ);
size_t size = 0;
if (s.ok()) {
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
if (get_impl_options.get_value) {
size = get_impl_options.value->size();
} else {
// Return all merge operands for get_impl_options.key
*get_impl_options.number_of_operands =
static_cast<int>(merge_context.GetNumOperands());
if (*get_impl_options.number_of_operands >
get_impl_options.get_merge_operands_options
->expected_max_number_of_operands) {
s = Status::Incomplete(
Status::SubCode::KMergeOperandsInsufficientCapacity);
} else {
for (const Slice& sl : merge_context.GetOperands()) {
size += sl.size();
get_impl_options.merge_operands->PinSelf(sl);
get_impl_options.merge_operands++;
}
}
}
RecordTick(stats_, BYTES_READ, size);
PERF_COUNTER_ADD(get_read_bytes, size);
}
RecordInHistogram(stats_, BYTES_PER_READ, size);
}
return s;
}
std::vector<Status> DBImpl::MultiGet(
const ReadOptions& read_options,
const std::vector<ColumnFamilyHandle*>& column_family,
const std::vector<Slice>& keys, std::vector<std::string>* values) {
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
return MultiGet(read_options, column_family, keys, values,
/*timestamps=*/nullptr);
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
}
std::vector<Status> DBImpl::MultiGet(
const ReadOptions& read_options,
const std::vector<ColumnFamilyHandle*>& column_family,
const std::vector<Slice>& keys, std::vector<std::string>* values,
std::vector<std::string>* timestamps) {
PERF_CPU_TIMER_GUARD(get_cpu_nanos, immutable_db_options_.clock);
StopWatch sw(immutable_db_options_.clock, stats_, DB_MULTIGET);
PERF_TIMER_GUARD(get_snapshot_time);
#ifndef NDEBUG
for (const auto* cfh : column_family) {
assert(cfh);
const Comparator* const ucmp = cfh->GetComparator();
assert(ucmp);
if (ucmp->timestamp_size() > 0) {
assert(read_options.timestamp);
assert(ucmp->timestamp_size() == read_options.timestamp->size());
} else {
assert(!read_options.timestamp);
}
}
#endif // NDEBUG
if (tracer_) {
// TODO: This mutex should be removed later, to improve performance when
// tracing is enabled.
InstrumentedMutexLock lock(&trace_mutex_);
if (tracer_) {
// TODO: maybe handle the tracing status?
tracer_->MultiGet(column_family, keys).PermitUncheckedError();
}
}
SequenceNumber consistent_seqnum;
std::unordered_map<uint32_t, MultiGetColumnFamilyData> multiget_cf_data(
column_family.size());
for (auto cf : column_family) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(cf);
auto cfd = cfh->cfd();
if (multiget_cf_data.find(cfd->GetID()) == multiget_cf_data.end()) {
multiget_cf_data.emplace(cfd->GetID(),
MultiGetColumnFamilyData(cfh, nullptr));
}
}
std::function<MultiGetColumnFamilyData*(
std::unordered_map<uint32_t, MultiGetColumnFamilyData>::iterator&)>
iter_deref_lambda =
[](std::unordered_map<uint32_t, MultiGetColumnFamilyData>::iterator&
cf_iter) { return &cf_iter->second; };
bool unref_only =
MultiCFSnapshot<std::unordered_map<uint32_t, MultiGetColumnFamilyData>>(
read_options, nullptr, iter_deref_lambda, &multiget_cf_data,
&consistent_seqnum);
TEST_SYNC_POINT("DBImpl::MultiGet:AfterGetSeqNum1");
TEST_SYNC_POINT("DBImpl::MultiGet:AfterGetSeqNum2");
// Contain a list of merge operations if merge occurs.
MergeContext merge_context;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
12 years ago
// Note: this always resizes the values array
size_t num_keys = keys.size();
std::vector<Status> stat_list(num_keys);
values->resize(num_keys);
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
if (timestamps) {
timestamps->resize(num_keys);
}
// Keep track of bytes that we read for statistics-recording later
uint64_t bytes_read = 0;
PERF_TIMER_STOP(get_snapshot_time);
// For each of the given keys, apply the entire "get" process as follows:
// First look in the memtable, then in the immutable memtable (if any).
// s is both in/out. When in, s could either be OK or MergeInProgress.
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
12 years ago
// merge_operands will contain the sequence of merges in the latter case.
size_t num_found = 0;
size_t keys_read;
uint64_t curr_value_size = 0;
GetWithTimestampReadCallback timestamp_read_callback(0);
ReadCallback* read_callback = nullptr;
if (read_options.timestamp && read_options.timestamp->size() > 0) {
timestamp_read_callback.Refresh(consistent_seqnum);
read_callback = &timestamp_read_callback;
}
for (keys_read = 0; keys_read < num_keys; ++keys_read) {
merge_context.Clear();
Status& s = stat_list[keys_read];
std::string* value = &(*values)[keys_read];
std::string* timestamp = timestamps ? &(*timestamps)[keys_read] : nullptr;
LookupKey lkey(keys[keys_read], consistent_seqnum, read_options.timestamp);
auto cfh =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family[keys_read]);
Use only "local" range tombstones during Get (#4449) Summary: Previously, range tombstones were accumulated from every level, which was necessary if a range tombstone in a higher level covered a key in a lower level. However, RangeDelAggregator::AddTombstones's complexity is based on the number of tombstones that are currently stored in it, which is wasteful in the Get case, where we only need to know the highest sequence number of range tombstones that cover the key from higher levels, and compute the highest covering sequence number at the current level. This change introduces this optimization, and removes the use of RangeDelAggregator from the Get path. In the benchmark results, the following command was used to initialize the database: ``` ./db_bench -db=/dev/shm/5k-rts -use_existing_db=false -benchmarks=filluniquerandom -write_buffer_size=1048576 -compression_type=lz4 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -value_size=112 -key_size=16 -block_size=4096 -level_compaction_dynamic_level_bytes=true -num=5000000 -max_background_jobs=12 -benchmark_write_rate_limit=20971520 -range_tombstone_width=100 -writes_per_range_tombstone=100 -max_num_range_tombstones=50000 -bloom_bits=8 ``` ...and the following command was used to measure read throughput: ``` ./db_bench -db=/dev/shm/5k-rts/ -use_existing_db=true -benchmarks=readrandom -disable_auto_compactions=true -num=5000000 -reads=100000 -threads=32 ``` The filluniquerandom command was only run once, and the resulting database was used to measure read performance before and after the PR. Both binaries were compiled with `DEBUG_LEVEL=0`. Readrandom results before PR: ``` readrandom : 4.544 micros/op 220090 ops/sec; 16.9 MB/s (63103 of 100000 found) ``` Readrandom results after PR: ``` readrandom : 11.147 micros/op 89707 ops/sec; 6.9 MB/s (63103 of 100000 found) ``` So it's actually slower right now, but this PR paves the way for future optimizations (see #4493). ---- Pull Request resolved: https://github.com/facebook/rocksdb/pull/4449 Differential Revision: D10370575 Pulled By: abhimadan fbshipit-source-id: 9a2e152be1ef36969055c0e9eb4beb0d96c11f4d
6 years ago
SequenceNumber max_covering_tombstone_seq = 0;
auto mgd_iter = multiget_cf_data.find(cfh->cfd()->GetID());
assert(mgd_iter != multiget_cf_data.end());
auto mgd = mgd_iter->second;
auto super_version = mgd.super_version;
bool skip_memtable =
(read_options.read_tier == kPersistedTier &&
has_unpersisted_data_.load(std::memory_order_relaxed));
bool done = false;
if (!skip_memtable) {
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
if (super_version->mem->Get(lkey, value, timestamp, &s, &merge_context,
&max_covering_tombstone_seq, read_options,
read_callback)) {
done = true;
RecordTick(stats_, MEMTABLE_HIT);
} else if (super_version->imm->Get(lkey, value, timestamp, &s,
&merge_context,
&max_covering_tombstone_seq,
read_options, read_callback)) {
done = true;
RecordTick(stats_, MEMTABLE_HIT);
}
}
if (!done) {
PinnableSlice pinnable_val;
PERF_TIMER_GUARD(get_from_output_files_time);
super_version->current->Get(
read_options, lkey, &pinnable_val, timestamp, &s, &merge_context,
&max_covering_tombstone_seq, /*value_found=*/nullptr,
/*key_exists=*/nullptr,
/*seq=*/nullptr, read_callback);
value->assign(pinnable_val.data(), pinnable_val.size());
RecordTick(stats_, MEMTABLE_MISS);
}
if (s.ok()) {
bytes_read += value->size();
num_found++;
curr_value_size += value->size();
if (curr_value_size > read_options.value_size_soft_limit) {
while (++keys_read < num_keys) {
stat_list[keys_read] = Status::Aborted();
}
break;
}
}
if (read_options.deadline.count() &&
immutable_db_options_.clock->NowMicros() >
static_cast<uint64_t>(read_options.deadline.count())) {
break;
}
}
if (keys_read < num_keys) {
// The only reason to break out of the loop is when the deadline is
// exceeded
assert(immutable_db_options_.clock->NowMicros() >
static_cast<uint64_t>(read_options.deadline.count()));
for (++keys_read; keys_read < num_keys; ++keys_read) {
stat_list[keys_read] = Status::TimedOut();
}
}
// Post processing (decrement reference counts and record statistics)
PERF_TIMER_GUARD(get_post_process_time);
autovector<SuperVersion*> superversions_to_delete;
for (auto mgd_iter : multiget_cf_data) {
auto mgd = mgd_iter.second;
if (!unref_only) {
ReturnAndCleanupSuperVersion(mgd.cfd, mgd.super_version);
} else {
mgd.cfd->GetSuperVersion()->Unref();
}
}
RecordTick(stats_, NUMBER_MULTIGET_CALLS);
RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
RecordTick(stats_, NUMBER_MULTIGET_KEYS_FOUND, num_found);
RecordTick(stats_, NUMBER_MULTIGET_BYTES_READ, bytes_read);
RecordInHistogram(stats_, BYTES_PER_MULTIGET, bytes_read);
PERF_COUNTER_ADD(multiget_read_bytes, bytes_read);
PERF_TIMER_STOP(get_post_process_time);
return stat_list;
}
template <class T>
bool DBImpl::MultiCFSnapshot(
const ReadOptions& read_options, ReadCallback* callback,
std::function<MultiGetColumnFamilyData*(typename T::iterator&)>&
iter_deref_func,
T* cf_list, SequenceNumber* snapshot) {
PERF_TIMER_GUARD(get_snapshot_time);
bool last_try = false;
if (cf_list->size() == 1) {
// Fast path for a single column family. We can simply get the thread loca
// super version
auto cf_iter = cf_list->begin();
auto node = iter_deref_func(cf_iter);
node->super_version = GetAndRefSuperVersion(node->cfd);
if (read_options.snapshot != nullptr) {
// Note: In WritePrepared txns this is not necessary but not harmful
// either. Because prep_seq > snapshot => commit_seq > snapshot so if
// a snapshot is specified we should be fine with skipping seq numbers
// that are greater than that.
//
// In WriteUnprepared, we cannot set snapshot in the lookup key because we
// may skip uncommitted data that should be visible to the transaction for
// reading own writes.
*snapshot =
static_cast<const SnapshotImpl*>(read_options.snapshot)->number_;
if (callback) {
*snapshot = std::max(*snapshot, callback->max_visible_seq());
}
} else {
// Since we get and reference the super version before getting
// the snapshot number, without a mutex protection, it is possible
// that a memtable switch happened in the middle and not all the
// data for this snapshot is available. But it will contain all
// the data available in the super version we have, which is also
// a valid snapshot to read from.
// We shouldn't get snapshot before finding and referencing the super
// version because a flush happening in between may compact away data for
// the snapshot, but the snapshot is earlier than the data overwriting it,
// so users may see wrong results.
Fix TSAN build and re-enable the tests (#7386) Summary: Resolve TSAN build warnings and re-enable disabled TSAN tests. Not sure if it's a compiler issue or TSAN check issue. Switching from conditional operator to if-else mitigated the problem. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7386 Test Plan: run TSAN check 10 times in circleci. ``` WARNING: ThreadSanitizer: data race (pid=27735) Atomic write of size 8 at 0x7b54000005e8 by thread T32: #0 __tsan_atomic64_store <null> (db_test+0x4cee95) https://github.com/facebook/rocksdb/issues/1 std::__atomic_base<unsigned long>::store(unsigned long, std::memory_order) /usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/bits/atomic_base.h:374:2 (db_test+0x78460e) https://github.com/facebook/rocksdb/issues/2 rocksdb::VersionSet::SetLastSequence(unsigned long) /home/circleci/project/./db/version_set.h:1058:20 (db_test+0x78460e) ... Previous read of size 8 at 0x7b54000005e8 by thread T31: #0 bool rocksdb::DBImpl::MultiCFSnapshot<std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > > >(rocksdb::ReadOptions const&, rocksdb::ReadCallback*, std::function<rocksdb::DBImpl::MultiGetColumnFamilyData* (std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >::iterator&)>&, std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >*, unsigned long*) /home/circleci/project/db/db_impl/db_impl.cc (db_test+0x715087) ``` Reviewed By: ltamasi Differential Revision: D23725226 Pulled By: jay-zhuang fbshipit-source-id: a6d662a5ea68111246cd32ec95f3411a25f76bc6
4 years ago
if (last_seq_same_as_publish_seq_) {
*snapshot = versions_->LastSequence();
} else {
*snapshot = versions_->LastPublishedSequence();
}
}
} else {
// If we end up with the same issue of memtable geting sealed during 2
// consecutive retries, it means the write rate is very high. In that case
// its probably ok to take the mutex on the 3rd try so we can succeed for
// sure
constexpr int num_retries = 3;
for (int i = 0; i < num_retries; ++i) {
last_try = (i == num_retries - 1);
bool retry = false;
if (i > 0) {
for (auto cf_iter = cf_list->begin(); cf_iter != cf_list->end();
++cf_iter) {
auto node = iter_deref_func(cf_iter);
SuperVersion* super_version = node->super_version;
ColumnFamilyData* cfd = node->cfd;
if (super_version != nullptr) {
ReturnAndCleanupSuperVersion(cfd, super_version);
}
node->super_version = nullptr;
}
}
if (read_options.snapshot == nullptr) {
if (last_try) {
TEST_SYNC_POINT("DBImpl::MultiGet::LastTry");
// We're close to max number of retries. For the last retry,
// acquire the lock so we're sure to succeed
mutex_.Lock();
}
Fix TSAN build and re-enable the tests (#7386) Summary: Resolve TSAN build warnings and re-enable disabled TSAN tests. Not sure if it's a compiler issue or TSAN check issue. Switching from conditional operator to if-else mitigated the problem. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7386 Test Plan: run TSAN check 10 times in circleci. ``` WARNING: ThreadSanitizer: data race (pid=27735) Atomic write of size 8 at 0x7b54000005e8 by thread T32: #0 __tsan_atomic64_store <null> (db_test+0x4cee95) https://github.com/facebook/rocksdb/issues/1 std::__atomic_base<unsigned long>::store(unsigned long, std::memory_order) /usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/bits/atomic_base.h:374:2 (db_test+0x78460e) https://github.com/facebook/rocksdb/issues/2 rocksdb::VersionSet::SetLastSequence(unsigned long) /home/circleci/project/./db/version_set.h:1058:20 (db_test+0x78460e) ... Previous read of size 8 at 0x7b54000005e8 by thread T31: #0 bool rocksdb::DBImpl::MultiCFSnapshot<std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > > >(rocksdb::ReadOptions const&, rocksdb::ReadCallback*, std::function<rocksdb::DBImpl::MultiGetColumnFamilyData* (std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >::iterator&)>&, std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >*, unsigned long*) /home/circleci/project/db/db_impl/db_impl.cc (db_test+0x715087) ``` Reviewed By: ltamasi Differential Revision: D23725226 Pulled By: jay-zhuang fbshipit-source-id: a6d662a5ea68111246cd32ec95f3411a25f76bc6
4 years ago
if (last_seq_same_as_publish_seq_) {
*snapshot = versions_->LastSequence();
} else {
*snapshot = versions_->LastPublishedSequence();
}
} else {
*snapshot =
static_cast_with_check<const SnapshotImpl>(read_options.snapshot)
->number_;
}
for (auto cf_iter = cf_list->begin(); cf_iter != cf_list->end();
++cf_iter) {
auto node = iter_deref_func(cf_iter);
if (!last_try) {
node->super_version = GetAndRefSuperVersion(node->cfd);
} else {
node->super_version = node->cfd->GetSuperVersion()->Ref();
}
TEST_SYNC_POINT("DBImpl::MultiGet::AfterRefSV");
if (read_options.snapshot != nullptr || last_try) {
// If user passed a snapshot, then we don't care if a memtable is
// sealed or compaction happens because the snapshot would ensure
// that older key versions are kept around. If this is the last
// retry, then we have the lock so nothing bad can happen
continue;
}
// We could get the earliest sequence number for the whole list of
// memtables, which will include immutable memtables as well, but that
// might be tricky to maintain in case we decide, in future, to do
// memtable compaction.
if (!last_try) {
SequenceNumber seq =
node->super_version->mem->GetEarliestSequenceNumber();
if (seq > *snapshot) {
retry = true;
break;
}
}
}
if (!retry) {
if (last_try) {
mutex_.Unlock();
}
break;
}
}
}
// Keep track of bytes that we read for statistics-recording later
PERF_TIMER_STOP(get_snapshot_time);
return last_try;
}
void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
ColumnFamilyHandle** column_families, const Slice* keys,
PinnableSlice* values, Status* statuses,
const bool sorted_input) {
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
return MultiGet(read_options, num_keys, column_families, keys, values,
/*timestamps=*/nullptr, statuses, sorted_input);
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
}
void DBImpl::MultiGet(const ReadOptions& read_options, const size_t num_keys,
ColumnFamilyHandle** column_families, const Slice* keys,
PinnableSlice* values, std::string* timestamps,
Status* statuses, const bool sorted_input) {
if (num_keys == 0) {
return;
}
#ifndef NDEBUG
for (size_t i = 0; i < num_keys; ++i) {
ColumnFamilyHandle* cfh = column_families[i];
assert(cfh);
const Comparator* const ucmp = cfh->GetComparator();
assert(ucmp);
if (ucmp->timestamp_size() > 0) {
assert(read_options.timestamp);
assert(read_options.timestamp->size() == ucmp->timestamp_size());
} else {
assert(!read_options.timestamp);
}
}
#endif // NDEBUG
if (tracer_) {
// TODO: This mutex should be removed later, to improve performance when
// tracing is enabled.
InstrumentedMutexLock lock(&trace_mutex_);
if (tracer_) {
// TODO: maybe handle the tracing status?
tracer_->MultiGet(num_keys, column_families, keys).PermitUncheckedError();
}
}
autovector<KeyContext, MultiGetContext::MAX_BATCH_SIZE> key_context;
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE> sorted_keys;
sorted_keys.resize(num_keys);
for (size_t i = 0; i < num_keys; ++i) {
key_context.emplace_back(column_families[i], keys[i], &values[i],
timestamps ? &timestamps[i] : nullptr,
&statuses[i]);
}
for (size_t i = 0; i < num_keys; ++i) {
sorted_keys[i] = &key_context[i];
}
PrepareMultiGetKeys(num_keys, sorted_input, &sorted_keys);
autovector<MultiGetColumnFamilyData, MultiGetContext::MAX_BATCH_SIZE>
multiget_cf_data;
size_t cf_start = 0;
ColumnFamilyHandle* cf = sorted_keys[0]->column_family;
for (size_t i = 0; i < num_keys; ++i) {
KeyContext* key_ctx = sorted_keys[i];
if (key_ctx->column_family != cf) {
multiget_cf_data.emplace_back(cf, cf_start, i - cf_start, nullptr);
cf_start = i;
cf = key_ctx->column_family;
}
}
multiget_cf_data.emplace_back(cf, cf_start, num_keys - cf_start, nullptr);
std::function<MultiGetColumnFamilyData*(
autovector<MultiGetColumnFamilyData,
MultiGetContext::MAX_BATCH_SIZE>::iterator&)>
iter_deref_lambda =
[](autovector<MultiGetColumnFamilyData,
MultiGetContext::MAX_BATCH_SIZE>::iterator& cf_iter) {
return &(*cf_iter);
};
SequenceNumber consistent_seqnum;
bool unref_only = MultiCFSnapshot<
autovector<MultiGetColumnFamilyData, MultiGetContext::MAX_BATCH_SIZE>>(
read_options, nullptr, iter_deref_lambda, &multiget_cf_data,
&consistent_seqnum);
GetWithTimestampReadCallback timestamp_read_callback(0);
ReadCallback* read_callback = nullptr;
if (read_options.timestamp && read_options.timestamp->size() > 0) {
timestamp_read_callback.Refresh(consistent_seqnum);
read_callback = &timestamp_read_callback;
}
Status s;
auto cf_iter = multiget_cf_data.begin();
for (; cf_iter != multiget_cf_data.end(); ++cf_iter) {
s = MultiGetImpl(read_options, cf_iter->start, cf_iter->num_keys,
&sorted_keys, cf_iter->super_version, consistent_seqnum,
read_callback);
if (!s.ok()) {
break;
}
}
if (!s.ok()) {
assert(s.IsTimedOut() || s.IsAborted());
for (++cf_iter; cf_iter != multiget_cf_data.end(); ++cf_iter) {
for (size_t i = cf_iter->start; i < cf_iter->start + cf_iter->num_keys;
++i) {
*sorted_keys[i]->s = s;
}
}
}
for (const auto& iter : multiget_cf_data) {
if (!unref_only) {
ReturnAndCleanupSuperVersion(iter.cfd, iter.super_version);
} else {
iter.cfd->GetSuperVersion()->Unref();
}
}
}
namespace {
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
// Order keys by CF ID, followed by key contents
struct CompareKeyContext {
inline bool operator()(const KeyContext* lhs, const KeyContext* rhs) {
ColumnFamilyHandleImpl* cfh =
static_cast<ColumnFamilyHandleImpl*>(lhs->column_family);
uint32_t cfd_id1 = cfh->cfd()->GetID();
const Comparator* comparator = cfh->cfd()->user_comparator();
cfh = static_cast<ColumnFamilyHandleImpl*>(rhs->column_family);
uint32_t cfd_id2 = cfh->cfd()->GetID();
if (cfd_id1 < cfd_id2) {
return true;
} else if (cfd_id1 > cfd_id2) {
return false;
}
// Both keys are from the same column family
int cmp = comparator->CompareWithoutTimestamp(
*(lhs->key), /*a_has_ts=*/false, *(rhs->key), /*b_has_ts=*/false);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
if (cmp < 0) {
return true;
}
return false;
}
};
} // anonymous namespace
void DBImpl::PrepareMultiGetKeys(
size_t num_keys, bool sorted_input,
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys) {
if (sorted_input) {
#ifndef NDEBUG
assert(std::is_sorted(sorted_keys->begin(), sorted_keys->end(),
CompareKeyContext()));
#endif
return;
}
std::sort(sorted_keys->begin(), sorted_keys->begin() + num_keys,
CompareKeyContext());
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
void DBImpl::MultiGet(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const size_t num_keys,
const Slice* keys, PinnableSlice* values,
Status* statuses, const bool sorted_input) {
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
return MultiGet(read_options, column_family, num_keys, keys, values,
/*timestamp=*/nullptr, statuses, sorted_input);
}
void DBImpl::MultiGet(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const size_t num_keys,
const Slice* keys, PinnableSlice* values,
std::string* timestamps, Status* statuses,
const bool sorted_input) {
if (tracer_) {
// TODO: This mutex should be removed later, to improve performance when
// tracing is enabled.
InstrumentedMutexLock lock(&trace_mutex_);
if (tracer_) {
// TODO: maybe handle the tracing status?
tracer_->MultiGet(num_keys, column_family, keys).PermitUncheckedError();
}
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
autovector<KeyContext, MultiGetContext::MAX_BATCH_SIZE> key_context;
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE> sorted_keys;
sorted_keys.resize(num_keys);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
for (size_t i = 0; i < num_keys; ++i) {
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
5 years ago
key_context.emplace_back(column_family, keys[i], &values[i],
timestamps ? &timestamps[i] : nullptr,
&statuses[i]);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
for (size_t i = 0; i < num_keys; ++i) {
sorted_keys[i] = &key_context[i];
}
PrepareMultiGetKeys(num_keys, sorted_input, &sorted_keys);
MultiGetWithCallback(read_options, column_family, nullptr, &sorted_keys);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
void DBImpl::MultiGetWithCallback(
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
const ReadOptions& read_options, ColumnFamilyHandle* column_family,
ReadCallback* callback,
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys) {
std::array<MultiGetColumnFamilyData, 1> multiget_cf_data;
multiget_cf_data[0] = MultiGetColumnFamilyData(column_family, nullptr);
std::function<MultiGetColumnFamilyData*(
std::array<MultiGetColumnFamilyData, 1>::iterator&)>
iter_deref_lambda =
[](std::array<MultiGetColumnFamilyData, 1>::iterator& cf_iter) {
return &(*cf_iter);
};
size_t num_keys = sorted_keys->size();
SequenceNumber consistent_seqnum;
bool unref_only = MultiCFSnapshot<std::array<MultiGetColumnFamilyData, 1>>(
read_options, callback, iter_deref_lambda, &multiget_cf_data,
&consistent_seqnum);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
#ifndef NDEBUG
assert(!unref_only);
#else
// Silence unused variable warning
(void)unref_only;
#endif // NDEBUG
if (callback && read_options.snapshot == nullptr) {
// The unprep_seqs are not published for write unprepared, so it could be
// that max_visible_seq is larger. Seek to the std::max of the two.
// However, we still want our callback to contain the actual snapshot so
// that it can do the correct visibility filtering.
callback->Refresh(consistent_seqnum);
// Internally, WriteUnpreparedTxnReadCallback::Refresh would set
// max_visible_seq = max(max_visible_seq, snapshot)
//
// Currently, the commented out assert is broken by
// InvalidSnapshotReadCallback, but if write unprepared recovery followed
// the regular transaction flow, then this special read callback would not
// be needed.
//
// assert(callback->max_visible_seq() >= snapshot);
consistent_seqnum = callback->max_visible_seq();
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
GetWithTimestampReadCallback timestamp_read_callback(0);
ReadCallback* read_callback = callback;
if (read_options.timestamp && read_options.timestamp->size() > 0) {
assert(!read_callback); // timestamp with callback is not supported
timestamp_read_callback.Refresh(consistent_seqnum);
read_callback = &timestamp_read_callback;
}
Status s = MultiGetImpl(read_options, 0, num_keys, sorted_keys,
multiget_cf_data[0].super_version, consistent_seqnum,
read_callback);
assert(s.ok() || s.IsTimedOut() || s.IsAborted());
ReturnAndCleanupSuperVersion(multiget_cf_data[0].cfd,
multiget_cf_data[0].super_version);
}
// The actual implementation of batched MultiGet. Parameters -
// start_key - Index in the sorted_keys vector to start processing from
// num_keys - Number of keys to lookup, starting with sorted_keys[start_key]
// sorted_keys - The entire batch of sorted keys for this CF
//
// The per key status is returned in the KeyContext structures pointed to by
// sorted_keys. An overall Status is also returned, with the only possible
// values being Status::OK() and Status::TimedOut(). The latter indicates
// that the call exceeded read_options.deadline
Status DBImpl::MultiGetImpl(
const ReadOptions& read_options, size_t start_key, size_t num_keys,
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys,
SuperVersion* super_version, SequenceNumber snapshot,
ReadCallback* callback) {
PERF_CPU_TIMER_GUARD(get_cpu_nanos, immutable_db_options_.clock);
StopWatch sw(immutable_db_options_.clock, stats_, DB_MULTIGET);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
// For each of the given keys, apply the entire "get" process as follows:
// First look in the memtable, then in the immutable memtable (if any).
// s is both in/out. When in, s could either be OK or MergeInProgress.
// merge_operands will contain the sequence of merges in the latter case.
size_t keys_left = num_keys;
Status s;
uint64_t curr_value_size = 0;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
while (keys_left) {
if (read_options.deadline.count() &&
immutable_db_options_.clock->NowMicros() >
static_cast<uint64_t>(read_options.deadline.count())) {
s = Status::TimedOut();
break;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
size_t batch_size = (keys_left > MultiGetContext::MAX_BATCH_SIZE)
? MultiGetContext::MAX_BATCH_SIZE
: keys_left;
MultiGetContext ctx(sorted_keys, start_key + num_keys - keys_left,
batch_size, snapshot, read_options);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
MultiGetRange range = ctx.GetMultiGetRange();
range.AddValueSize(curr_value_size);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
bool lookup_current = false;
keys_left -= batch_size;
for (auto mget_iter = range.begin(); mget_iter != range.end();
++mget_iter) {
MultiGet batching in memtable (#5818) Summary: RocksDB has a MultiGet() API that implements batched key lookup for higher performance (https://github.com/facebook/rocksdb/blob/master/include/rocksdb/db.h#L468). Currently, batching is implemented in BlockBasedTableReader::MultiGet() for SST file lookups. One of the ways it improves performance is by pipelining bloom filter lookups (by prefetching required cachelines for all the keys in the batch, and then doing the probe) and thus hiding the cache miss latency. The same concept can be extended to the memtable as well. This PR involves implementing a pipelined bloom filter lookup in DynamicBloom, and implementing MemTable::MultiGet() that can leverage it. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5818 Test Plan: Existing tests Performance Test: Ran the below command which fills up the memtable and makes sure there are no flushes and then call multiget. Ran it on master and on the new change and see atleast 1% performance improvement across all the test runs I did. Sometimes the improvement was upto 5%. TEST_TMPDIR=/data/users/$USER/benchmarks/feature/ numactl -C 10 ./db_bench -benchmarks="fillseq,multireadrandom" -num=600000 -compression_type="none" -level_compaction_dynamic_level_bytes -write_buffer_size=200000000 -target_file_size_base=200000000 -max_bytes_for_level_base=16777216 -reads=90000 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 -statistics -memtable_whole_key_filtering=true -memtable_bloom_size_ratio=10 Differential Revision: D17578869 Pulled By: vjnadimpalli fbshipit-source-id: 23dc651d9bf49db11d22375bf435708875a1f192
5 years ago
mget_iter->merge_context.Clear();
*mget_iter->s = Status::OK();
}
bool skip_memtable =
(read_options.read_tier == kPersistedTier &&
has_unpersisted_data_.load(std::memory_order_relaxed));
if (!skip_memtable) {
super_version->mem->MultiGet(read_options, &range, callback);
MultiGet batching in memtable (#5818) Summary: RocksDB has a MultiGet() API that implements batched key lookup for higher performance (https://github.com/facebook/rocksdb/blob/master/include/rocksdb/db.h#L468). Currently, batching is implemented in BlockBasedTableReader::MultiGet() for SST file lookups. One of the ways it improves performance is by pipelining bloom filter lookups (by prefetching required cachelines for all the keys in the batch, and then doing the probe) and thus hiding the cache miss latency. The same concept can be extended to the memtable as well. This PR involves implementing a pipelined bloom filter lookup in DynamicBloom, and implementing MemTable::MultiGet() that can leverage it. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5818 Test Plan: Existing tests Performance Test: Ran the below command which fills up the memtable and makes sure there are no flushes and then call multiget. Ran it on master and on the new change and see atleast 1% performance improvement across all the test runs I did. Sometimes the improvement was upto 5%. TEST_TMPDIR=/data/users/$USER/benchmarks/feature/ numactl -C 10 ./db_bench -benchmarks="fillseq,multireadrandom" -num=600000 -compression_type="none" -level_compaction_dynamic_level_bytes -write_buffer_size=200000000 -target_file_size_base=200000000 -max_bytes_for_level_base=16777216 -reads=90000 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 -statistics -memtable_whole_key_filtering=true -memtable_bloom_size_ratio=10 Differential Revision: D17578869 Pulled By: vjnadimpalli fbshipit-source-id: 23dc651d9bf49db11d22375bf435708875a1f192
5 years ago
if (!range.empty()) {
super_version->imm->MultiGet(read_options, &range, callback);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
MultiGet batching in memtable (#5818) Summary: RocksDB has a MultiGet() API that implements batched key lookup for higher performance (https://github.com/facebook/rocksdb/blob/master/include/rocksdb/db.h#L468). Currently, batching is implemented in BlockBasedTableReader::MultiGet() for SST file lookups. One of the ways it improves performance is by pipelining bloom filter lookups (by prefetching required cachelines for all the keys in the batch, and then doing the probe) and thus hiding the cache miss latency. The same concept can be extended to the memtable as well. This PR involves implementing a pipelined bloom filter lookup in DynamicBloom, and implementing MemTable::MultiGet() that can leverage it. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5818 Test Plan: Existing tests Performance Test: Ran the below command which fills up the memtable and makes sure there are no flushes and then call multiget. Ran it on master and on the new change and see atleast 1% performance improvement across all the test runs I did. Sometimes the improvement was upto 5%. TEST_TMPDIR=/data/users/$USER/benchmarks/feature/ numactl -C 10 ./db_bench -benchmarks="fillseq,multireadrandom" -num=600000 -compression_type="none" -level_compaction_dynamic_level_bytes -write_buffer_size=200000000 -target_file_size_base=200000000 -max_bytes_for_level_base=16777216 -reads=90000 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 -statistics -memtable_whole_key_filtering=true -memtable_bloom_size_ratio=10 Differential Revision: D17578869 Pulled By: vjnadimpalli fbshipit-source-id: 23dc651d9bf49db11d22375bf435708875a1f192
5 years ago
if (!range.empty()) {
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
lookup_current = true;
MultiGet batching in memtable (#5818) Summary: RocksDB has a MultiGet() API that implements batched key lookup for higher performance (https://github.com/facebook/rocksdb/blob/master/include/rocksdb/db.h#L468). Currently, batching is implemented in BlockBasedTableReader::MultiGet() for SST file lookups. One of the ways it improves performance is by pipelining bloom filter lookups (by prefetching required cachelines for all the keys in the batch, and then doing the probe) and thus hiding the cache miss latency. The same concept can be extended to the memtable as well. This PR involves implementing a pipelined bloom filter lookup in DynamicBloom, and implementing MemTable::MultiGet() that can leverage it. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5818 Test Plan: Existing tests Performance Test: Ran the below command which fills up the memtable and makes sure there are no flushes and then call multiget. Ran it on master and on the new change and see atleast 1% performance improvement across all the test runs I did. Sometimes the improvement was upto 5%. TEST_TMPDIR=/data/users/$USER/benchmarks/feature/ numactl -C 10 ./db_bench -benchmarks="fillseq,multireadrandom" -num=600000 -compression_type="none" -level_compaction_dynamic_level_bytes -write_buffer_size=200000000 -target_file_size_base=200000000 -max_bytes_for_level_base=16777216 -reads=90000 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 -statistics -memtable_whole_key_filtering=true -memtable_bloom_size_ratio=10 Differential Revision: D17578869 Pulled By: vjnadimpalli fbshipit-source-id: 23dc651d9bf49db11d22375bf435708875a1f192
5 years ago
uint64_t left = range.KeysLeft();
RecordTick(stats_, MEMTABLE_MISS, left);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
}
if (lookup_current) {
PERF_TIMER_GUARD(get_from_output_files_time);
super_version->current->MultiGet(read_options, &range, callback);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
curr_value_size = range.GetValueSize();
if (curr_value_size > read_options.value_size_soft_limit) {
s = Status::Aborted();
break;
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
// Post processing (decrement reference counts and record statistics)
PERF_TIMER_GUARD(get_post_process_time);
size_t num_found = 0;
uint64_t bytes_read = 0;
for (size_t i = start_key; i < start_key + num_keys - keys_left; ++i) {
KeyContext* key = (*sorted_keys)[i];
if (key->s->ok()) {
bytes_read += key->value->size();
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
num_found++;
}
}
if (keys_left) {
assert(s.IsTimedOut() || s.IsAborted());
for (size_t i = start_key + num_keys - keys_left; i < start_key + num_keys;
++i) {
KeyContext* key = (*sorted_keys)[i];
*key->s = s;
}
}
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
RecordTick(stats_, NUMBER_MULTIGET_CALLS);
RecordTick(stats_, NUMBER_MULTIGET_KEYS_READ, num_keys);
RecordTick(stats_, NUMBER_MULTIGET_KEYS_FOUND, num_found);
RecordTick(stats_, NUMBER_MULTIGET_BYTES_READ, bytes_read);
RecordInHistogram(stats_, BYTES_PER_MULTIGET, bytes_read);
PERF_COUNTER_ADD(multiget_read_bytes, bytes_read);
PERF_TIMER_STOP(get_post_process_time);
return s;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
6 years ago
}
Status DBImpl::CreateColumnFamily(const ColumnFamilyOptions& cf_options,
const std::string& column_family,
ColumnFamilyHandle** handle) {
assert(handle != nullptr);
Status s = CreateColumnFamilyImpl(cf_options, column_family, handle);
if (s.ok()) {
s = WriteOptionsFile(true /*need_mutex_lock*/,
true /*need_enter_write_thread*/);
}
return s;
}
Status DBImpl::CreateColumnFamilies(
const ColumnFamilyOptions& cf_options,
const std::vector<std::string>& column_family_names,
std::vector<ColumnFamilyHandle*>* handles) {
assert(handles != nullptr);
handles->clear();
size_t num_cf = column_family_names.size();
Status s;
bool success_once = false;
for (size_t i = 0; i < num_cf; i++) {
ColumnFamilyHandle* handle;
s = CreateColumnFamilyImpl(cf_options, column_family_names[i], &handle);
if (!s.ok()) {
break;
}
handles->push_back(handle);
success_once = true;
}
if (success_once) {
Status persist_options_status = WriteOptionsFile(
true /*need_mutex_lock*/, true /*need_enter_write_thread*/);
if (s.ok() && !persist_options_status.ok()) {
s = persist_options_status;
}
}
return s;
}
Status DBImpl::CreateColumnFamilies(
const std::vector<ColumnFamilyDescriptor>& column_families,
std::vector<ColumnFamilyHandle*>* handles) {
assert(handles != nullptr);
handles->clear();
size_t num_cf = column_families.size();
Status s;
bool success_once = false;
for (size_t i = 0; i < num_cf; i++) {
ColumnFamilyHandle* handle;
s = CreateColumnFamilyImpl(column_families[i].options,
column_families[i].name, &handle);
if (!s.ok()) {
break;
}
handles->push_back(handle);
success_once = true;
}
if (success_once) {
Status persist_options_status = WriteOptionsFile(
true /*need_mutex_lock*/, true /*need_enter_write_thread*/);
if (s.ok() && !persist_options_status.ok()) {
s = persist_options_status;
}
}
return s;
}
Status DBImpl::CreateColumnFamilyImpl(const ColumnFamilyOptions& cf_options,
const std::string& column_family_name,
ColumnFamilyHandle** handle) {
Status s;
*handle = nullptr;
DBOptions db_options =
BuildDBOptions(immutable_db_options_, mutable_db_options_);
s = ColumnFamilyData::ValidateOptions(db_options, cf_options);
if (s.ok()) {
for (auto& cf_path : cf_options.cf_paths) {
s = env_->CreateDirIfMissing(cf_path.path);
if (!s.ok()) {
break;
}
}
}
if (!s.ok()) {
return s;
}
SuperVersionContext sv_context(/* create_superversion */ true);
{
InstrumentedMutexLock l(&mutex_);
if (versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name) !=
nullptr) {
return Status::InvalidArgument("Column family already exists");
}
VersionEdit edit;
edit.AddColumnFamily(column_family_name);
uint32_t new_id = versions_->GetColumnFamilySet()->GetNextColumnFamilyID();
edit.SetColumnFamily(new_id);
edit.SetLogNumber(logfile_number_);
edit.SetComparatorName(cf_options.comparator->Name());
// LogAndApply will both write the creation in MANIFEST and create
// ColumnFamilyData object
{ // write thread
WriteThread::Writer w;
write_thread_.EnterUnbatched(&w, &mutex_);
// LogAndApply will both write the creation in MANIFEST and create
// ColumnFamilyData object
s = versions_->LogAndApply(nullptr, MutableCFOptions(cf_options), &edit,
&mutex_, directories_.GetDbDir(), false,
&cf_options);
write_thread_.ExitUnbatched(&w);
}
if (s.ok()) {
auto* cfd =
versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name);
assert(cfd != nullptr);
std::map<std::string, std::shared_ptr<FSDirectory>> dummy_created_dirs;
s = cfd->AddDirectories(&dummy_created_dirs);
}
if (s.ok()) {
single_column_family_mode_ = false;
auto* cfd =
versions_->GetColumnFamilySet()->GetColumnFamily(column_family_name);
assert(cfd != nullptr);
InstallSuperVersionAndScheduleWork(cfd, &sv_context,
*cfd->GetLatestMutableCFOptions());
if (!cfd->mem()->IsSnapshotSupported()) {
is_snapshot_supported_ = false;
}
cfd->set_initialized();
*handle = new ColumnFamilyHandleImpl(cfd, this, &mutex_);
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Created column family [%s] (ID %u)",
column_family_name.c_str(), (unsigned)cfd->GetID());
} else {
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Creating column family [%s] FAILED -- %s",
column_family_name.c_str(), s.ToString().c_str());
}
} // InstrumentedMutexLock l(&mutex_)
sv_context.Clean();
// this is outside the mutex
if (s.ok()) {
NewThreadStatusCfInfo(
static_cast_with_check<ColumnFamilyHandleImpl>(*handle)->cfd());
}
return s;
}
Status DBImpl::DropColumnFamily(ColumnFamilyHandle* column_family) {
assert(column_family != nullptr);
Status s = DropColumnFamilyImpl(column_family);
if (s.ok()) {
s = WriteOptionsFile(true /*need_mutex_lock*/,
true /*need_enter_write_thread*/);
}
return s;
}
Status DBImpl::DropColumnFamilies(
const std::vector<ColumnFamilyHandle*>& column_families) {
Status s;
bool success_once = false;
for (auto* handle : column_families) {
s = DropColumnFamilyImpl(handle);
if (!s.ok()) {
break;
}
success_once = true;
}
if (success_once) {
Status persist_options_status = WriteOptionsFile(
true /*need_mutex_lock*/, true /*need_enter_write_thread*/);
if (s.ok() && !persist_options_status.ok()) {
s = persist_options_status;
}
}
return s;
}
Status DBImpl::DropColumnFamilyImpl(ColumnFamilyHandle* column_family) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
auto cfd = cfh->cfd();
if (cfd->GetID() == 0) {
return Status::InvalidArgument("Can't drop default column family");
}
bool cf_support_snapshot = cfd->mem()->IsSnapshotSupported();
VersionEdit edit;
edit.DropColumnFamily();
edit.SetColumnFamily(cfd->GetID());
Status s;
{
InstrumentedMutexLock l(&mutex_);
if (cfd->IsDropped()) {
s = Status::InvalidArgument("Column family already dropped!\n");
}
if (s.ok()) {
// we drop column family from a single write thread
WriteThread::Writer w;
write_thread_.EnterUnbatched(&w, &mutex_);
s = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(), &edit,
&mutex_);
write_thread_.ExitUnbatched(&w);
}
if (s.ok()) {
auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
max_total_in_memory_state_ -= mutable_cf_options->write_buffer_size *
mutable_cf_options->max_write_buffer_number;
}
if (!cf_support_snapshot) {
// Dropped Column Family doesn't support snapshot. Need to recalculate
// is_snapshot_supported_.
bool new_is_snapshot_supported = true;
for (auto c : *versions_->GetColumnFamilySet()) {
if (!c->IsDropped() && !c->mem()->IsSnapshotSupported()) {
new_is_snapshot_supported = false;
break;
}
}
is_snapshot_supported_ = new_is_snapshot_supported;
}
bg_cv_.SignalAll();
}
if (s.ok()) {
// Note that here we erase the associated cf_info of the to-be-dropped
// cfd before its ref-count goes to zero to avoid having to erase cf_info
// later inside db_mutex.
EraseThreadStatusCfInfo(cfd);
assert(cfd->IsDropped());
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Dropped column family with id %u\n", cfd->GetID());
} else {
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Dropping column family with id %u FAILED -- %s\n",
cfd->GetID(), s.ToString().c_str());
}
return s;
}
bool DBImpl::KeyMayExist(const ReadOptions& read_options,
ColumnFamilyHandle* column_family, const Slice& key,
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
std::string* value, std::string* timestamp,
bool* value_found) {
assert(value != nullptr);
if (value_found != nullptr) {
// falsify later if key-may-exist but can't fetch value
*value_found = true;
}
ReadOptions roptions = read_options;
roptions.read_tier = kBlockCacheTier; // read from block cache only
PinnableSlice pinnable_val;
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
GetImplOptions get_impl_options;
get_impl_options.column_family = column_family;
get_impl_options.value = &pinnable_val;
get_impl_options.value_found = value_found;
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
get_impl_options.timestamp = timestamp;
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
5 years ago
auto s = GetImpl(roptions, key, get_impl_options);
value->assign(pinnable_val.data(), pinnable_val.size());
// If block_cache is enabled and the index block of the table didn't
// not present in block_cache, the return value will be Status::Incomplete.
// In this case, key may still exist in the table.
return s.ok() || s.IsIncomplete();
}
Iterator* DBImpl::NewIterator(const ReadOptions& read_options,
ColumnFamilyHandle* column_family) {
if (read_options.managed) {
return NewErrorIterator(
Status::NotSupported("Managed iterator is not supported anymore."));
}
Iterator* result = nullptr;
if (read_options.read_tier == kPersistedTier) {
return NewErrorIterator(Status::NotSupported(
"ReadTier::kPersistedData is not yet supported in iterators."));
}
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
7 years ago
// if iterator wants internal keys, we can only proceed if
// we can guarantee the deletes haven't been processed yet
if (immutable_db_options_.preserve_deletes &&
read_options.iter_start_seqnum > 0 &&
read_options.iter_start_seqnum < preserve_deletes_seqnum_.load()) {
return NewErrorIterator(Status::InvalidArgument(
"Iterator requested internal keys which are too old and are not"
" guaranteed to be preserved, try larger iter_start_seqnum opt."));
}
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
ColumnFamilyData* cfd = cfh->cfd();
assert(cfd != nullptr);
ReadCallback* read_callback = nullptr; // No read callback provided.
if (read_options.tailing) {
#ifdef ROCKSDB_LITE
// not supported in lite version
result = nullptr;
#else
SuperVersion* sv = cfd->GetReferencedSuperVersion(this);
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
5 years ago
auto iter = new ForwardIterator(this, read_options, cfd, sv,
/* allow_unprepared_value */ true);
result = NewDBIterator(
env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
cfd->user_comparator(), iter, sv->current, kMaxSequenceNumber,
sv->mutable_cf_options.max_sequential_skip_in_iterations, read_callback,
this, cfd);
#endif
} else {
// Note: no need to consider the special case of
// last_seq_same_as_publish_seq_==false since NewIterator is overridden in
// WritePreparedTxnDB
result = NewIteratorImpl(read_options, cfd,
(read_options.snapshot != nullptr)
? read_options.snapshot->GetSequenceNumber()
: kMaxSequenceNumber,
read_callback);
}
return result;
}
ArenaWrappedDBIter* DBImpl::NewIteratorImpl(const ReadOptions& read_options,
ColumnFamilyData* cfd,
SequenceNumber snapshot,
ReadCallback* read_callback,
bool expose_blob_index,
bool allow_refresh) {
SuperVersion* sv = cfd->GetReferencedSuperVersion(this);
TEST_SYNC_POINT("DBImpl::NewIterator:1");
TEST_SYNC_POINT("DBImpl::NewIterator:2");
if (snapshot == kMaxSequenceNumber) {
// Note that the snapshot is assigned AFTER referencing the super
// version because otherwise a flush happening in between may compact away
// data for the snapshot, so the reader would see neither data that was be
// visible to the snapshot before compaction nor the newer data inserted
// afterwards.
// Note that the super version might not contain all the data available
// to this snapshot, but in that case it can see all the data in the
// super version, which is a valid consistent state after the user
// calls NewIterator().
snapshot = versions_->LastSequence();
TEST_SYNC_POINT("DBImpl::NewIterator:3");
TEST_SYNC_POINT("DBImpl::NewIterator:4");
}
// Try to generate a DB iterator tree in continuous memory area to be
// cache friendly. Here is an example of result:
// +-------------------------------+
// | |
// | ArenaWrappedDBIter |
// | + |
// | +---> Inner Iterator ------------+
// | | | |
// | | +-- -- -- -- -- -- -- --+ |
// | +--- | Arena | |
// | | | |
// | Allocated Memory: | |
// | | +-------------------+ |
// | | | DBIter | <---+
// | | + |
// | | | +-> iter_ ------------+
// | | | | |
// | | +-------------------+ |
// | | | MergingIterator | <---+
// | | + |
// | | | +->child iter1 ------------+
// | | | | | |
// | | +->child iter2 ----------+ |
// | | | | | | |
// | | | +->child iter3 --------+ | |
// | | | | | |
// | | +-------------------+ | | |
// | | | Iterator1 | <--------+
// | | +-------------------+ | |
// | | | Iterator2 | <------+
// | | +-------------------+ |
// | | | Iterator3 | <----+
// | | +-------------------+
// | | |
// +-------+-----------------------+
//
// ArenaWrappedDBIter inlines an arena area where all the iterators in
// the iterator tree are allocated in the order of being accessed when
// querying.
// Laying out the iterators in the order of being accessed makes it more
// likely that any iterator pointer is close to the iterator it points to so
// that they are likely to be in the same cache line and/or page.
ArenaWrappedDBIter* db_iter = NewArenaWrappedDbIterator(
env_, read_options, *cfd->ioptions(), sv->mutable_cf_options, sv->current,
snapshot, sv->mutable_cf_options.max_sequential_skip_in_iterations,
sv->version_number, read_callback, this, cfd, expose_blob_index,
read_options.snapshot != nullptr ? false : allow_refresh);
InternalIterator* internal_iter = NewInternalIterator(
db_iter->GetReadOptions(), cfd, sv, db_iter->GetArena(),
db_iter->GetRangeDelAggregator(), snapshot,
/* allow_unprepared_value */ true);
db_iter->SetIterUnderDBIter(internal_iter);
return db_iter;
}
Status DBImpl::NewIterators(
const ReadOptions& read_options,
const std::vector<ColumnFamilyHandle*>& column_families,
std::vector<Iterator*>* iterators) {
if (read_options.managed) {
return Status::NotSupported("Managed iterator is not supported anymore.");
}
if (read_options.read_tier == kPersistedTier) {
return Status::NotSupported(
"ReadTier::kPersistedData is not yet supported in iterators.");
}
ReadCallback* read_callback = nullptr; // No read callback provided.
iterators->clear();
iterators->reserve(column_families.size());
if (read_options.tailing) {
#ifdef ROCKSDB_LITE
return Status::InvalidArgument(
"Tailing iterator not supported in RocksDB lite");
#else
for (auto cfh : column_families) {
auto cfd = static_cast_with_check<ColumnFamilyHandleImpl>(cfh)->cfd();
SuperVersion* sv = cfd->GetReferencedSuperVersion(this);
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
5 years ago
auto iter = new ForwardIterator(this, read_options, cfd, sv,
/* allow_unprepared_value */ true);
iterators->push_back(NewDBIterator(
env_, read_options, *cfd->ioptions(), sv->mutable_cf_options,
cfd->user_comparator(), iter, sv->current, kMaxSequenceNumber,
sv->mutable_cf_options.max_sequential_skip_in_iterations,
read_callback, this, cfd));
}
#endif
} else {
// Note: no need to consider the special case of
// last_seq_same_as_publish_seq_==false since NewIterators is overridden in
// WritePreparedTxnDB
auto snapshot = read_options.snapshot != nullptr
? read_options.snapshot->GetSequenceNumber()
: versions_->LastSequence();
for (size_t i = 0; i < column_families.size(); ++i) {
auto* cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_families[i])
->cfd();
iterators->push_back(
NewIteratorImpl(read_options, cfd, snapshot, read_callback));
}
}
return Status::OK();
}
const Snapshot* DBImpl::GetSnapshot() { return GetSnapshotImpl(false); }
#ifndef ROCKSDB_LITE
const Snapshot* DBImpl::GetSnapshotForWriteConflictBoundary() {
return GetSnapshotImpl(true);
}
#endif // ROCKSDB_LITE
SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
bool lock) {
int64_t unix_time = 0;
immutable_db_options_.clock->GetCurrentTime(&unix_time)
.PermitUncheckedError(); // Ignore error
SnapshotImpl* s = new SnapshotImpl;
if (lock) {
mutex_.Lock();
}
// returns null if the underlying memtable does not support snapshot.
if (!is_snapshot_supported_) {
if (lock) {
mutex_.Unlock();
}
delete s;
return nullptr;
}
auto snapshot_seq = last_seq_same_as_publish_seq_
? versions_->LastSequence()
: versions_->LastPublishedSequence();
SnapshotImpl* snapshot =
snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
if (lock) {
mutex_.Unlock();
}
return snapshot;
}
namespace {
using CfdList = autovector<ColumnFamilyData*, 2>;
bool CfdListContains(const CfdList& list, ColumnFamilyData* cfd) {
for (const ColumnFamilyData* t : list) {
if (t == cfd) {
return true;
}
}
return false;
}
} // namespace
void DBImpl::ReleaseSnapshot(const Snapshot* s) {
const SnapshotImpl* casted_s = reinterpret_cast<const SnapshotImpl*>(s);
{
InstrumentedMutexLock l(&mutex_);
snapshots_.Delete(casted_s);
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
7 years ago
uint64_t oldest_snapshot;
if (snapshots_.empty()) {
Fix TSAN build and re-enable the tests (#7386) Summary: Resolve TSAN build warnings and re-enable disabled TSAN tests. Not sure if it's a compiler issue or TSAN check issue. Switching from conditional operator to if-else mitigated the problem. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7386 Test Plan: run TSAN check 10 times in circleci. ``` WARNING: ThreadSanitizer: data race (pid=27735) Atomic write of size 8 at 0x7b54000005e8 by thread T32: #0 __tsan_atomic64_store <null> (db_test+0x4cee95) https://github.com/facebook/rocksdb/issues/1 std::__atomic_base<unsigned long>::store(unsigned long, std::memory_order) /usr/bin/../lib/gcc/x86_64-linux-gnu/5.4.0/../../../../include/c++/5.4.0/bits/atomic_base.h:374:2 (db_test+0x78460e) https://github.com/facebook/rocksdb/issues/2 rocksdb::VersionSet::SetLastSequence(unsigned long) /home/circleci/project/./db/version_set.h:1058:20 (db_test+0x78460e) ... Previous read of size 8 at 0x7b54000005e8 by thread T31: #0 bool rocksdb::DBImpl::MultiCFSnapshot<std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > > >(rocksdb::ReadOptions const&, rocksdb::ReadCallback*, std::function<rocksdb::DBImpl::MultiGetColumnFamilyData* (std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >::iterator&)>&, std::unordered_map<unsigned int, rocksdb::DBImpl::MultiGetColumnFamilyData, std::hash<unsigned int>, std::equal_to<unsigned int>, std::allocator<std::pair<unsigned int const, rocksdb::DBImpl::MultiGetColumnFamilyData> > >*, unsigned long*) /home/circleci/project/db/db_impl/db_impl.cc (db_test+0x715087) ``` Reviewed By: ltamasi Differential Revision: D23725226 Pulled By: jay-zhuang fbshipit-source-id: a6d662a5ea68111246cd32ec95f3411a25f76bc6
4 years ago
if (last_seq_same_as_publish_seq_) {
oldest_snapshot = versions_->LastSequence();
} else {
oldest_snapshot = versions_->LastPublishedSequence();
}
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
7 years ago
} else {
oldest_snapshot = snapshots_.oldest()->number_;
}
// Avoid to go through every column family by checking a global threshold
// first.
if (oldest_snapshot > bottommost_files_mark_threshold_) {
CfdList cf_scheduled;
for (auto* cfd : *versions_->GetColumnFamilySet()) {
cfd->current()->storage_info()->UpdateOldestSnapshot(oldest_snapshot);
if (!cfd->current()
->storage_info()
->BottommostFilesMarkedForCompaction()
.empty()) {
SchedulePendingCompaction(cfd);
MaybeScheduleFlushOrCompaction();
cf_scheduled.push_back(cfd);
}
}
// Calculate a new threshold, skipping those CFs where compactions are
// scheduled. We do not do the same pass as the previous loop because
// mutex might be unlocked during the loop, making the result inaccurate.
SequenceNumber new_bottommost_files_mark_threshold = kMaxSequenceNumber;
for (auto* cfd : *versions_->GetColumnFamilySet()) {
if (CfdListContains(cf_scheduled, cfd)) {
continue;
}
new_bottommost_files_mark_threshold = std::min(
new_bottommost_files_mark_threshold,
cfd->current()->storage_info()->bottommost_files_mark_threshold());
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
7 years ago
}
bottommost_files_mark_threshold_ = new_bottommost_files_mark_threshold;
single-file bottom-level compaction when snapshot released Summary: When snapshots are held for a long time, files may reach the bottom level containing overwritten/deleted keys. We previously had no mechanism to trigger compaction on such files. This particularly impacted DBs that write to different parts of the keyspace over time, as such files would never be naturally compacted due to second-last level files moving down. This PR introduces a mechanism for bottommost files to be recompacted upon releasing all snapshots that prevent them from dropping their deleted/overwritten keys. - Changed `CompactionPicker` to compact files in `BottommostFilesMarkedForCompaction()`. These are the last choice when picking. Each file will be compacted alone and output to the same level in which it originated. The goal of this type of compaction is to rewrite the data excluding deleted/overwritten keys. - Changed `ReleaseSnapshot()` to recompute the bottom files marked for compaction when the oldest existing snapshot changes, and schedule a compaction if needed. We cache the value that oldest existing snapshot needs to exceed in order for another file to be marked in `bottommost_files_mark_threshold_`, which allows us to avoid recomputing marked files for most snapshot releases. - Changed `VersionStorageInfo` to track the list of bottommost files, which is recomputed every time the version changes by `UpdateBottommostFiles()`. The list of marked bottommost files is first computed in `ComputeBottommostFilesMarkedForCompaction()` when the version changes, but may also be recomputed when `ReleaseSnapshot()` is called. - Extracted core logic of `Compaction::IsBottommostLevel()` into `VersionStorageInfo::RangeMightExistAfterSortedRun()` since logic to check whether a file is bottommost is now necessary outside of compaction. Closes https://github.com/facebook/rocksdb/pull/3009 Differential Revision: D6062044 Pulled By: ajkr fbshipit-source-id: 123d201cf140715a7d5928e8b3cb4f9cd9f7ad21
7 years ago
}
}
delete casted_s;
}
#ifndef ROCKSDB_LITE
Status DBImpl::GetPropertiesOfAllTables(ColumnFamilyHandle* column_family,
TablePropertiesCollection* props) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
auto cfd = cfh->cfd();
// Increment the ref count
mutex_.Lock();
auto version = cfd->current();
version->Ref();
mutex_.Unlock();
auto s = version->GetPropertiesOfAllTables(props);
// Decrement the ref count
mutex_.Lock();
version->Unref();
mutex_.Unlock();
return s;
}
Status DBImpl::GetPropertiesOfTablesInRange(ColumnFamilyHandle* column_family,
const Range* range, std::size_t n,
TablePropertiesCollection* props) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
auto cfd = cfh->cfd();
// Increment the ref count
mutex_.Lock();
auto version = cfd->current();
version->Ref();
mutex_.Unlock();
auto s = version->GetPropertiesOfTablesInRange(range, n, props);
// Decrement the ref count
mutex_.Lock();
version->Unref();
mutex_.Unlock();
return s;
}
#endif // ROCKSDB_LITE
const std::string& DBImpl::GetName() const { return dbname_; }
[RocksDB] BackupableDB Summary: In this diff I present you BackupableDB v1. You can easily use it to backup your DB and it will do incremental snapshots for you. Let's first describe how you would use BackupableDB. It's inheriting StackableDB interface so you can easily construct it with your DB object -- it will add a method RollTheSnapshot() to the DB object. When you call RollTheSnapshot(), current snapshot of the DB will be stored in the backup dir. To restore, you can just call RestoreDBFromBackup() on a BackupableDB (which is a static method) and it will restore all files from the backup dir. In the next version, it will even support automatic backuping every X minutes. There are multiple things you can configure: 1. backup_env and db_env can be different, which is awesome because then you can easily backup to HDFS or wherever you feel like. 2. sync - if true, it *guarantees* backup consistency on machine reboot 3. number of snapshots to keep - this will keep last N snapshots around if you want, for some reason, be able to restore from an earlier snapshot. All the backuping is done in incremental fashion - if we already have 00010.sst, we will not copy it again. *IMPORTANT* -- This is based on assumption that 00010.sst never changes - two files named 00010.sst from the same DB will always be exactly the same. Is this true? I always copy manifest, current and log files. 4. You can decide if you want to flush the memtables before you backup, or you're fine with backing up the log files -- either way, you get a complete and consistent view of the database at a time of backup. 5. More things you can find in BackupableDBOptions Here is the directory structure I use: backup_dir/CURRENT_SNAPSHOT - just 4 bytes holding the latest snapshot 0, 1, 2, ... - files containing serialized version of each snapshot - containing a list of files files/*.sst - sst files shared between snapshots - if one snapshot references 00010.sst and another one needs to backup it from the DB, it will just reference the same file files/ 0/, 1/, 2/, ... - snapshot directories containing private snapshot files - current, manifest and log files All the files are ref counted and deleted immediatelly when they get out of scope. Some other stuff in this diff: 1. Added GetEnv() method to the DB. Discussed with @haobo and we agreed that it seems right thing to do. 2. Fixed StackableDB interface. The way it was set up before, I was not able to implement BackupableDB. Test Plan: I have a unittest, but please don't look at this yet. I just hacked it up to help me with debugging. I will write a lot of good tests and update the diff. Also, `make asan_check` Reviewers: dhruba, haobo, emayanke Reviewed By: dhruba CC: leveldb, haobo Differential Revision: https://reviews.facebook.net/D14295
11 years ago
Env* DBImpl::GetEnv() const { return env_; }
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
FileSystem* DB::GetFileSystem() const {
Create a CustomEnv class; Add WinFileSystem; Make LegacyFileSystemWrapper private (#7703) Summary: This PR does the following: -> Creates a WinFileSystem class. This class is the Windows equivalent of the PosixFileSystem and will be used on Windows systems. -> Introduces a CustomEnv class. A CustomEnv is an Env that takes a FileSystem as constructor argument. I believe there will only ever be two implementations of this class (PosixEnv and WinEnv). There is still a CustomEnvWrapper class that takes an Env and a FileSystem and wraps the Env calls with the input Env but uses the FileSystem for the FileSystem calls -> Eliminates the public uses of the LegacyFileSystemWrapper. With this change in place, there are effectively the following patterns of Env: - "Base Env classes" (PosixEnv, WinEnv). These classes implement the core Env functions (e.g. Threads) and have a hard-coded input FileSystem. These classes inherit from CompositeEnv, implement the core Env functions (threads) and delegate the FileSystem-like calls to the input file system. - Wrapped Composite Env classes (MemEnv). These classes take in an Env and a FileSystem. The core env functions are re-directed to the wrapped env. The file system calls are redirected to the input file system - Legacy Wrapped Env classes. These classes take in an Env input (but no FileSystem). The core env functions are re-directed to the wrapped env. A "Legacy File System" is created using this env and the file system calls directed to the env itself. With these changes in place, the PosixEnv becomes a singleton -- there is only ever one created. Any other use of the PosixEnv is via another wrapped env. This cleans up some of the issues with the env construction and destruction. Additionally, there were places in the code that required had an Env when they required a FileSystem. Many of these places would wrap the Env with a LegacyFileSystemWrapper instead of using the env->GetFileSystem(). These places were changed, thereby removing layers of additional redirection (LegacyFileSystem --> Env --> Env::FileSystem). Pull Request resolved: https://github.com/facebook/rocksdb/pull/7703 Reviewed By: zhichao-cao Differential Revision: D25762190 Pulled By: anand1976 fbshipit-source-id: 1a088e97fc916f28ac69c149cd1dcad0ab31704b
4 years ago
const auto& fs = GetEnv()->GetFileSystem();
return fs.get();
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
}
FileSystem* DBImpl::GetFileSystem() const {
return immutable_db_options_.fs.get();
}
SystemClock* DBImpl::GetSystemClock() const {
return immutable_db_options_.clock;
}
#ifndef ROCKSDB_LITE
Status DBImpl::StartIOTrace(const TraceOptions& trace_options,
std::unique_ptr<TraceWriter>&& trace_writer) {
assert(trace_writer != nullptr);
return io_tracer_->StartIOTrace(GetSystemClock(), trace_options,
std::move(trace_writer));
}
Status DBImpl::EndIOTrace() {
io_tracer_->EndIOTrace();
return Status::OK();
}
#endif // ROCKSDB_LITE
Options DBImpl::GetOptions(ColumnFamilyHandle* column_family) const {
InstrumentedMutexLock l(&mutex_);
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
return Options(BuildDBOptions(immutable_db_options_, mutable_db_options_),
cfh->cfd()->GetLatestCFOptions());
}
DBOptions DBImpl::GetDBOptions() const {
InstrumentedMutexLock l(&mutex_);
return BuildDBOptions(immutable_db_options_, mutable_db_options_);
}
bool DBImpl::GetProperty(ColumnFamilyHandle* column_family,
const Slice& property, std::string* value) {
const DBPropertyInfo* property_info = GetPropertyInfo(property);
value->clear();
auto cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
if (property_info == nullptr) {
return false;
} else if (property_info->handle_int) {
uint64_t int_value;
bool ret_value =
GetIntPropertyInternal(cfd, *property_info, false, &int_value);
if (ret_value) {
*value = ToString(int_value);
}
return ret_value;
} else if (property_info->handle_string) {
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
if (property_info->need_out_of_mutex) {
return cfd->internal_stats()->GetStringProperty(*property_info, property,
value);
} else {
InstrumentedMutexLock l(&mutex_);
return cfd->internal_stats()->GetStringProperty(*property_info, property,
value);
}
} else if (property_info->handle_string_dbimpl) {
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
if (property_info->need_out_of_mutex) {
return (this->*(property_info->handle_string_dbimpl))(value);
} else {
InstrumentedMutexLock l(&mutex_);
return (this->*(property_info->handle_string_dbimpl))(value);
}
}
// Shouldn't reach here since exactly one of handle_string and handle_int
// should be non-nullptr.
assert(false);
return false;
}
bool DBImpl::GetMapProperty(ColumnFamilyHandle* column_family,
const Slice& property,
std::map<std::string, std::string>* value) {
const DBPropertyInfo* property_info = GetPropertyInfo(property);
value->clear();
auto cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
if (property_info == nullptr) {
return false;
} else if (property_info->handle_map) {
Don't hold DB mutex for block cache entry stat scans (#8538) Summary: I previously didn't notice the DB mutex was being held during block cache entry stat scans, probably because I primarily checked for read performance regressions, because they require the block cache and are traditionally latency-sensitive. This change does some refactoring to avoid holding DB mutex and to avoid triggering and waiting for a scan in GetProperty("rocksdb.cfstats"). Some tests have to be updated because now the stats collector is populated in the Cache aggressively on DB startup rather than lazily. (I hope to clean up some of this added complexity in the future.) This change also ensures proper treatment of need_out_of_mutex for non-int DB properties. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8538 Test Plan: Added unit test logic that uses sync points to fail if the DB mutex is held during a scan, covering the various ways that a scan might be triggered. Performance test - the known impact to holding the DB mutex is on TransactionDB, and the easiest way to see the impact is to hack the scan code to almost always miss and take an artificially long time scanning. Here I've injected an unconditional 5s sleep at the call to ApplyToAllEntries. Before (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 433.219 micros/op 2308 ops/sec; 0.1 MB/s ( transactions:78999 aborts:0) rocksdb.db.write.micros P50 : 16.135883 P95 : 36.622503 P99 : 66.036115 P100 : 5000614.000000 COUNT : 149677 SUM : 8364856 $ TEST_TMPDIR=/dev/shm ./db_bench.base_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 448.802 micros/op 2228 ops/sec; 0.1 MB/s ( transactions:75999 aborts:0) rocksdb.db.write.micros P50 : 16.629221 P95 : 37.320607 P99 : 72.144341 P100 : 5000871.000000 COUNT : 143995 SUM : 13472323 Notice the 5s P100 write time. After (hacked): $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 303.645 micros/op 3293 ops/sec; 0.1 MB/s ( transactions:98999 aborts:0) rocksdb.db.write.micros P50 : 16.061871 P95 : 33.978834 P99 : 60.018017 P100 : 616315.000000 COUNT : 187619 SUM : 4097407 $ TEST_TMPDIR=/dev/shm ./db_bench.new_xxx -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 310.383 micros/op 3221 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.270026 P95 : 35.786844 P99 : 64.302878 P100 : 603088.000000 COUNT : 183819 SUM : 4095918 P100 write is now ~0.6s. Not good, but it's the same even if I completely bypass all the scanning code: $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 311.365 micros/op 3211 ops/sec; 0.1 MB/s ( transactions:96999 aborts:0) rocksdb.db.write.micros P50 : 16.274362 P95 : 36.221184 P99 : 68.809783 P100 : 649808.000000 COUNT : 183819 SUM : 4156767 $ TEST_TMPDIR=/dev/shm ./db_bench.new_skip -benchmarks=randomtransaction,stats -cache_index_and_filter_blocks=1 -bloom_bits=10 -partition_index_and_filters=1 -duration=30 -stats_dump_period_sec=12 -cache_size=100000000 -statistics -transaction_db 2>&1 | egrep 'db.db.write.micros|micros/op' randomtransaction : 308.395 micros/op 3242 ops/sec; 0.1 MB/s ( transactions:97999 aborts:0) rocksdb.db.write.micros P50 : 16.106222 P95 : 37.202403 P99 : 67.081875 P100 : 598091.000000 COUNT : 185714 SUM : 4098832 No substantial difference. Reviewed By: siying Differential Revision: D29738847 Pulled By: pdillinger fbshipit-source-id: 1c5c155f5a1b62e4fea0fd4eeb515a8b7474027b
3 years ago
if (property_info->need_out_of_mutex) {
return cfd->internal_stats()->GetMapProperty(*property_info, property,
value);
} else {
InstrumentedMutexLock l(&mutex_);
return cfd->internal_stats()->GetMapProperty(*property_info, property,
value);
}
}
// If we reach this point it means that handle_map is not provided for the
// requested property
return false;
}
bool DBImpl::GetIntProperty(ColumnFamilyHandle* column_family,
const Slice& property, uint64_t* value) {
const DBPropertyInfo* property_info = GetPropertyInfo(property);
if (property_info == nullptr || property_info->handle_int == nullptr) {
return false;
}
auto cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
return GetIntPropertyInternal(cfd, *property_info, false, value);
}
bool DBImpl::GetIntPropertyInternal(ColumnFamilyData* cfd,
const DBPropertyInfo& property_info,
bool is_locked, uint64_t* value) {
assert(property_info.handle_int != nullptr);
if (!property_info.need_out_of_mutex) {
if (is_locked) {
mutex_.AssertHeld();
return cfd->internal_stats()->GetIntProperty(property_info, value, this);
} else {
InstrumentedMutexLock l(&mutex_);
return cfd->internal_stats()->GetIntProperty(property_info, value, this);
}
} else {
SuperVersion* sv = nullptr;
if (is_locked) {
mutex_.Unlock();
}
sv = GetAndRefSuperVersion(cfd);
bool ret = cfd->internal_stats()->GetIntPropertyOutOfMutex(
property_info, sv->current, value);
ReturnAndCleanupSuperVersion(cfd, sv);
if (is_locked) {
mutex_.Lock();
}
return ret;
}
}
bool DBImpl::GetPropertyHandleOptionsStatistics(std::string* value) {
assert(value != nullptr);
Statistics* statistics = immutable_db_options_.stats;
if (!statistics) {
return false;
}
*value = statistics->ToString();
return true;
}
#ifndef ROCKSDB_LITE
Status DBImpl::ResetStats() {
InstrumentedMutexLock l(&mutex_);
for (auto* cfd : *versions_->GetColumnFamilySet()) {
if (cfd->initialized()) {
cfd->internal_stats()->Clear();
}
}
return Status::OK();
}
#endif // ROCKSDB_LITE
bool DBImpl::GetAggregatedIntProperty(const Slice& property,
uint64_t* aggregated_value) {
const DBPropertyInfo* property_info = GetPropertyInfo(property);
if (property_info == nullptr || property_info->handle_int == nullptr) {
return false;
}
uint64_t sum = 0;
bool ret = true;
{
// Needs mutex to protect the list of column families.
InstrumentedMutexLock l(&mutex_);
uint64_t value;
for (auto* cfd : *versions_->GetColumnFamilySet()) {
if (!cfd->initialized()) {
continue;
}
cfd->Ref();
ret = GetIntPropertyInternal(cfd, *property_info, true, &value);
// GetIntPropertyInternal may release db mutex and re-acquire it.
mutex_.AssertHeld();
cfd->UnrefAndTryDelete();
if (ret) {
sum += value;
} else {
ret = false;
break;
}
}
}
*aggregated_value = sum;
return ret;
}
SuperVersion* DBImpl::GetAndRefSuperVersion(ColumnFamilyData* cfd) {
// TODO(ljin): consider using GetReferencedSuperVersion() directly
return cfd->GetThreadLocalSuperVersion(this);
}
// REQUIRED: this function should only be called on the write thread or if the
// mutex is held.
SuperVersion* DBImpl::GetAndRefSuperVersion(uint32_t column_family_id) {
auto column_family_set = versions_->GetColumnFamilySet();
auto cfd = column_family_set->GetColumnFamily(column_family_id);
if (!cfd) {
return nullptr;
}
return GetAndRefSuperVersion(cfd);
}
void DBImpl::CleanupSuperVersion(SuperVersion* sv) {
// Release SuperVersion
if (sv->Unref()) {
bool defer_purge =
immutable_db_options().avoid_unnecessary_blocking_io;
{
InstrumentedMutexLock l(&mutex_);
sv->Cleanup();
if (defer_purge) {
AddSuperVersionsToFreeQueue(sv);
SchedulePurge();
}
}
if (!defer_purge) {
delete sv;
}
RecordTick(stats_, NUMBER_SUPERVERSION_CLEANUPS);
}
RecordTick(stats_, NUMBER_SUPERVERSION_RELEASES);
}
void DBImpl::ReturnAndCleanupSuperVersion(ColumnFamilyData* cfd,
SuperVersion* sv) {
if (!cfd->ReturnThreadLocalSuperVersion(sv)) {
CleanupSuperVersion(sv);
}
}
// REQUIRED: this function should only be called on the write thread.
void DBImpl::ReturnAndCleanupSuperVersion(uint32_t column_family_id,
SuperVersion* sv) {
auto column_family_set = versions_->GetColumnFamilySet();
auto cfd = column_family_set->GetColumnFamily(column_family_id);
// If SuperVersion is held, and we successfully fetched a cfd using
// GetAndRefSuperVersion(), it must still exist.
assert(cfd != nullptr);
ReturnAndCleanupSuperVersion(cfd, sv);
}
// REQUIRED: this function should only be called on the write thread or if the
// mutex is held.
ColumnFamilyHandle* DBImpl::GetColumnFamilyHandle(uint32_t column_family_id) {
ColumnFamilyMemTables* cf_memtables = column_family_memtables_.get();
if (!cf_memtables->Seek(column_family_id)) {
return nullptr;
}
return cf_memtables->GetColumnFamilyHandle();
}
// REQUIRED: mutex is NOT held.
std::unique_ptr<ColumnFamilyHandle> DBImpl::GetColumnFamilyHandleUnlocked(
uint32_t column_family_id) {
InstrumentedMutexLock l(&mutex_);
auto* cfd =
versions_->GetColumnFamilySet()->GetColumnFamily(column_family_id);
if (cfd == nullptr) {
return nullptr;
}
return std::unique_ptr<ColumnFamilyHandleImpl>(
new ColumnFamilyHandleImpl(cfd, this, &mutex_));
}
void DBImpl::GetApproximateMemTableStats(ColumnFamilyHandle* column_family,
const Range& range,
uint64_t* const count,
uint64_t* const size) {
ColumnFamilyHandleImpl* cfh =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
ColumnFamilyData* cfd = cfh->cfd();
SuperVersion* sv = GetAndRefSuperVersion(cfd);
// Convert user_key into a corresponding internal key.
InternalKey k1(range.start, kMaxSequenceNumber, kValueTypeForSeek);
InternalKey k2(range.limit, kMaxSequenceNumber, kValueTypeForSeek);
MemTable::MemTableStats memStats =
sv->mem->ApproximateStats(k1.Encode(), k2.Encode());
MemTable::MemTableStats immStats =
sv->imm->ApproximateStats(k1.Encode(), k2.Encode());
*count = memStats.count + immStats.count;
*size = memStats.size + immStats.size;
ReturnAndCleanupSuperVersion(cfd, sv);
}
Status DBImpl::GetApproximateSizes(const SizeApproximationOptions& options,
ColumnFamilyHandle* column_family,
const Range* range, int n, uint64_t* sizes) {
if (!options.include_memtabtles && !options.include_files) {
return Status::InvalidArgument("Invalid options");
}
const Comparator* const ucmp = column_family->GetComparator();
assert(ucmp);
size_t ts_sz = ucmp->timestamp_size();
Version* v;
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
auto cfd = cfh->cfd();
SuperVersion* sv = GetAndRefSuperVersion(cfd);
v = sv->current;
for (int i = 0; i < n; i++) {
Slice start = range[i].start;
Slice limit = range[i].limit;
// Add timestamp if needed
std::string start_with_ts, limit_with_ts;
if (ts_sz > 0) {
// Maximum timestamp means including all key with any timestamp
AppendKeyWithMaxTimestamp(&start_with_ts, start, ts_sz);
// Append a maximum timestamp as the range limit is exclusive:
// [start, limit)
AppendKeyWithMaxTimestamp(&limit_with_ts, limit, ts_sz);
start = start_with_ts;
limit = limit_with_ts;
}
// Convert user_key into a corresponding internal key.
InternalKey k1(start, kMaxSequenceNumber, kValueTypeForSeek);
InternalKey k2(limit, kMaxSequenceNumber, kValueTypeForSeek);
sizes[i] = 0;
if (options.include_files) {
Create a BlockCacheLookupContext to enable fine-grained block cache tracing. (#5421) Summary: BlockCacheLookupContext only contains the caller for now. We will trace block accesses at five places: 1. BlockBasedTable::GetFilter. 2. BlockBasedTable::GetUncompressedDict. 3. BlockBasedTable::MaybeReadAndLoadToCache. (To trace access on data, index, and range deletion block.) 4. BlockBasedTable::Get. (To trace the referenced key and whether the referenced key exists in a fetched data block.) 5. BlockBasedTable::MultiGet. (To trace the referenced key and whether the referenced key exists in a fetched data block.) We create the context at: 1. BlockBasedTable::Get. (kUserGet) 2. BlockBasedTable::MultiGet. (kUserMGet) 3. BlockBasedTable::NewIterator. (either kUserIterator, kCompaction, or external SST ingestion calls this function.) 4. BlockBasedTable::Open. (kPrefetch) 5. Index/Filter::CacheDependencies. (kPrefetch) 6. BlockBasedTable::ApproximateOffsetOf. (kCompaction or kUserApproximateSize). I loaded 1 million key-value pairs into the database and ran the readrandom benchmark with a single thread. I gave the block cache 10 GB to make sure all reads hit the block cache after warmup. The throughput is comparable. Throughput of this PR: 231334 ops/s. Throughput of the master branch: 238428 ops/s. Experiment setup: RocksDB: version 6.2 Date: Mon Jun 10 10:42:51 2019 CPU: 24 * Intel Core Processor (Skylake) CPUCache: 16384 KB Keys: 20 bytes each Values: 100 bytes each (100 bytes after compression) Entries: 1000000 Prefix: 20 bytes Keys per prefix: 0 RawSize: 114.4 MB (estimated) FileSize: 114.4 MB (estimated) Write rate: 0 bytes/second Read rate: 0 ops/second Compression: NoCompression Compression sampling rate: 0 Memtablerep: skip_list Perf Level: 1 Load command: ./db_bench --benchmarks="fillseq" --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --statistics --cache_index_and_filter_blocks --cache_size=10737418240 --disable_auto_compactions=1 --disable_wal=1 --compression_type=none --min_level_to_compress=-1 --compression_ratio=1 --num=1000000 Run command: ./db_bench --benchmarks="readrandom,stats" --use_existing_db --threads=1 --duration=120 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --statistics --cache_index_and_filter_blocks --cache_size=10737418240 --disable_auto_compactions=1 --disable_wal=1 --compression_type=none --min_level_to_compress=-1 --compression_ratio=1 --num=1000000 --duration=120 TODOs: 1. Create a caller for external SST file ingestion and differentiate the callers for iterator. 2. Integrate tracer to trace block cache accesses. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5421 Differential Revision: D15704258 Pulled By: HaoyuHuang fbshipit-source-id: 4aa8a55f8cb1576ffb367bfa3186a91d8f06d93a
6 years ago
sizes[i] += versions_->ApproximateSize(
options, v, k1.Encode(), k2.Encode(), /*start_level=*/0,
/*end_level=*/-1, TableReaderCaller::kUserApproximateSize);
}
if (options.include_memtabtles) {
sizes[i] += sv->mem->ApproximateStats(k1.Encode(), k2.Encode()).size;
sizes[i] += sv->imm->ApproximateStats(k1.Encode(), k2.Encode()).size;
}
}
ReturnAndCleanupSuperVersion(cfd, sv);
return Status::OK();
}
std::list<uint64_t>::iterator
DBImpl::CaptureCurrentFileNumberInPendingOutputs() {
// We need to remember the iterator of our insert, because after the
// background job is done, we need to remove that element from
// pending_outputs_.
pending_outputs_.push_back(versions_->current_next_file_number());
auto pending_outputs_inserted_elem = pending_outputs_.end();
--pending_outputs_inserted_elem;
return pending_outputs_inserted_elem;
}
void DBImpl::ReleaseFileNumberFromPendingOutputs(
std::unique_ptr<std::list<uint64_t>::iterator>& v) {
if (v.get() != nullptr) {
pending_outputs_.erase(*v.get());
v.reset();
}
}
#ifndef ROCKSDB_LITE
Status DBImpl::GetUpdatesSince(
SequenceNumber seq, std::unique_ptr<TransactionLogIterator>* iter,
const TransactionLogIterator::ReadOptions& read_options) {
RecordTick(stats_, GET_UPDATES_SINCE_CALLS);
if (seq > versions_->LastSequence()) {
return Status::NotFound("Requested sequence not yet written in the db");
}
return wal_manager_.GetUpdatesSince(seq, iter, read_options, versions_.get());
}
Status DBImpl::DeleteFile(std::string name) {
uint64_t number;
FileType type;
WalFileType log_type;
if (!ParseFileName(name, &number, &type, &log_type) ||
(type != kTableFile && type != kWalFile)) {
ROCKS_LOG_ERROR(immutable_db_options_.info_log, "DeleteFile %s failed.\n",
name.c_str());
return Status::InvalidArgument("Invalid file name");
}
if (type == kWalFile) {
// Only allow deleting archived log files
if (log_type != kArchivedLogFile) {
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"DeleteFile %s failed - not archived log.\n",
name.c_str());
return Status::NotSupported("Delete only supported for archived logs");
}
Status status = wal_manager_.DeleteFile(name, number);
if (!status.ok()) {
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"DeleteFile %s failed -- %s.\n", name.c_str(),
status.ToString().c_str());
}
return status;
}
Status status;
int level;
FileMetaData* metadata;
ColumnFamilyData* cfd;
VersionEdit edit;
JobContext job_context(next_job_id_.fetch_add(1), true);
{
InstrumentedMutexLock l(&mutex_);
status = versions_->GetMetadataForFile(number, &level, &metadata, &cfd);
if (!status.ok()) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"DeleteFile %s failed. File not found\n", name.c_str());
job_context.Clean();
return Status::InvalidArgument("File not found");
}
assert(level < cfd->NumberLevels());
// If the file is being compacted no need to delete.
if (metadata->being_compacted) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"DeleteFile %s Skipped. File about to be compacted\n",
name.c_str());
job_context.Clean();
return Status::OK();
}
// Only the files in the last level can be deleted externally.
// This is to make sure that any deletion tombstones are not
// lost. Check that the level passed is the last level.
auto* vstoreage = cfd->current()->storage_info();
for (int i = level + 1; i < cfd->NumberLevels(); i++) {
if (vstoreage->NumLevelFiles(i) != 0) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"DeleteFile %s FAILED. File not in last level\n",
name.c_str());
job_context.Clean();
return Status::InvalidArgument("File not in last level");
}
}
// if level == 0, it has to be the oldest file
if (level == 0 &&
vstoreage->LevelFiles(0).back()->fd.GetNumber() != number) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"DeleteFile %s failed ---"
" target file in level 0 must be the oldest.",
name.c_str());
job_context.Clean();
return Status::InvalidArgument("File in level 0, but not oldest");
}
edit.SetColumnFamily(cfd->GetID());
edit.DeleteFile(level, number);
status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
&edit, &mutex_, directories_.GetDbDir());
if (status.ok()) {
InstallSuperVersionAndScheduleWork(cfd,
&job_context.superversion_contexts[0],
*cfd->GetLatestMutableCFOptions());
}
FindObsoleteFiles(&job_context, false);
} // lock released here
LogFlush(immutable_db_options_.info_log);
// remove files outside the db-lock
if (job_context.HaveSomethingToDelete()) {
// Call PurgeObsoleteFiles() without holding mutex.
PurgeObsoleteFiles(job_context);
}
job_context.Clean();
return status;
}
Status DBImpl::DeleteFilesInRanges(ColumnFamilyHandle* column_family,
const RangePtr* ranges, size_t n,
bool include_end) {
Status status = Status::OK();
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
ColumnFamilyData* cfd = cfh->cfd();
VersionEdit edit;
std::set<FileMetaData*> deleted_files;
JobContext job_context(next_job_id_.fetch_add(1), true);
{
InstrumentedMutexLock l(&mutex_);
Version* input_version = cfd->current();
auto* vstorage = input_version->storage_info();
for (size_t r = 0; r < n; r++) {
auto begin = ranges[r].start, end = ranges[r].limit;
for (int i = 1; i < cfd->NumberLevels(); i++) {
if (vstorage->LevelFiles(i).empty() ||
!vstorage->OverlapInLevel(i, begin, end)) {
continue;
}
std::vector<FileMetaData*> level_files;
InternalKey begin_storage, end_storage, *begin_key, *end_key;
if (begin == nullptr) {
begin_key = nullptr;
} else {
begin_storage.SetMinPossibleForUserKey(*begin);
begin_key = &begin_storage;
}
if (end == nullptr) {
end_key = nullptr;
} else {
end_storage.SetMaxPossibleForUserKey(*end);
end_key = &end_storage;
}
vstorage->GetCleanInputsWithinInterval(
i, begin_key, end_key, &level_files, -1 /* hint_index */,
nullptr /* file_index */);
FileMetaData* level_file;
for (uint32_t j = 0; j < level_files.size(); j++) {
level_file = level_files[j];
if (level_file->being_compacted) {
continue;
}
if (deleted_files.find(level_file) != deleted_files.end()) {
continue;
}
if (!include_end && end != nullptr &&
cfd->user_comparator()->Compare(level_file->largest.user_key(),
*end) == 0) {
continue;
}
edit.SetColumnFamily(cfd->GetID());
edit.DeleteFile(i, level_file->fd.GetNumber());
deleted_files.insert(level_file);
level_file->being_compacted = true;
}
vstorage->ComputeCompactionScore(*cfd->ioptions(),
*cfd->GetLatestMutableCFOptions());
}
}
if (edit.GetDeletedFiles().empty()) {
job_context.Clean();
return status;
}
input_version->Ref();
status = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
&edit, &mutex_, directories_.GetDbDir());
if (status.ok()) {
InstallSuperVersionAndScheduleWork(cfd,
&job_context.superversion_contexts[0],
*cfd->GetLatestMutableCFOptions());
}
for (auto* deleted_file : deleted_files) {
deleted_file->being_compacted = false;
}
input_version->Unref();
FindObsoleteFiles(&job_context, false);
} // lock released here
LogFlush(immutable_db_options_.info_log);
// remove files outside the db-lock
if (job_context.HaveSomethingToDelete()) {
// Call PurgeObsoleteFiles() without holding mutex.
PurgeObsoleteFiles(job_context);
}
job_context.Clean();
return status;
}
void DBImpl::GetLiveFilesMetaData(std::vector<LiveFileMetaData>* metadata) {
InstrumentedMutexLock l(&mutex_);
versions_->GetLiveFilesMetaData(metadata);
}
Status DBImpl::GetLiveFilesChecksumInfo(FileChecksumList* checksum_list) {
InstrumentedMutexLock l(&mutex_);
return versions_->GetLiveFilesChecksumInfo(checksum_list);
}
void DBImpl::GetColumnFamilyMetaData(ColumnFamilyHandle* column_family,
ColumnFamilyMetaData* cf_meta) {
assert(column_family);
auto* cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
auto* sv = GetAndRefSuperVersion(cfd);
{
// Without mutex, Version::GetColumnFamilyMetaData will have data race with
// Compaction::MarkFilesBeingCompacted. One solution is to use mutex, but
// this may cause regression. An alternative is to make
// FileMetaData::being_compacted atomic, but it will make FileMetaData
// non-copy-able. Another option is to separate these variables from
// original FileMetaData struct, and this requires re-organization of data
// structures. For now, we take the easy approach. If
// DB::GetColumnFamilyMetaData is not called frequently, the regression
// should not be big. We still need to keep an eye on it.
InstrumentedMutexLock l(&mutex_);
sv->current->GetColumnFamilyMetaData(cf_meta);
}
ReturnAndCleanupSuperVersion(cfd, sv);
}
void DBImpl::GetAllColumnFamilyMetaData(
std::vector<ColumnFamilyMetaData>* metadata) {
InstrumentedMutexLock l(&mutex_);
for (auto cfd : *(versions_->GetColumnFamilySet())) {
{
metadata->emplace_back();
cfd->current()->GetColumnFamilyMetaData(&metadata->back());
}
}
}
#endif // ROCKSDB_LITE
Status DBImpl::CheckConsistency() {
mutex_.AssertHeld();
std::vector<LiveFileMetaData> metadata;
versions_->GetLiveFilesMetaData(&metadata);
TEST_SYNC_POINT("DBImpl::CheckConsistency:AfterGetLiveFilesMetaData");
std::string corruption_messages;
if (immutable_db_options_.skip_checking_sst_file_sizes_on_db_open) {
// Instead of calling GetFileSize() for each expected file, call
// GetChildren() for the DB directory and check that all expected files
// are listed, without checking their sizes.
// Since sst files might be in different directories, do it for each
// directory separately.
std::map<std::string, std::vector<std::string>> files_by_directory;
for (const auto& md : metadata) {
// md.name has a leading "/". Remove it.
std::string fname = md.name;
if (!fname.empty() && fname[0] == '/') {
fname = fname.substr(1);
}
files_by_directory[md.db_path].push_back(fname);
}
for (const auto& dir_files : files_by_directory) {
std::string directory = dir_files.first;
std::vector<std::string> existing_files;
Status s = env_->GetChildren(directory, &existing_files);
if (!s.ok()) {
corruption_messages +=
"Can't list files in " + directory + ": " + s.ToString() + "\n";
continue;
}
std::sort(existing_files.begin(), existing_files.end());
for (const std::string& fname : dir_files.second) {
if (!std::binary_search(existing_files.begin(), existing_files.end(),
fname) &&
!std::binary_search(existing_files.begin(), existing_files.end(),
Rocks2LevelTableFileName(fname))) {
corruption_messages +=
"Missing sst file " + fname + " in " + directory + "\n";
}
}
}
} else {
for (const auto& md : metadata) {
// md.name has a leading "/".
std::string file_path = md.db_path + md.name;
uint64_t fsize = 0;
TEST_SYNC_POINT("DBImpl::CheckConsistency:BeforeGetFileSize");
Status s = env_->GetFileSize(file_path, &fsize);
if (!s.ok() &&
env_->GetFileSize(Rocks2LevelTableFileName(file_path), &fsize).ok()) {
s = Status::OK();
}
if (!s.ok()) {
corruption_messages +=
"Can't access " + md.name + ": " + s.ToString() + "\n";
} else if (fsize != md.size) {
corruption_messages += "Sst file size mismatch: " + file_path +
". Size recorded in manifest " +
ToString(md.size) + ", actual size " +
ToString(fsize) + "\n";
}
}
}
if (corruption_messages.size() == 0) {
return Status::OK();
} else {
return Status::Corruption(corruption_messages);
}
}
Status DBImpl::GetDbIdentity(std::string& identity) const {
identity.assign(db_id_);
return Status::OK();
}
Status DBImpl::GetDbIdentityFromIdentityFile(std::string* identity) const {
std::string idfilename = IdentityFileName(dbname_);
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
const FileOptions soptions;
Status s = ReadFileToString(fs_.get(), idfilename, identity);
if (!s.ok()) {
return s;
}
Built-in support for generating unique IDs, bug fix (#8708) Summary: Env::GenerateUniqueId() works fine on Windows and on POSIX where /proc/sys/kernel/random/uuid exists. Our other implementation is flawed and easily produces collision in a new multi-threaded test. As we rely more heavily on DB session ID uniqueness, this becomes a serious issue. This change combines several individually suitable entropy sources for reliable generation of random unique IDs, with goal of uniqueness and portability, not cryptographic strength nor maximum speed. Specifically: * Moves code for getting UUIDs from the OS to port::GenerateRfcUuid rather than in Env implementation details. Callers are now told whether the operation fails or succeeds. * Adds an internal API GenerateRawUniqueId for generating high-quality 128-bit unique identifiers, by combining entropy from three "tracks": * Lots of info from default Env like time, process id, and hostname. * std::random_device * port::GenerateRfcUuid (when working) * Built-in implementations of Env::GenerateUniqueId() will now always produce an RFC 4122 UUID string, either from platform-specific API or by converting the output of GenerateRawUniqueId. DB session IDs now use GenerateRawUniqueId while DB IDs (not as critical) try to use port::GenerateRfcUuid but fall back on GenerateRawUniqueId with conversion to an RFC 4122 UUID. GenerateRawUniqueId is declared and defined under env/ rather than util/ or even port/ because of the Env dependency. Likely follow-up: enhance GenerateRawUniqueId to be faster after the first call and to guarantee uniqueness within the lifetime of a single process (imparting the same property onto DB session IDs). Pull Request resolved: https://github.com/facebook/rocksdb/pull/8708 Test Plan: A new mini-stress test in env_test checks the various public and internal APIs for uniqueness, including each track of GenerateRawUniqueId individually. We can't hope to verify anywhere close to 128 bits of entropy, but it can at least detect flaws as bad as the old code. Serial execution of the new tests takes about 350 ms on my machine. Reviewed By: zhichao-cao, mrambacher Differential Revision: D30563780 Pulled By: pdillinger fbshipit-source-id: de4c9ff4b2f581cf784fcedb5f39f16e5185c364
3 years ago
// If last character is '\n' remove it from identity. (Old implementations
// of Env::GenerateUniqueId() would include a trailing '\n'.)
if (identity->size() > 0 && identity->back() == '\n') {
identity->pop_back();
}
return s;
}
Status DBImpl::GetDbSessionId(std::string& session_id) const {
session_id.assign(db_session_id_);
return Status::OK();
}
Built-in support for generating unique IDs, bug fix (#8708) Summary: Env::GenerateUniqueId() works fine on Windows and on POSIX where /proc/sys/kernel/random/uuid exists. Our other implementation is flawed and easily produces collision in a new multi-threaded test. As we rely more heavily on DB session ID uniqueness, this becomes a serious issue. This change combines several individually suitable entropy sources for reliable generation of random unique IDs, with goal of uniqueness and portability, not cryptographic strength nor maximum speed. Specifically: * Moves code for getting UUIDs from the OS to port::GenerateRfcUuid rather than in Env implementation details. Callers are now told whether the operation fails or succeeds. * Adds an internal API GenerateRawUniqueId for generating high-quality 128-bit unique identifiers, by combining entropy from three "tracks": * Lots of info from default Env like time, process id, and hostname. * std::random_device * port::GenerateRfcUuid (when working) * Built-in implementations of Env::GenerateUniqueId() will now always produce an RFC 4122 UUID string, either from platform-specific API or by converting the output of GenerateRawUniqueId. DB session IDs now use GenerateRawUniqueId while DB IDs (not as critical) try to use port::GenerateRfcUuid but fall back on GenerateRawUniqueId with conversion to an RFC 4122 UUID. GenerateRawUniqueId is declared and defined under env/ rather than util/ or even port/ because of the Env dependency. Likely follow-up: enhance GenerateRawUniqueId to be faster after the first call and to guarantee uniqueness within the lifetime of a single process (imparting the same property onto DB session IDs). Pull Request resolved: https://github.com/facebook/rocksdb/pull/8708 Test Plan: A new mini-stress test in env_test checks the various public and internal APIs for uniqueness, including each track of GenerateRawUniqueId individually. We can't hope to verify anywhere close to 128 bits of entropy, but it can at least detect flaws as bad as the old code. Serial execution of the new tests takes about 350 ms on my machine. Reviewed By: zhichao-cao, mrambacher Differential Revision: D30563780 Pulled By: pdillinger fbshipit-source-id: de4c9ff4b2f581cf784fcedb5f39f16e5185c364
3 years ago
std::string DBImpl::GenerateDbSessionId(Env*) {
Experimental support for SST unique IDs (#8990) Summary: * New public header unique_id.h and function GetUniqueIdFromTableProperties which computes a universally unique identifier based on table properties of table files from recent RocksDB versions. * Generation of DB session IDs is refactored so that they are guaranteed unique in the lifetime of a process running RocksDB. (SemiStructuredUniqueIdGen, new test included.) Along with file numbers, this enables SST unique IDs to be guaranteed unique among SSTs generated in a single process, and "better than random" between processes. See https://github.com/pdillinger/unique_id * In addition to public API producing 'external' unique IDs, there is a function for producing 'internal' unique IDs, with functions for converting between the two. In short, the external ID is "safe" for things people might do with it, and the internal ID enables more "power user" features for the future. Specifically, the external ID goes through a hashing layer so that any subset of bits in the external ID can be used as a hash of the full ID, while also preserving uniqueness guarantees in the first 128 bits (bijective both on first 128 bits and on full 192 bits). Intended follow-up: * Use the internal unique IDs in cache keys. (Avoid conflicts with https://github.com/facebook/rocksdb/issues/8912) (The file offset can be XORed into the third 64-bit value of the unique ID.) * Publish the external unique IDs in FileStorageInfo (https://github.com/facebook/rocksdb/issues/8968) Pull Request resolved: https://github.com/facebook/rocksdb/pull/8990 Test Plan: Unit tests added, and checking of unique ids in stress test. NOTE in stress test we do not generate nearly enough files to thoroughly stress uniqueness, but the test trims off pieces of the ID to check for uniqueness so that we can infer (with some assumptions) stronger properties in the aggregate. Reviewed By: zhichao-cao, mrambacher Differential Revision: D31582865 Pulled By: pdillinger fbshipit-source-id: 1f620c4c86af9abe2a8d177b9ccf2ad2b9f48243
3 years ago
// See SemiStructuredUniqueIdGen for its desirable properties.
static SemiStructuredUniqueIdGen gen;
uint64_t lo, hi;
gen.GenerateNext(&hi, &lo);
if (lo == 0) {
// Avoid emitting session ID with lo==0, so that SST unique
// IDs can be more easily ensured non-zero
gen.GenerateNext(&hi, &lo);
assert(lo != 0);
}
return EncodeSessionId(hi, lo);
}
void DBImpl::SetDbSessionId() {
db_session_id_ = GenerateDbSessionId(env_);
Restore file size in backup table file names (and other cleanup) (#7400) Summary: Prior to 6.12, backup files using share_files_with_checksum had the file size encoded in the file name, after the last '\_' and before the last '.'. We considered this an implementation detail subject to change, and indeed removed this information from the file name (with an option to use old behavior) because it was considered ineffective/inefficient for file name uniqueness. However, some downstream RocksDB users were relying on this information since the file size is not explicitly in the backup manifest file. This primary purpose of this change is "retrofitting" the 6.12 release (not yet a public release) to simultaneously support the benefits of the new naming scheme (I/O performance and data correctness at scale) and preserve the file size information, both as default behaviors. With this change, we are essentially making the file size information encoded in the file name an official, though obscure, extension of the backup meta file format. We preserve an option (kLegacyCrc32cAndFileSize) to use the original "legacy" naming scheme, with its caveats, and make it easy to omit the file size information (no kFlagIncludeFileSize), for more compact file names. But note that changing the naming scheme used on an existing db and backup directory can lead to transient space amplification, as some files will be stored under two names in the shared_checksum directory. Because some backups were saved using the original 6.12 naming scheme, we offer two ways of dealing with those files: SST files generated by older 6.12 versions can either use the default naming scheme in effect when the SST files were generated (kFlagMatchInterimNaming, default, no transient space amplification) or can use a new naming scheme (no kFlagMatchInterimNaming, potential space amplification because some already stored files getting a new name). We don't have a natural way to detect which files were generated by previous 6.12 versions, but this change hacks one in by changing DB session ids to now use a more concise encoding, reducing file name length, saving ~dozen bytes from SST files, and making them visually distinct from DB ids so that they are less likely to be mixed up. Two final auxiliary notes: Recognizing that the backup file names have become a de facto part of the backup meta schema, this change makes them easier to parse and extend by putting a distinct marker, 's', before DB session ids embedded in the name. When we extend this to allow custom checksums in the name, they can get their own marker to ensure safe parsing. For backward compatibility, file size does not get a marker but is assumed for `_[0-9]+[.]` Another change from initial 6.12 default behavior is never including file custom checksum in the file name. Looking ahead to 6.13, we do not want the default behavior to cause backup space amplification for someone turning on file custom checksum checking in BackupEngine; we want that to be an easy decision. When implemented, including file custom checksums in backup file names will be a non-default option. Actual file name patterns and priorities, as regexes: kLegacyCrc32cAndFileSize OR pre-6.12 SST file -> [0-9]+_[0-9]+_[0-9]+[.]sst kFlagMatchInterimNaming set (default) AND early 6.12 SST file -> [0-9]+_[0-9a-fA-F-]+[.]sst kUseDbSessionId AND NOT kFlagIncludeFileSize -> [0-9]+_s[0-9A-Z]{20}[.]sst kUseDbSessionId AND kFlagIncludeFileSize (default) -> [0-9]+_s[0-9A-Z]{20}_[0-9]+[.]sst We might add opt-in options for more '\_' separated data in the name, but embedded file size, if present, will always be after last '\_' and before '.sst'. This change was originally applied to version 6.12. (See https://github.com/facebook/rocksdb/issues/7390) Pull Request resolved: https://github.com/facebook/rocksdb/pull/7400 Test Plan: unit tests included. Sync point callbacks are used to mimic previous version SST files. Reviewed By: ajkr Differential Revision: D23759587 Pulled By: pdillinger fbshipit-source-id: f62d8af4e0978de0a34f26288cfbe66049b70025
4 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::SetDbSessionId", &db_session_id_);
}
// Default implementation -- returns not supported status
Status DB::CreateColumnFamily(const ColumnFamilyOptions& /*cf_options*/,
const std::string& /*column_family_name*/,
ColumnFamilyHandle** /*handle*/) {
return Status::NotSupported("");
}
Status DB::CreateColumnFamilies(
const ColumnFamilyOptions& /*cf_options*/,
const std::vector<std::string>& /*column_family_names*/,
std::vector<ColumnFamilyHandle*>* /*handles*/) {
return Status::NotSupported("");
}
Status DB::CreateColumnFamilies(
const std::vector<ColumnFamilyDescriptor>& /*column_families*/,
std::vector<ColumnFamilyHandle*>* /*handles*/) {
return Status::NotSupported("");
}
Status DB::DropColumnFamily(ColumnFamilyHandle* /*column_family*/) {
return Status::NotSupported("");
}
Status DB::DropColumnFamilies(
const std::vector<ColumnFamilyHandle*>& /*column_families*/) {
return Status::NotSupported("");
}
Status DB::DestroyColumnFamilyHandle(ColumnFamilyHandle* column_family) {
delete column_family;
return Status::OK();
}
DB::~DB() {}
Status DBImpl::Close() {
InstrumentedMutexLock closing_lock_guard(&closing_mutex_);
if (closed_) {
return closing_status_;
}
{
InstrumentedMutexLock l(&mutex_);
// If there is unreleased snapshot, fail the close call
if (!snapshots_.empty()) {
return Status::Aborted("Cannot close DB with unreleased snapshot.");
}
}
closing_status_ = CloseImpl();
closed_ = true;
return closing_status_;
}
Status DB::ListColumnFamilies(const DBOptions& db_options,
const std::string& name,
std::vector<std::string>* column_families) {
Simplify migration to FileSystem API (#6552) Summary: The current Env/FileSystem API separation has a couple of issues - 1. It requires the user to specify 2 options - ```Options::env``` and ```Options::file_system``` - which means they have to make code changes to benefit from the new APIs. Furthermore, there is a risk of accessing the same APIs in two different ways, through Env in the old way and through FileSystem in the new way. The two may not always match, for example, if env is ```PosixEnv``` and FileSystem is a custom implementation. Any stray RocksDB calls to env will use the ```PosixEnv``` implementation rather than the file_system implementation. 2. There needs to be a simple way for the FileSystem developer to instantiate an Env for backward compatibility purposes. This PR solves the above issues and simplifies the migration in the following ways - 1. Embed a shared_ptr to the ```FileSystem``` in the ```Env```, and remove ```Options::file_system``` as a configurable option. This way, no code changes will be required in application code to benefit from the new API. The default Env constructor uses a ```LegacyFileSystemWrapper``` as the embedded ```FileSystem```. 1a. - This also makes it more robust by ensuring that even if RocksDB has some stray calls to Env APIs rather than FileSystem, they will go through the same object and thus there is no risk of getting out of sync. 2. Provide a ```NewCompositeEnv()``` API that can be used to construct a PosixEnv with a custom FileSystem implementation. This eliminates an indirection to call Env APIs, and relieves the FileSystem developer of the burden of having to implement wrappers for the Env APIs. 3. Add a couple of missing FileSystem APIs - ```SanitizeEnvOptions()``` and ```NewLogger()``` Tests: 1. New unit tests 2. make check and make asan_check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6552 Reviewed By: riversand963 Differential Revision: D20592038 Pulled By: anand1976 fbshipit-source-id: c3801ad4153f96d21d5a3ae26c92ba454d1bf1f7
5 years ago
const std::shared_ptr<FileSystem>& fs = db_options.env->GetFileSystem();
return VersionSet::ListColumnFamilies(column_families, name, fs.get());
}
Snapshot::~Snapshot() {}
Status DestroyDB(const std::string& dbname, const Options& options,
const std::vector<ColumnFamilyDescriptor>& column_families) {
ImmutableDBOptions soptions(SanitizeOptions(dbname, options));
Env* env = soptions.env;
std::vector<std::string> filenames;
bool wal_in_db_path = soptions.IsWalDirSameAsDBPath();
// Reset the logger because it holds a handle to the
// log file and prevents cleanup and directory removal
soptions.info_log.reset();
// Ignore error in case directory does not exist
env->GetChildren(dbname, &filenames).PermitUncheckedError();
FileLock* lock;
const std::string lockname = LockFileName(dbname);
Status result = env->LockFile(lockname, &lock);
if (result.ok()) {
uint64_t number;
FileType type;
InfoLogPrefix info_log_prefix(!soptions.db_log_dir.empty(), dbname);
for (const auto& fname : filenames) {
if (ParseFileName(fname, &number, info_log_prefix.prefix, &type) &&
type != kDBLockFile) { // Lock file will be deleted at end
Status del;
std::string path_to_delete = dbname + "/" + fname;
if (type == kMetaDatabase) {
del = DestroyDB(path_to_delete, options);
} else if (type == kTableFile || type == kWalFile ||
type == kBlobFile) {
del = DeleteDBFile(
&soptions, path_to_delete, dbname,
/*force_bg=*/false,
/*force_fg=*/(type == kWalFile) ? !wal_in_db_path : false);
} else {
del = env->DeleteFile(path_to_delete);
}
if (!del.ok() && result.ok()) {
result = del;
}
}
}
std::set<std::string> paths;
for (const DbPath& db_path : options.db_paths) {
paths.insert(db_path.path);
}
for (const ColumnFamilyDescriptor& cf : column_families) {
for (const DbPath& cf_path : cf.options.cf_paths) {
paths.insert(cf_path.path);
}
}
for (const auto& path : paths) {
if (env->GetChildren(path, &filenames).ok()) {
for (const auto& fname : filenames) {
if (ParseFileName(fname, &number, &type) &&
(type == kTableFile ||
type == kBlobFile)) { // Lock file will be deleted at end
std::string file_path = path + "/" + fname;
Status del = DeleteDBFile(&soptions, file_path, dbname,
/*force_bg=*/false, /*force_fg=*/false);
if (!del.ok() && result.ok()) {
result = del;
}
}
}
// TODO: Should we return an error if we cannot delete the directory?
env->DeleteDir(path).PermitUncheckedError();
}
}
std::vector<std::string> walDirFiles;
std::string archivedir = ArchivalDirectory(dbname);
bool wal_dir_exists = false;
if (!soptions.IsWalDirSameAsDBPath(dbname)) {
wal_dir_exists = env->GetChildren(soptions.wal_dir, &walDirFiles).ok();
archivedir = ArchivalDirectory(soptions.wal_dir);
}
// Archive dir may be inside wal dir or dbname and should be
// processed and removed before those otherwise we have issues
// removing them
std::vector<std::string> archiveFiles;
if (env->GetChildren(archivedir, &archiveFiles).ok()) {
// Delete archival files.
for (const auto& file : archiveFiles) {
if (ParseFileName(file, &number, &type) && type == kWalFile) {
Status del =
DeleteDBFile(&soptions, archivedir + "/" + file, archivedir,
/*force_bg=*/false, /*force_fg=*/!wal_in_db_path);
if (!del.ok() && result.ok()) {
result = del;
}
}
}
// Ignore error in case dir contains other files
env->DeleteDir(archivedir).PermitUncheckedError();
}
// Delete log files in the WAL dir
if (wal_dir_exists) {
for (const auto& file : walDirFiles) {
if (ParseFileName(file, &number, &type) && type == kWalFile) {
Status del =
DeleteDBFile(&soptions, LogFileName(soptions.wal_dir, number),
soptions.wal_dir, /*force_bg=*/false,
/*force_fg=*/!wal_in_db_path);
if (!del.ok() && result.ok()) {
result = del;
}
}
}
// Ignore error in case dir contains other files
env->DeleteDir(soptions.wal_dir).PermitUncheckedError();
}
// Ignore error since state is already gone
env->UnlockFile(lock).PermitUncheckedError();
env->DeleteFile(lockname).PermitUncheckedError();
// sst_file_manager holds a ref to the logger. Make sure the logger is
// gone before trying to remove the directory.
soptions.sst_file_manager.reset();
// Ignore error in case dir contains other files
env->DeleteDir(dbname).PermitUncheckedError();
;
}
return result;
}
Status DBImpl::WriteOptionsFile(bool need_mutex_lock,
bool need_enter_write_thread) {
#ifndef ROCKSDB_LITE
WriteThread::Writer w;
if (need_mutex_lock) {
mutex_.Lock();
} else {
mutex_.AssertHeld();
}
if (need_enter_write_thread) {
write_thread_.EnterUnbatched(&w, &mutex_);
}
std::vector<std::string> cf_names;
std::vector<ColumnFamilyOptions> cf_opts;
// This part requires mutex to protect the column family options
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (cfd->IsDropped()) {
continue;
}
cf_names.push_back(cfd->GetName());
cf_opts.push_back(cfd->GetLatestCFOptions());
}
// Unlock during expensive operations. New writes cannot get here
// because the single write thread ensures all new writes get queued.
DBOptions db_options =
BuildDBOptions(immutable_db_options_, mutable_db_options_);
mutex_.Unlock();
TEST_SYNC_POINT("DBImpl::WriteOptionsFile:1");
TEST_SYNC_POINT("DBImpl::WriteOptionsFile:2");
TEST_SYNC_POINT_CALLBACK("DBImpl::WriteOptionsFile:PersistOptions",
&db_options);
std::string file_name =
TempOptionsFileName(GetName(), versions_->NewFileNumber());
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
Status s = PersistRocksDBOptions(db_options, cf_names, cf_opts, file_name,
fs_.get());
if (s.ok()) {
s = RenameTempFileToOptionsFile(file_name);
}
// restore lock
if (!need_mutex_lock) {
mutex_.Lock();
}
if (need_enter_write_thread) {
write_thread_.ExitUnbatched(&w);
}
if (!s.ok()) {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"Unnable to persist options -- %s", s.ToString().c_str());
if (immutable_db_options_.fail_if_options_file_error) {
return Status::IOError("Unable to persist options.",
s.ToString().c_str());
}
}
#else
(void)need_mutex_lock;
(void)need_enter_write_thread;
#endif // !ROCKSDB_LITE
return Status::OK();
}
#ifndef ROCKSDB_LITE
namespace {
void DeleteOptionsFilesHelper(const std::map<uint64_t, std::string>& filenames,
const size_t num_files_to_keep,
const std::shared_ptr<Logger>& info_log,
Env* env) {
if (filenames.size() <= num_files_to_keep) {
return;
}
for (auto iter = std::next(filenames.begin(), num_files_to_keep);
iter != filenames.end(); ++iter) {
if (!env->DeleteFile(iter->second).ok()) {
ROCKS_LOG_WARN(info_log, "Unable to delete options file %s",
iter->second.c_str());
}
}
}
} // namespace
#endif // !ROCKSDB_LITE
Status DBImpl::DeleteObsoleteOptionsFiles() {
#ifndef ROCKSDB_LITE
std::vector<std::string> filenames;
// use ordered map to store keep the filenames sorted from the newest
// to the oldest.
std::map<uint64_t, std::string> options_filenames;
Status s;
s = GetEnv()->GetChildren(GetName(), &filenames);
if (!s.ok()) {
return s;
}
for (auto& filename : filenames) {
uint64_t file_number;
FileType type;
if (ParseFileName(filename, &file_number, &type) && type == kOptionsFile) {
options_filenames.insert(
{std::numeric_limits<uint64_t>::max() - file_number,
GetName() + "/" + filename});
}
}
// Keeps the latest 2 Options file
const size_t kNumOptionsFilesKept = 2;
DeleteOptionsFilesHelper(options_filenames, kNumOptionsFilesKept,
immutable_db_options_.info_log, GetEnv());
return Status::OK();
#else
return Status::OK();
#endif // !ROCKSDB_LITE
}
Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
#ifndef ROCKSDB_LITE
Status s;
uint64_t options_file_number = versions_->NewFileNumber();
std::string options_file_name =
OptionsFileName(GetName(), options_file_number);
Add (Live)FileStorageInfo API (#8968) Summary: New classes FileStorageInfo and LiveFileStorageInfo and 'experimental' function DB::GetLiveFilesStorageInfo, which is intended to largely replace several fragmented DB functions needed to create checkpoints and backups. This function is now used to create checkpoints and backups, because it fixes many (probably not all) of the prior complexities of checkpoint not having atomic access to DB metadata. This also ensures strong functional test coverage of the new API. Specifically, much of the old CheckpointImpl::CreateCustomCheckpoint has been migrated to and updated in DBImpl::GetLiveFilesStorageInfo, with the former now calling the latter. Also, the class FileStorageInfo in metadata.h compatibly replaces BackupFileInfo and serves as a new base class for SstFileMetaData. Some old fields of SstFileMetaData are still provided (for now) but deprecated. Although FileStorageInfo::directory is accurate when using db_paths and/or cf_paths, these have never been supported by Checkpoint nor BackupEngine and still are not. This change does now detect these cases and return NotSupported when appropriate. (More work needed for support.) Somehow this change broke ProgressCallbackDuringBackup, but the progress_callback logic was dubious to begin with because it would call the callback based on copy buffer size, not size actually copied. Logic and test updated to track size actually copied per-thread. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8968 Test Plan: tests updated. DB::GetLiveFilesStorageInfo mostly tested by use in CheckpointImpl. DBTest.SnapshotFiles updated to also test GetLiveFilesStorageInfo, including reading the data after DB close. Added CheckpointTest.CheckpointWithDbPath (NotSupported). Reviewed By: siying Differential Revision: D31242045 Pulled By: pdillinger fbshipit-source-id: b183d1ce9799e220daaefd6b3b5365d98de676c0
3 years ago
uint64_t options_file_size = 0;
s = GetEnv()->GetFileSize(file_name, &options_file_size);
if (s.ok()) {
// Retry if the file name happen to conflict with an existing one.
s = GetEnv()->RenameFile(file_name, options_file_name);
std::unique_ptr<FSDirectory> dir_obj;
if (s.ok()) {
s = fs_->NewDirectory(GetName(), IOOptions(), &dir_obj, nullptr);
}
if (s.ok()) {
s = dir_obj->FsyncWithDirOptions(IOOptions(), nullptr,
DirFsyncOptions(options_file_name));
}
Add (Live)FileStorageInfo API (#8968) Summary: New classes FileStorageInfo and LiveFileStorageInfo and 'experimental' function DB::GetLiveFilesStorageInfo, which is intended to largely replace several fragmented DB functions needed to create checkpoints and backups. This function is now used to create checkpoints and backups, because it fixes many (probably not all) of the prior complexities of checkpoint not having atomic access to DB metadata. This also ensures strong functional test coverage of the new API. Specifically, much of the old CheckpointImpl::CreateCustomCheckpoint has been migrated to and updated in DBImpl::GetLiveFilesStorageInfo, with the former now calling the latter. Also, the class FileStorageInfo in metadata.h compatibly replaces BackupFileInfo and serves as a new base class for SstFileMetaData. Some old fields of SstFileMetaData are still provided (for now) but deprecated. Although FileStorageInfo::directory is accurate when using db_paths and/or cf_paths, these have never been supported by Checkpoint nor BackupEngine and still are not. This change does now detect these cases and return NotSupported when appropriate. (More work needed for support.) Somehow this change broke ProgressCallbackDuringBackup, but the progress_callback logic was dubious to begin with because it would call the callback based on copy buffer size, not size actually copied. Logic and test updated to track size actually copied per-thread. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8968 Test Plan: tests updated. DB::GetLiveFilesStorageInfo mostly tested by use in CheckpointImpl. DBTest.SnapshotFiles updated to also test GetLiveFilesStorageInfo, including reading the data after DB close. Added CheckpointTest.CheckpointWithDbPath (NotSupported). Reviewed By: siying Differential Revision: D31242045 Pulled By: pdillinger fbshipit-source-id: b183d1ce9799e220daaefd6b3b5365d98de676c0
3 years ago
}
if (s.ok()) {
InstrumentedMutexLock l(&mutex_);
versions_->options_file_number_ = options_file_number;
Add (Live)FileStorageInfo API (#8968) Summary: New classes FileStorageInfo and LiveFileStorageInfo and 'experimental' function DB::GetLiveFilesStorageInfo, which is intended to largely replace several fragmented DB functions needed to create checkpoints and backups. This function is now used to create checkpoints and backups, because it fixes many (probably not all) of the prior complexities of checkpoint not having atomic access to DB metadata. This also ensures strong functional test coverage of the new API. Specifically, much of the old CheckpointImpl::CreateCustomCheckpoint has been migrated to and updated in DBImpl::GetLiveFilesStorageInfo, with the former now calling the latter. Also, the class FileStorageInfo in metadata.h compatibly replaces BackupFileInfo and serves as a new base class for SstFileMetaData. Some old fields of SstFileMetaData are still provided (for now) but deprecated. Although FileStorageInfo::directory is accurate when using db_paths and/or cf_paths, these have never been supported by Checkpoint nor BackupEngine and still are not. This change does now detect these cases and return NotSupported when appropriate. (More work needed for support.) Somehow this change broke ProgressCallbackDuringBackup, but the progress_callback logic was dubious to begin with because it would call the callback based on copy buffer size, not size actually copied. Logic and test updated to track size actually copied per-thread. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8968 Test Plan: tests updated. DB::GetLiveFilesStorageInfo mostly tested by use in CheckpointImpl. DBTest.SnapshotFiles updated to also test GetLiveFilesStorageInfo, including reading the data after DB close. Added CheckpointTest.CheckpointWithDbPath (NotSupported). Reviewed By: siying Differential Revision: D31242045 Pulled By: pdillinger fbshipit-source-id: b183d1ce9799e220daaefd6b3b5365d98de676c0
3 years ago
versions_->options_file_size_ = options_file_size;
}
if (0 == disable_delete_obsolete_files_) {
// TODO: Should we check for errors here?
DeleteObsoleteOptionsFiles().PermitUncheckedError();
}
return s;
#else
(void)file_name;
return Status::OK();
#endif // !ROCKSDB_LITE
}
#ifdef ROCKSDB_USING_THREAD_STATUS
void DBImpl::NewThreadStatusCfInfo(ColumnFamilyData* cfd) const {
if (immutable_db_options_.enable_thread_tracking) {
ThreadStatusUtil::NewColumnFamilyInfo(this, cfd, cfd->GetName(),
cfd->ioptions()->env);
}
}
void DBImpl::EraseThreadStatusCfInfo(ColumnFamilyData* cfd) const {
if (immutable_db_options_.enable_thread_tracking) {
ThreadStatusUtil::EraseColumnFamilyInfo(cfd);
}
}
void DBImpl::EraseThreadStatusDbInfo() const {
if (immutable_db_options_.enable_thread_tracking) {
ThreadStatusUtil::EraseDatabaseInfo(this);
}
}
#else
void DBImpl::NewThreadStatusCfInfo(ColumnFamilyData* /*cfd*/) const {}
void DBImpl::EraseThreadStatusCfInfo(ColumnFamilyData* /*cfd*/) const {}
void DBImpl::EraseThreadStatusDbInfo() const {}
#endif // ROCKSDB_USING_THREAD_STATUS
//
// A global method that can dump out the build version
void DumpRocksDBBuildVersion(Logger* log) {
ROCKS_LOG_HEADER(log, "RocksDB version: %s\n",
GetRocksVersionAsString().c_str());
const auto& props = GetRocksBuildProperties();
const auto& sha = props.find("rocksdb_build_git_sha");
if (sha != props.end()) {
ROCKS_LOG_HEADER(log, "Git sha %s", sha->second.c_str());
}
const auto date = props.find("rocksdb_build_date");
if (date != props.end()) {
ROCKS_LOG_HEADER(log, "Compile date %s", date->second.c_str());
}
}
#ifndef ROCKSDB_LITE
SequenceNumber DBImpl::GetEarliestMemTableSequenceNumber(SuperVersion* sv,
bool include_history) {
// Find the earliest sequence number that we know we can rely on reading
// from the memtable without needing to check sst files.
SequenceNumber earliest_seq =
sv->imm->GetEarliestSequenceNumber(include_history);
if (earliest_seq == kMaxSequenceNumber) {
earliest_seq = sv->mem->GetEarliestSequenceNumber();
}
assert(sv->mem->GetEarliestSequenceNumber() >= earliest_seq);
return earliest_seq;
}
Status DBImpl::GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
bool cache_only,
SequenceNumber lower_bound_seq,
SequenceNumber* seq,
bool* found_record_for_key,
bool* is_blob_index) {
Status s;
MergeContext merge_context;
Use only "local" range tombstones during Get (#4449) Summary: Previously, range tombstones were accumulated from every level, which was necessary if a range tombstone in a higher level covered a key in a lower level. However, RangeDelAggregator::AddTombstones's complexity is based on the number of tombstones that are currently stored in it, which is wasteful in the Get case, where we only need to know the highest sequence number of range tombstones that cover the key from higher levels, and compute the highest covering sequence number at the current level. This change introduces this optimization, and removes the use of RangeDelAggregator from the Get path. In the benchmark results, the following command was used to initialize the database: ``` ./db_bench -db=/dev/shm/5k-rts -use_existing_db=false -benchmarks=filluniquerandom -write_buffer_size=1048576 -compression_type=lz4 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -value_size=112 -key_size=16 -block_size=4096 -level_compaction_dynamic_level_bytes=true -num=5000000 -max_background_jobs=12 -benchmark_write_rate_limit=20971520 -range_tombstone_width=100 -writes_per_range_tombstone=100 -max_num_range_tombstones=50000 -bloom_bits=8 ``` ...and the following command was used to measure read throughput: ``` ./db_bench -db=/dev/shm/5k-rts/ -use_existing_db=true -benchmarks=readrandom -disable_auto_compactions=true -num=5000000 -reads=100000 -threads=32 ``` The filluniquerandom command was only run once, and the resulting database was used to measure read performance before and after the PR. Both binaries were compiled with `DEBUG_LEVEL=0`. Readrandom results before PR: ``` readrandom : 4.544 micros/op 220090 ops/sec; 16.9 MB/s (63103 of 100000 found) ``` Readrandom results after PR: ``` readrandom : 11.147 micros/op 89707 ops/sec; 6.9 MB/s (63103 of 100000 found) ``` So it's actually slower right now, but this PR paves the way for future optimizations (see #4493). ---- Pull Request resolved: https://github.com/facebook/rocksdb/pull/4449 Differential Revision: D10370575 Pulled By: abhimadan fbshipit-source-id: 9a2e152be1ef36969055c0e9eb4beb0d96c11f4d
6 years ago
SequenceNumber max_covering_tombstone_seq = 0;
ReadOptions read_options;
SequenceNumber current_seq = versions_->LastSequence();
LookupKey lkey(key, current_seq);
*seq = kMaxSequenceNumber;
*found_record_for_key = false;
// Check if there is a record for this key in the latest memtable
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
sv->mem->Get(lkey, nullptr, nullptr, &s, &merge_context,
&max_covering_tombstone_seq, seq, read_options,
nullptr /*read_callback*/, is_blob_index);
if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
// unexpected error reading memtable.
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Unexpected status returned from MemTable::Get: %s\n",
s.ToString().c_str());
return s;
}
if (*seq != kMaxSequenceNumber) {
// Found a sequence number, no need to check immutable memtables
*found_record_for_key = true;
return Status::OK();
}
SequenceNumber lower_bound_in_mem = sv->mem->GetEarliestSequenceNumber();
if (lower_bound_in_mem != kMaxSequenceNumber &&
lower_bound_in_mem < lower_bound_seq) {
*found_record_for_key = false;
return Status::OK();
}
// Check if there is a record for this key in the immutable memtables
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
sv->imm->Get(lkey, nullptr, nullptr, &s, &merge_context,
&max_covering_tombstone_seq, seq, read_options,
nullptr /*read_callback*/, is_blob_index);
if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
// unexpected error reading memtable.
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Unexpected status returned from MemTableList::Get: %s\n",
s.ToString().c_str());
return s;
}
if (*seq != kMaxSequenceNumber) {
// Found a sequence number, no need to check memtable history
*found_record_for_key = true;
return Status::OK();
}
SequenceNumber lower_bound_in_imm = sv->imm->GetEarliestSequenceNumber();
if (lower_bound_in_imm != kMaxSequenceNumber &&
lower_bound_in_imm < lower_bound_seq) {
*found_record_for_key = false;
return Status::OK();
}
// Check if there is a record for this key in the immutable memtables
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
sv->imm->GetFromHistory(lkey, nullptr, nullptr, &s, &merge_context,
Use only "local" range tombstones during Get (#4449) Summary: Previously, range tombstones were accumulated from every level, which was necessary if a range tombstone in a higher level covered a key in a lower level. However, RangeDelAggregator::AddTombstones's complexity is based on the number of tombstones that are currently stored in it, which is wasteful in the Get case, where we only need to know the highest sequence number of range tombstones that cover the key from higher levels, and compute the highest covering sequence number at the current level. This change introduces this optimization, and removes the use of RangeDelAggregator from the Get path. In the benchmark results, the following command was used to initialize the database: ``` ./db_bench -db=/dev/shm/5k-rts -use_existing_db=false -benchmarks=filluniquerandom -write_buffer_size=1048576 -compression_type=lz4 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -value_size=112 -key_size=16 -block_size=4096 -level_compaction_dynamic_level_bytes=true -num=5000000 -max_background_jobs=12 -benchmark_write_rate_limit=20971520 -range_tombstone_width=100 -writes_per_range_tombstone=100 -max_num_range_tombstones=50000 -bloom_bits=8 ``` ...and the following command was used to measure read throughput: ``` ./db_bench -db=/dev/shm/5k-rts/ -use_existing_db=true -benchmarks=readrandom -disable_auto_compactions=true -num=5000000 -reads=100000 -threads=32 ``` The filluniquerandom command was only run once, and the resulting database was used to measure read performance before and after the PR. Both binaries were compiled with `DEBUG_LEVEL=0`. Readrandom results before PR: ``` readrandom : 4.544 micros/op 220090 ops/sec; 16.9 MB/s (63103 of 100000 found) ``` Readrandom results after PR: ``` readrandom : 11.147 micros/op 89707 ops/sec; 6.9 MB/s (63103 of 100000 found) ``` So it's actually slower right now, but this PR paves the way for future optimizations (see #4493). ---- Pull Request resolved: https://github.com/facebook/rocksdb/pull/4449 Differential Revision: D10370575 Pulled By: abhimadan fbshipit-source-id: 9a2e152be1ef36969055c0e9eb4beb0d96c11f4d
6 years ago
&max_covering_tombstone_seq, seq, read_options,
is_blob_index);
if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
// unexpected error reading memtable.
ROCKS_LOG_ERROR(
immutable_db_options_.info_log,
"Unexpected status returned from MemTableList::GetFromHistory: %s\n",
s.ToString().c_str());
return s;
}
if (*seq != kMaxSequenceNumber) {
// Found a sequence number, no need to check SST files
*found_record_for_key = true;
return Status::OK();
}
// We could do a sv->imm->GetEarliestSequenceNumber(/*include_history*/ true)
// check here to skip the history if possible. But currently the caller
// already does that. Maybe we should move the logic here later.
// TODO(agiardullo): possible optimization: consider checking cached
// SST files if cache_only=true?
if (!cache_only) {
// Check tables
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
5 years ago
sv->current->Get(read_options, lkey, nullptr, nullptr, &s, &merge_context,
Use only "local" range tombstones during Get (#4449) Summary: Previously, range tombstones were accumulated from every level, which was necessary if a range tombstone in a higher level covered a key in a lower level. However, RangeDelAggregator::AddTombstones's complexity is based on the number of tombstones that are currently stored in it, which is wasteful in the Get case, where we only need to know the highest sequence number of range tombstones that cover the key from higher levels, and compute the highest covering sequence number at the current level. This change introduces this optimization, and removes the use of RangeDelAggregator from the Get path. In the benchmark results, the following command was used to initialize the database: ``` ./db_bench -db=/dev/shm/5k-rts -use_existing_db=false -benchmarks=filluniquerandom -write_buffer_size=1048576 -compression_type=lz4 -target_file_size_base=1048576 -max_bytes_for_level_base=4194304 -value_size=112 -key_size=16 -block_size=4096 -level_compaction_dynamic_level_bytes=true -num=5000000 -max_background_jobs=12 -benchmark_write_rate_limit=20971520 -range_tombstone_width=100 -writes_per_range_tombstone=100 -max_num_range_tombstones=50000 -bloom_bits=8 ``` ...and the following command was used to measure read throughput: ``` ./db_bench -db=/dev/shm/5k-rts/ -use_existing_db=true -benchmarks=readrandom -disable_auto_compactions=true -num=5000000 -reads=100000 -threads=32 ``` The filluniquerandom command was only run once, and the resulting database was used to measure read performance before and after the PR. Both binaries were compiled with `DEBUG_LEVEL=0`. Readrandom results before PR: ``` readrandom : 4.544 micros/op 220090 ops/sec; 16.9 MB/s (63103 of 100000 found) ``` Readrandom results after PR: ``` readrandom : 11.147 micros/op 89707 ops/sec; 6.9 MB/s (63103 of 100000 found) ``` So it's actually slower right now, but this PR paves the way for future optimizations (see #4493). ---- Pull Request resolved: https://github.com/facebook/rocksdb/pull/4449 Differential Revision: D10370575 Pulled By: abhimadan fbshipit-source-id: 9a2e152be1ef36969055c0e9eb4beb0d96c11f4d
6 years ago
&max_covering_tombstone_seq, nullptr /* value_found */,
found_record_for_key, seq, nullptr /*read_callback*/,
is_blob_index);
if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
// unexpected error reading SST files
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Unexpected status returned from Version::Get: %s\n",
s.ToString().c_str());
}
}
return s;
}
Status DBImpl::IngestExternalFile(
ColumnFamilyHandle* column_family,
const std::vector<std::string>& external_files,
const IngestExternalFileOptions& ingestion_options) {
IngestExternalFileArg arg;
arg.column_family = column_family;
arg.external_files = external_files;
arg.options = ingestion_options;
return IngestExternalFiles({arg});
}
Status DBImpl::IngestExternalFiles(
const std::vector<IngestExternalFileArg>& args) {
if (args.empty()) {
return Status::InvalidArgument("ingestion arg list is empty");
}
{
std::unordered_set<ColumnFamilyHandle*> unique_cfhs;
for (const auto& arg : args) {
if (arg.column_family == nullptr) {
return Status::InvalidArgument("column family handle is null");
} else if (unique_cfhs.count(arg.column_family) > 0) {
return Status::InvalidArgument(
"ingestion args have duplicate column families");
}
unique_cfhs.insert(arg.column_family);
}
}
// Ingest multiple external SST files atomically.
const size_t num_cfs = args.size();
for (size_t i = 0; i != num_cfs; ++i) {
if (args[i].external_files.empty()) {
char err_msg[128] = {0};
snprintf(err_msg, 128, "external_files[%zu] is empty", i);
return Status::InvalidArgument(err_msg);
}
}
for (const auto& arg : args) {
const IngestExternalFileOptions& ingest_opts = arg.options;
if (ingest_opts.ingest_behind &&
!immutable_db_options_.allow_ingest_behind) {
return Status::InvalidArgument(
"can't ingest_behind file in DB with allow_ingest_behind=false");
}
}
// TODO (yanqin) maybe handle the case in which column_families have
// duplicates
std::unique_ptr<std::list<uint64_t>::iterator> pending_output_elem;
size_t total = 0;
for (const auto& arg : args) {
total += arg.external_files.size();
}
uint64_t next_file_number = 0;
Status status = ReserveFileNumbersBeforeIngestion(
static_cast<ColumnFamilyHandleImpl*>(args[0].column_family)->cfd(), total,
pending_output_elem, &next_file_number);
if (!status.ok()) {
InstrumentedMutexLock l(&mutex_);
ReleaseFileNumberFromPendingOutputs(pending_output_elem);
return status;
}
std::vector<ExternalSstFileIngestionJob> ingestion_jobs;
for (const auto& arg : args) {
auto* cfd = static_cast<ColumnFamilyHandleImpl*>(arg.column_family)->cfd();
ingestion_jobs.emplace_back(versions_.get(), cfd, immutable_db_options_,
file_options_, &snapshots_, arg.options,
&directories_, &event_logger_, io_tracer_);
}
// TODO(yanqin) maybe make jobs run in parallel
uint64_t start_file_number = next_file_number;
for (size_t i = 1; i != num_cfs; ++i) {
start_file_number += args[i - 1].external_files.size();
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[i].column_family)->cfd();
SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
Status es = ingestion_jobs[i].Prepare(
Ingest SST files with checksum information (#6891) Summary: Application can ingest SST files with file checksum information, such that during ingestion, DB is able to check data integrity and identify of the SST file. The PR introduces generate_and_verify_file_checksum to IngestExternalFileOption to control if the ingested checksum information should be verified with the generated checksum. 1. If generate_and_verify_file_checksum options is *FALSE*: *1)* if DB does not enable SST file checksum, the checksum information ingested will be ignored; *2)* if DB enables the SST file checksum and the checksum function name matches the checksum function name in DB, we trust the ingested checksum, store it in Manifest. If the checksum function name does not match, we treat that as an error and fail the IngestExternalFile() call. 2. If generate_and_verify_file_checksum options is *TRUE*: *1)* if DB does not enable SST file checksum, the checksum information ingested will be ignored; *2)* if DB enable the SST file checksum, we will use the checksum generator from DB to calculate the checksum for each ingested SST files after they are copied or moved. Then, compare the checksum results with the ingested checksum information: _A)_ if the checksum function name does not match, _verification always report true_ and we store the DB generated checksum information in Manifest. _B)_ if the checksum function name mach, and checksum match, ingestion continues and stores the checksum information in the Manifest. Otherwise, terminate file ingestion and report file corruption. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6891 Test Plan: added unit test, pass make asan_check Reviewed By: pdillinger Differential Revision: D21935988 Pulled By: zhichao-cao fbshipit-source-id: 7b55f486632db467e76d72602218d0658aa7f6ed
5 years ago
args[i].external_files, args[i].files_checksums,
args[i].files_checksum_func_names, args[i].file_temperature,
start_file_number, super_version);
// capture first error only
if (!es.ok() && status.ok()) {
status = es;
}
CleanupSuperVersion(super_version);
}
TEST_SYNC_POINT("DBImpl::IngestExternalFiles:BeforeLastJobPrepare:0");
TEST_SYNC_POINT("DBImpl::IngestExternalFiles:BeforeLastJobPrepare:1");
{
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[0].column_family)->cfd();
SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
Status es = ingestion_jobs[0].Prepare(
Ingest SST files with checksum information (#6891) Summary: Application can ingest SST files with file checksum information, such that during ingestion, DB is able to check data integrity and identify of the SST file. The PR introduces generate_and_verify_file_checksum to IngestExternalFileOption to control if the ingested checksum information should be verified with the generated checksum. 1. If generate_and_verify_file_checksum options is *FALSE*: *1)* if DB does not enable SST file checksum, the checksum information ingested will be ignored; *2)* if DB enables the SST file checksum and the checksum function name matches the checksum function name in DB, we trust the ingested checksum, store it in Manifest. If the checksum function name does not match, we treat that as an error and fail the IngestExternalFile() call. 2. If generate_and_verify_file_checksum options is *TRUE*: *1)* if DB does not enable SST file checksum, the checksum information ingested will be ignored; *2)* if DB enable the SST file checksum, we will use the checksum generator from DB to calculate the checksum for each ingested SST files after they are copied or moved. Then, compare the checksum results with the ingested checksum information: _A)_ if the checksum function name does not match, _verification always report true_ and we store the DB generated checksum information in Manifest. _B)_ if the checksum function name mach, and checksum match, ingestion continues and stores the checksum information in the Manifest. Otherwise, terminate file ingestion and report file corruption. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6891 Test Plan: added unit test, pass make asan_check Reviewed By: pdillinger Differential Revision: D21935988 Pulled By: zhichao-cao fbshipit-source-id: 7b55f486632db467e76d72602218d0658aa7f6ed
5 years ago
args[0].external_files, args[0].files_checksums,
args[0].files_checksum_func_names, args[0].file_temperature,
next_file_number, super_version);
if (!es.ok()) {
status = es;
}
CleanupSuperVersion(super_version);
}
if (!status.ok()) {
for (size_t i = 0; i != num_cfs; ++i) {
ingestion_jobs[i].Cleanup(status);
}
InstrumentedMutexLock l(&mutex_);
ReleaseFileNumberFromPendingOutputs(pending_output_elem);
return status;
}
std::vector<SuperVersionContext> sv_ctxs;
for (size_t i = 0; i != num_cfs; ++i) {
sv_ctxs.emplace_back(true /* create_superversion */);
}
TEST_SYNC_POINT("DBImpl::IngestExternalFiles:BeforeJobsRun:0");
TEST_SYNC_POINT("DBImpl::IngestExternalFiles:BeforeJobsRun:1");
TEST_SYNC_POINT("DBImpl::AddFile:Start");
{
InstrumentedMutexLock l(&mutex_);
TEST_SYNC_POINT("DBImpl::AddFile:MutexLock");
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
// Stop writes to the DB by entering both write threads
WriteThread::Writer w;
write_thread_.EnterUnbatched(&w, &mutex_);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
WriteThread::Writer nonmem_w;
if (two_write_queues_) {
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
}
// When unordered_write is enabled, the keys are writing to memtable in an
// unordered way. If the ingestion job checks memtable key range before the
// key landing in memtable, the ingestion job may skip the necessary
// memtable flush.
// So wait here to ensure there is no pending write to memtable.
WaitForPendingWrites();
num_running_ingest_file_ += static_cast<int>(num_cfs);
TEST_SYNC_POINT("DBImpl::IngestExternalFile:AfterIncIngestFileCounter");
bool at_least_one_cf_need_flush = false;
std::vector<bool> need_flush(num_cfs, false);
for (size_t i = 0; i != num_cfs; ++i) {
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[i].column_family)->cfd();
if (cfd->IsDropped()) {
// TODO (yanqin) investigate whether we should abort ingestion or
// proceed with other non-dropped column families.
status = Status::InvalidArgument(
"cannot ingest an external file into a dropped CF");
break;
}
bool tmp = false;
status = ingestion_jobs[i].NeedsFlush(&tmp, cfd->GetSuperVersion());
need_flush[i] = tmp;
at_least_one_cf_need_flush = (at_least_one_cf_need_flush || tmp);
if (!status.ok()) {
break;
}
}
TEST_SYNC_POINT_CALLBACK("DBImpl::IngestExternalFile:NeedFlush",
&at_least_one_cf_need_flush);
if (status.ok() && at_least_one_cf_need_flush) {
FlushOptions flush_opts;
flush_opts.allow_write_stall = true;
if (immutable_db_options_.atomic_flush) {
autovector<ColumnFamilyData*> cfds_to_flush;
SelectColumnFamiliesForAtomicFlush(&cfds_to_flush);
mutex_.Unlock();
status = AtomicFlushMemTables(cfds_to_flush, flush_opts,
FlushReason::kExternalFileIngestion,
true /* writes_stopped */);
mutex_.Lock();
} else {
for (size_t i = 0; i != num_cfs; ++i) {
if (need_flush[i]) {
mutex_.Unlock();
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[i].column_family)
->cfd();
status = FlushMemTable(cfd, flush_opts,
FlushReason::kExternalFileIngestion,
true /* writes_stopped */);
mutex_.Lock();
if (!status.ok()) {
break;
}
}
}
}
}
// Run ingestion jobs.
if (status.ok()) {
for (size_t i = 0; i != num_cfs; ++i) {
status = ingestion_jobs[i].Run();
if (!status.ok()) {
break;
}
}
}
if (status.ok()) {
int consumed_seqno_count =
ingestion_jobs[0].ConsumedSequenceNumbersCount();
for (size_t i = 1; i != num_cfs; ++i) {
consumed_seqno_count =
std::max(consumed_seqno_count,
ingestion_jobs[i].ConsumedSequenceNumbersCount());
}
if (consumed_seqno_count > 0) {
const SequenceNumber last_seqno = versions_->LastSequence();
versions_->SetLastAllocatedSequence(last_seqno + consumed_seqno_count);
versions_->SetLastPublishedSequence(last_seqno + consumed_seqno_count);
versions_->SetLastSequence(last_seqno + consumed_seqno_count);
}
autovector<ColumnFamilyData*> cfds_to_commit;
autovector<const MutableCFOptions*> mutable_cf_options_list;
autovector<autovector<VersionEdit*>> edit_lists;
uint32_t num_entries = 0;
for (size_t i = 0; i != num_cfs; ++i) {
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[i].column_family)->cfd();
if (cfd->IsDropped()) {
continue;
}
cfds_to_commit.push_back(cfd);
mutable_cf_options_list.push_back(cfd->GetLatestMutableCFOptions());
autovector<VersionEdit*> edit_list;
edit_list.push_back(ingestion_jobs[i].edit());
edit_lists.push_back(edit_list);
++num_entries;
}
// Mark the version edits as an atomic group if the number of version
// edits exceeds 1.
if (cfds_to_commit.size() > 1) {
for (auto& edits : edit_lists) {
assert(edits.size() == 1);
edits[0]->MarkAtomicGroup(--num_entries);
}
assert(0 == num_entries);
}
status =
versions_->LogAndApply(cfds_to_commit, mutable_cf_options_list,
edit_lists, &mutex_, directories_.GetDbDir());
}
if (status.ok()) {
for (size_t i = 0; i != num_cfs; ++i) {
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[i].column_family)->cfd();
if (!cfd->IsDropped()) {
InstallSuperVersionAndScheduleWork(cfd, &sv_ctxs[i],
*cfd->GetLatestMutableCFOptions());
#ifndef NDEBUG
if (0 == i && num_cfs > 1) {
TEST_SYNC_POINT(
"DBImpl::IngestExternalFiles:InstallSVForFirstCF:0");
TEST_SYNC_POINT(
"DBImpl::IngestExternalFiles:InstallSVForFirstCF:1");
}
#endif // !NDEBUG
}
}
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
5 years ago
} else if (versions_->io_status().IsIOError()) {
// Error while writing to MANIFEST.
// In fact, versions_->io_status() can also be the result of renaming
// CURRENT file. With current code, it's just difficult to tell. So just
// be pessimistic and try write to a new MANIFEST.
// TODO: distinguish between MANIFEST write and CURRENT renaming
const IOStatus& io_s = versions_->io_status();
// Should handle return error?
error_handler_.SetBGError(io_s, BackgroundErrorReason::kManifestWrite);
}
// Resume writes to the DB
if (two_write_queues_) {
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
8 years ago
nonmem_write_thread_.ExitUnbatched(&nonmem_w);
}
write_thread_.ExitUnbatched(&w);
if (status.ok()) {
for (auto& job : ingestion_jobs) {
job.UpdateStats();
}
}
ReleaseFileNumberFromPendingOutputs(pending_output_elem);
num_running_ingest_file_ -= static_cast<int>(num_cfs);
if (0 == num_running_ingest_file_) {
bg_cv_.SignalAll();
}
TEST_SYNC_POINT("DBImpl::AddFile:MutexUnlock");
}
// mutex_ is unlocked here
// Cleanup
for (size_t i = 0; i != num_cfs; ++i) {
sv_ctxs[i].Clean();
// This may rollback jobs that have completed successfully. This is
// intended for atomicity.
ingestion_jobs[i].Cleanup(status);
}
if (status.ok()) {
for (size_t i = 0; i != num_cfs; ++i) {
auto* cfd =
static_cast<ColumnFamilyHandleImpl*>(args[i].column_family)->cfd();
if (!cfd->IsDropped()) {
NotifyOnExternalFileIngested(cfd, ingestion_jobs[i]);
}
}
}
return status;
}
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
Status DBImpl::CreateColumnFamilyWithImport(
const ColumnFamilyOptions& options, const std::string& column_family_name,
const ImportColumnFamilyOptions& import_options,
const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle) {
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
assert(handle != nullptr);
assert(*handle == nullptr);
std::string cf_comparator_name = options.comparator->Name();
if (cf_comparator_name != metadata.db_comparator_name) {
return Status::InvalidArgument("Comparator name mismatch");
}
// Create column family.
auto status = CreateColumnFamily(options, column_family_name, handle);
if (!status.ok()) {
return status;
}
// Import sst files from metadata.
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(*handle);
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
auto cfd = cfh->cfd();
ImportColumnFamilyJob import_job(versions_.get(), cfd, immutable_db_options_,
file_options_, import_options,
metadata.files, io_tracer_);
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
SuperVersionContext dummy_sv_ctx(/* create_superversion */ true);
VersionEdit dummy_edit;
uint64_t next_file_number = 0;
std::unique_ptr<std::list<uint64_t>::iterator> pending_output_elem;
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
{
// Lock db mutex
InstrumentedMutexLock l(&mutex_);
if (error_handler_.IsDBStopped()) {
// Don't import files when there is a bg_error
status = error_handler_.GetBGError();
}
// Make sure that bg cleanup wont delete the files that we are importing
pending_output_elem.reset(new std::list<uint64_t>::iterator(
CaptureCurrentFileNumberInPendingOutputs()));
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
if (status.ok()) {
// If crash happen after a hard link established, Recover function may
// reuse the file number that has already assigned to the internal file,
// and this will overwrite the external file. To protect the external
// file, we have to make sure the file number will never being reused.
next_file_number = versions_->FetchAddFileNumber(metadata.files.size());
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
auto cf_options = cfd->GetLatestMutableCFOptions();
status = versions_->LogAndApply(cfd, *cf_options, &dummy_edit, &mutex_,
directories_.GetDbDir());
if (status.ok()) {
InstallSuperVersionAndScheduleWork(cfd, &dummy_sv_ctx, *cf_options);
}
}
}
dummy_sv_ctx.Clean();
if (status.ok()) {
SuperVersion* sv = cfd->GetReferencedSuperVersion(this);
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
status = import_job.Prepare(next_file_number, sv);
CleanupSuperVersion(sv);
}
if (status.ok()) {
SuperVersionContext sv_context(true /*create_superversion*/);
{
// Lock db mutex
InstrumentedMutexLock l(&mutex_);
// Stop writes to the DB by entering both write threads
WriteThread::Writer w;
write_thread_.EnterUnbatched(&w, &mutex_);
WriteThread::Writer nonmem_w;
if (two_write_queues_) {
nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
}
num_running_ingest_file_++;
assert(!cfd->IsDropped());
status = import_job.Run();
// Install job edit [Mutex will be unlocked here]
if (status.ok()) {
auto cf_options = cfd->GetLatestMutableCFOptions();
status = versions_->LogAndApply(cfd, *cf_options, import_job.edit(),
&mutex_, directories_.GetDbDir());
if (status.ok()) {
InstallSuperVersionAndScheduleWork(cfd, &sv_context, *cf_options);
}
}
// Resume writes to the DB
if (two_write_queues_) {
nonmem_write_thread_.ExitUnbatched(&nonmem_w);
}
write_thread_.ExitUnbatched(&w);
num_running_ingest_file_--;
if (num_running_ingest_file_ == 0) {
bg_cv_.SignalAll();
}
}
// mutex_ is unlocked here
sv_context.Clean();
}
{
InstrumentedMutexLock l(&mutex_);
ReleaseFileNumberFromPendingOutputs(pending_output_elem);
}
import_job.Cleanup(status);
if (!status.ok()) {
Status temp_s = DropColumnFamily(*handle);
if (!temp_s.ok()) {
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"DropColumnFamily failed with error %s",
temp_s.ToString().c_str());
}
// Always returns Status::OK()
temp_s = DestroyColumnFamilyHandle(*handle);
assert(temp_s.ok());
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
6 years ago
*handle = nullptr;
}
return status;
}
Status DBImpl::VerifyFileChecksums(const ReadOptions& read_options) {
return VerifyChecksumInternal(read_options, /*use_file_checksum=*/true);
}
Status DBImpl::VerifyChecksum(const ReadOptions& read_options) {
return VerifyChecksumInternal(read_options, /*use_file_checksum=*/false);
}
Status DBImpl::VerifyChecksumInternal(const ReadOptions& read_options,
bool use_file_checksum) {
// `bytes_read` stat is enabled based on compile-time support and cannot
// be dynamically toggled. So we do not need to worry about `PerfLevel`
// here, unlike many other `IOStatsContext` / `PerfContext` stats.
uint64_t prev_bytes_read = IOSTATS(bytes_read);
Status s;
if (use_file_checksum) {
FileChecksumGenFactory* const file_checksum_gen_factory =
immutable_db_options_.file_checksum_gen_factory.get();
if (!file_checksum_gen_factory) {
s = Status::InvalidArgument(
"Cannot verify file checksum if options.file_checksum_gen_factory is "
"null");
return s;
}
}
std::vector<ColumnFamilyData*> cfd_list;
{
InstrumentedMutexLock l(&mutex_);
for (auto cfd : *versions_->GetColumnFamilySet()) {
if (!cfd->IsDropped() && cfd->initialized()) {
cfd->Ref();
cfd_list.push_back(cfd);
}
}
}
std::vector<SuperVersion*> sv_list;
for (auto cfd : cfd_list) {
sv_list.push_back(cfd->GetReferencedSuperVersion(this));
}
for (auto& sv : sv_list) {
VersionStorageInfo* vstorage = sv->current->storage_info();
ColumnFamilyData* cfd = sv->current->cfd();
Options opts;
if (!use_file_checksum) {
InstrumentedMutexLock l(&mutex_);
opts = Options(BuildDBOptions(immutable_db_options_, mutable_db_options_),
cfd->GetLatestCFOptions());
}
for (int i = 0; i < vstorage->num_non_empty_levels() && s.ok(); i++) {
for (size_t j = 0; j < vstorage->LevelFilesBrief(i).num_files && s.ok();
j++) {
const auto& fd_with_krange = vstorage->LevelFilesBrief(i).files[j];
const auto& fd = fd_with_krange.fd;
const FileMetaData* fmeta = fd_with_krange.file_metadata;
assert(fmeta);
std::string fname = TableFileName(cfd->ioptions()->cf_paths,
fd.GetNumber(), fd.GetPathId());
if (use_file_checksum) {
s = VerifyFullFileChecksum(fmeta->file_checksum,
fmeta->file_checksum_func_name, fname,
read_options);
} else {
s = ROCKSDB_NAMESPACE::VerifySstFileChecksum(opts, file_options_,
read_options, fname);
}
RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
IOSTATS(bytes_read) - prev_bytes_read);
prev_bytes_read = IOSTATS(bytes_read);
}
}
if (s.ok() && use_file_checksum) {
const auto& blob_files = vstorage->GetBlobFiles();
for (const auto& pair : blob_files) {
const uint64_t blob_file_number = pair.first;
const auto& meta = pair.second;
assert(meta);
const std::string blob_file_name = BlobFileName(
cfd->ioptions()->cf_paths.front().path, blob_file_number);
s = VerifyFullFileChecksum(meta->GetChecksumValue(),
meta->GetChecksumMethod(), blob_file_name,
read_options);
RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
IOSTATS(bytes_read) - prev_bytes_read);
prev_bytes_read = IOSTATS(bytes_read);
if (!s.ok()) {
break;
}
}
}
if (!s.ok()) {
break;
}
}
bool defer_purge =
immutable_db_options().avoid_unnecessary_blocking_io;
{
InstrumentedMutexLock l(&mutex_);
for (auto sv : sv_list) {
if (sv && sv->Unref()) {
sv->Cleanup();
if (defer_purge) {
AddSuperVersionsToFreeQueue(sv);
} else {
delete sv;
}
}
}
if (defer_purge) {
SchedulePurge();
}
for (auto cfd : cfd_list) {
cfd->UnrefAndTryDelete();
}
}
RecordTick(stats_, VERIFY_CHECKSUM_READ_BYTES,
IOSTATS(bytes_read) - prev_bytes_read);
return s;
}
Status DBImpl::VerifyFullFileChecksum(const std::string& file_checksum_expected,
const std::string& func_name_expected,
const std::string& fname,
const ReadOptions& read_options) {
Status s;
if (file_checksum_expected == kUnknownFileChecksum) {
return s;
}
std::string file_checksum;
std::string func_name;
s = ROCKSDB_NAMESPACE::GenerateOneFileChecksum(
fs_.get(), fname, immutable_db_options_.file_checksum_gen_factory.get(),
func_name_expected, &file_checksum, &func_name,
read_options.readahead_size, immutable_db_options_.allow_mmap_reads,
io_tracer_, immutable_db_options_.rate_limiter.get());
if (s.ok()) {
assert(func_name_expected == func_name);
if (file_checksum != file_checksum_expected) {
std::ostringstream oss;
oss << fname << " file checksum mismatch, ";
oss << "expecting "
<< Slice(file_checksum_expected).ToString(/*hex=*/true);
oss << ", but actual " << Slice(file_checksum).ToString(/*hex=*/true);
s = Status::Corruption(oss.str());
TEST_SYNC_POINT_CALLBACK("DBImpl::VerifyFullFileChecksum:mismatch", &s);
}
}
return s;
}
void DBImpl::NotifyOnExternalFileIngested(
ColumnFamilyData* cfd, const ExternalSstFileIngestionJob& ingestion_job) {
if (immutable_db_options_.listeners.empty()) {
return;
}
for (const IngestedFileInfo& f : ingestion_job.files_to_ingest()) {
ExternalFileIngestionInfo info;
info.cf_name = cfd->GetName();
info.external_file_path = f.external_file_path;
info.internal_file_path = f.internal_file_path;
info.global_seqno = f.assigned_seqno;
info.table_properties = f.table_properties;
for (auto listener : immutable_db_options_.listeners) {
listener->OnExternalFileIngested(this, info);
}
}
}
void DBImpl::WaitForIngestFile() {
mutex_.AssertHeld();
while (num_running_ingest_file_ > 0) {
bg_cv_.Wait();
}
}
Status DBImpl::StartTrace(const TraceOptions& trace_options,
std::unique_ptr<TraceWriter>&& trace_writer) {
InstrumentedMutexLock lock(&trace_mutex_);
tracer_.reset(new Tracer(immutable_db_options_.clock, trace_options,
std::move(trace_writer)));
return Status::OK();
}
Status DBImpl::EndTrace() {
InstrumentedMutexLock lock(&trace_mutex_);
Status s;
if (tracer_ != nullptr) {
s = tracer_->Close();
tracer_.reset();
} else {
s = Status::IOError("No trace file to close");
}
return s;
}
Status DBImpl::NewDefaultReplayer(
const std::vector<ColumnFamilyHandle*>& handles,
std::unique_ptr<TraceReader>&& reader,
std::unique_ptr<Replayer>* replayer) {
replayer->reset(new ReplayerImpl(this, handles, std::move(reader)));
return Status::OK();
}
Status DBImpl::StartBlockCacheTrace(
const TraceOptions& trace_options,
std::unique_ptr<TraceWriter>&& trace_writer) {
return block_cache_tracer_.StartTrace(immutable_db_options_.clock,
trace_options, std::move(trace_writer));
}
Status DBImpl::EndBlockCacheTrace() {
block_cache_tracer_.EndTrace();
return Status::OK();
}
Status DBImpl::TraceIteratorSeek(const uint32_t& cf_id, const Slice& key,
const Slice& lower_bound,
const Slice upper_bound) {
Status s;
if (tracer_) {
InstrumentedMutexLock lock(&trace_mutex_);
if (tracer_) {
s = tracer_->IteratorSeek(cf_id, key, lower_bound, upper_bound);
}
}
return s;
}
Status DBImpl::TraceIteratorSeekForPrev(const uint32_t& cf_id, const Slice& key,
const Slice& lower_bound,
const Slice upper_bound) {
Status s;
if (tracer_) {
InstrumentedMutexLock lock(&trace_mutex_);
if (tracer_) {
s = tracer_->IteratorSeekForPrev(cf_id, key, lower_bound, upper_bound);
}
}
return s;
}
Status DBImpl::ReserveFileNumbersBeforeIngestion(
ColumnFamilyData* cfd, uint64_t num,
std::unique_ptr<std::list<uint64_t>::iterator>& pending_output_elem,
uint64_t* next_file_number) {
Status s;
SuperVersionContext dummy_sv_ctx(true /* create_superversion */);
assert(nullptr != next_file_number);
InstrumentedMutexLock l(&mutex_);
if (error_handler_.IsDBStopped()) {
// Do not ingest files when there is a bg_error
return error_handler_.GetBGError();
}
pending_output_elem.reset(new std::list<uint64_t>::iterator(
CaptureCurrentFileNumberInPendingOutputs()));
*next_file_number = versions_->FetchAddFileNumber(static_cast<uint64_t>(num));
auto cf_options = cfd->GetLatestMutableCFOptions();
VersionEdit dummy_edit;
// If crash happen after a hard link established, Recover function may
// reuse the file number that has already assigned to the internal file,
// and this will overwrite the external file. To protect the external
// file, we have to make sure the file number will never being reused.
s = versions_->LogAndApply(cfd, *cf_options, &dummy_edit, &mutex_,
directories_.GetDbDir());
if (s.ok()) {
InstallSuperVersionAndScheduleWork(cfd, &dummy_sv_ctx, *cf_options);
}
dummy_sv_ctx.Clean();
return s;
}
Status DBImpl::GetCreationTimeOfOldestFile(uint64_t* creation_time) {
if (mutable_db_options_.max_open_files == -1) {
uint64_t oldest_time = port::kMaxUint64;
for (auto cfd : *versions_->GetColumnFamilySet()) {
Fix data race of GetCreationTimeOfOldestFile() (#6473) Summary: When DBImpl::GetCreationTimeOfOldestFile() calls Version::GetCreationTimeOfOldestFile(), the version is not directly or indirectly referenced, so an event like compaction can race with the operation and cause DBImpl::GetCreationTimeOfOldestFile() to access delocated data. This was caught by an ASAN run: ==268==ERROR: AddressSanitizer: heap-use-after-free on address 0x612000b7d198 at pc 0x000018332913 bp 0x7f391510d310 sp 0x7f391510d308 READ of size 8 at 0x612000b7d198 thread T845 (store_load-33) SCARINESS: 51 (8-byte-read-heap-use-after-free) #0 0x18332912 in rocksdb::Version::GetCreationTimeOfOldestFile(unsigned long*) rocksdb/src/db/version_set.cc:1488 https://github.com/facebook/rocksdb/issues/1 0x1803ddaa in rocksdb::DBImpl::GetCreationTimeOfOldestFile(unsigned long*) rocksdb/src/db/db_impl/db_impl.cc:4499 https://github.com/facebook/rocksdb/issues/2 0xe24ca09 in rocksdb::StackableDB::GetCreationTimeOfOldestFile(unsigned long*) rocksdb/utilities/stackable_db.h:392 ...... 0x612000b7d198 is located 216 bytes inside of 296-byte region [0x612000b7d0c0,0x612000b7d1e8) freed by thread T28 here: ...... https://github.com/facebook/rocksdb/issues/5 0x1832c73f in std::vector<rocksdb::FileMetaData*, std::allocator<rocksdb::FileMetaData*> >::~vector() third-party-buck/platform007/build/libgcc/include/c++/trunk/bits/stl_vector.h:435 https://github.com/facebook/rocksdb/issues/6 0x1832c73f in rocksdb::VersionStorageInfo::~VersionStorageInfo() rocksdb/src/db/version_set.cc:734 https://github.com/facebook/rocksdb/issues/7 0x1832cf42 in rocksdb::Version::~Version() rocksdb/src/db/version_set.cc:758 https://github.com/facebook/rocksdb/issues/8 0x9d1bb5 in rocksdb::Version::Unref() rocksdb/src/db/version_set.cc:2869 https://github.com/facebook/rocksdb/issues/9 0x183e7631 in rocksdb::Compaction::~Compaction() rocksdb/src/db/compaction/compaction.cc:275 https://github.com/facebook/rocksdb/issues/10 0x9e6de6 in std::default_delete<rocksdb::Compaction>::operator()(rocksdb::Compaction*) const third-party-buck/platform007/build/libgcc/include/c++/trunk/bits/unique_ptr.h:78 https://github.com/facebook/rocksdb/issues/11 0x9e6de6 in std::unique_ptr<rocksdb::Compaction, std::default_delete<rocksdb::Compaction> >::reset(rocksdb::Compaction*) third-party-buck/platform007/build/libgcc/include/c++/trunk/bits/unique_ptr.h:376 https://github.com/facebook/rocksdb/issues/12 0x9e6de6 in rocksdb::DBImpl::BackgroundCompaction(bool*, rocksdb::JobContext*, rocksdb::LogBuffer*, rocksdb::DBImpl::PrepickedCompaction*, rocksdb::Env::Priority) rocksdb/src/db/db_impl/db_impl_compaction_flush.cc:2826 https://github.com/facebook/rocksdb/issues/13 0x9ac3b8 in rocksdb::DBImpl::BackgroundCallCompaction(rocksdb::DBImpl::PrepickedCompaction*, rocksdb::Env::Priority) rocksdb/src/db/db_impl/db_impl_compaction_flush.cc:2320 https://github.com/facebook/rocksdb/issues/14 0x9abff7 in rocksdb::DBImpl::BGWorkCompaction(void*) rocksdb/src/db/db_impl/db_impl_compaction_flush.cc:2096 ...... Fix the issue by reference the super version and use the referenced version from it. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6473 Test Plan: Run ASAN for all existing tests. Differential Revision: D20196416 fbshipit-source-id: 5f4a7918110fc7b8dd7841932d376bc9d1e59d6f
5 years ago
if (!cfd->IsDropped()) {
uint64_t ctime;
{
SuperVersion* sv = GetAndRefSuperVersion(cfd);
Version* version = sv->current;
version->GetCreationTimeOfOldestFile(&ctime);
ReturnAndCleanupSuperVersion(cfd, sv);
}
if (ctime < oldest_time) {
oldest_time = ctime;
}
if (oldest_time == 0) {
break;
}
}
}
*creation_time = oldest_time;
return Status::OK();
} else {
return Status::NotSupported("This API only works if max_open_files = -1");
}
}
#endif // ROCKSDB_LITE
} // namespace ROCKSDB_NAMESPACE