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/table_cache.cc

474 lines
18 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/table_cache.h"
#include "db/dbformat.h"
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
#include "db/range_tombstone_fragmenter.h"
#include "db/version_edit.h"
#include "util/filename.h"
#include "monitoring/perf_context_imp.h"
#include "rocksdb/statistics.h"
#include "table/get_context.h"
#include "table/internal_iterator.h"
#include "table/iterator_wrapper.h"
#include "table/table_builder.h"
#include "table/table_reader.h"
#include "util/coding.h"
#include "util/file_reader_writer.h"
#include "util/stop_watch.h"
#include "util/sync_point.h"
namespace rocksdb {
namespace {
template <class T>
static void DeleteEntry(const Slice& /*key*/, void* value) {
T* typed_value = reinterpret_cast<T*>(value);
delete typed_value;
}
static void UnrefEntry(void* arg1, void* arg2) {
Cache* cache = reinterpret_cast<Cache*>(arg1);
Cache::Handle* h = reinterpret_cast<Cache::Handle*>(arg2);
cache->Release(h);
}
static void DeleteTableReader(void* arg1, void* /*arg2*/) {
TableReader* table_reader = reinterpret_cast<TableReader*>(arg1);
delete table_reader;
}
static Slice GetSliceForFileNumber(const uint64_t* file_number) {
return Slice(reinterpret_cast<const char*>(file_number),
sizeof(*file_number));
}
#ifndef ROCKSDB_LITE
void AppendVarint64(IterKey* key, uint64_t v) {
char buf[10];
auto ptr = EncodeVarint64(buf, v);
key->TrimAppend(key->Size(), buf, ptr - buf);
}
#endif // ROCKSDB_LITE
} // namespace
TableCache::TableCache(const ImmutableCFOptions& ioptions,
const EnvOptions& env_options, Cache* const cache)
: ioptions_(ioptions),
env_options_(env_options),
cache_(cache),
immortal_tables_(false) {
if (ioptions_.row_cache) {
// If the same cache is shared by multiple instances, we need to
// disambiguate its entries.
PutVarint64(&row_cache_id_, ioptions_.row_cache->NewId());
}
}
TableCache::~TableCache() {
}
TableReader* TableCache::GetTableReaderFromHandle(Cache::Handle* handle) {
return reinterpret_cast<TableReader*>(cache_->Value(handle));
}
void TableCache::ReleaseHandle(Cache::Handle* handle) {
cache_->Release(handle);
}
Status TableCache::GetTableReader(
const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
bool sequential_mode, size_t readahead, bool record_read_stats,
HistogramImpl* file_read_hist, unique_ptr<TableReader>* table_reader,
const SliceTransform* prefix_extractor, bool skip_filters, int level,
bool prefetch_index_and_filter_in_cache, bool for_compaction) {
std::string fname =
TableFileName(ioptions_.cf_paths, fd.GetNumber(), fd.GetPathId());
unique_ptr<RandomAccessFile> file;
Status s = ioptions_.env->NewRandomAccessFile(fname, &file, env_options);
RecordTick(ioptions_.statistics, NO_FILE_OPENS);
if (s.ok()) {
if (readahead > 0 && !env_options.use_mmap_reads) {
// Not compatible with mmap files since ReadaheadRandomAccessFile requires
// its wrapped file's Read() to copy data into the provided scratch
// buffer, which mmap files don't use.
// TODO(ajkr): try madvise for mmap files in place of buffered readahead.
Fix a crash when compaction fails to open a file Summary: We've got a crash with this stack trace: Program terminated with signal SIGTRAP, Trace/breakpoint trap. #0 0x00007fc85f2f4009 in raise () from /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libpthread.so.0 #1 0x00000000005c8f61 in facebook::logdevice::handle_sigsegv(int) () at logdevice/server/sigsegv.cpp:159 #2 0x00007fc85f2f4150 in <signal handler called> () at /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libpthread.so.0 #3 0x00000000031ed80c in rocksdb::NewReadaheadRandomAccessFile() at util/file_reader_writer.cc:383 #4 0x00000000031ed80c in rocksdb::NewReadaheadRandomAccessFile() at util/file_reader_writer.cc:472 #5 0x00000000031558e7 in rocksdb::TableCache::GetTableReader() at db/table_cache.cc:99 #6 0x0000000003156329 in rocksdb::TableCache::NewIterator() at db/table_cache.cc:198 #7 0x0000000003166568 in rocksdb::VersionSet::MakeInputIterator() at db/version_set.cc:3345 #8 0x000000000324a94f in rocksdb::CompactionJob::ProcessKeyValueCompaction(rocksdb::CompactionJob::SubcompactionState*) () at db/compaction_job.cc:650 #9 0x000000000324c2f6 in rocksdb::CompactionJob::Run() () at db/compaction_job.cc:530 #10 0x00000000030f5ae5 in rocksdb::DBImpl::BackgroundCompaction() at db/db_impl.cc:3269 #11 0x0000000003108d36 in rocksdb::DBImpl::BackgroundCallCompaction(void*) () at db/db_impl.cc:2970 #12 0x00000000029a2a9a in facebook::logdevice::RocksDBEnv::callback(void*) () at logdevice/server/locallogstore/RocksDBEnv.cpp:26 #13 0x00000000029a2a9a in facebook::logdevice::RocksDBEnv::callback(void*) () at logdevice/server/locallogstore/RocksDBEnv.cpp:30 #14 0x00000000031e7521 in rocksdb::ThreadPool::BGThread() at util/threadpool.cc:230 #15 0x00000000031e7663 in rocksdb::BGThreadWrapper(void*) () at util/threadpool.cc:254 #16 0x00007fc85f2ea7f1 in start_thread () at /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libpthread.so.0 #17 0x00007fc85e8fb46d in clone () at /usr/local/fbcode/gcc-4.9-glibc-2.20-fb/lib/libc.so.6 From looking at the code, probably what happened is this: - `TableCache::GetTableReader()` called `Env::NewRandomAccessFile()`, which dispatched to a `PosixEnv::NewRandomAccessFile()`, where probably an `open()` call failed, so the `NewRandomAccessFile()` left a nullptr in the resulting file, - `TableCache::GetTableReader()` called `NewReadaheadRandomAccessFile()` with that `nullptr` file, - it tried to call file's method and crashed. This diff is a trivial fix to this crash. Test Plan: `make -j check` Reviewers: sdong, andrewkr, IslamAbdelRahman Reviewed By: IslamAbdelRahman Subscribers: andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D62451
8 years ago
file = NewReadaheadRandomAccessFile(std::move(file), readahead);
}
if (!sequential_mode && ioptions_.advise_random_on_open) {
file->Hint(RandomAccessFile::RANDOM);
}
StopWatch sw(ioptions_.env, ioptions_.statistics, TABLE_OPEN_IO_MICROS);
std::unique_ptr<RandomAccessFileReader> file_reader(
new RandomAccessFileReader(
std::move(file), fname, ioptions_.env,
record_read_stats ? ioptions_.statistics : nullptr, SST_READ_MICROS,
file_read_hist, ioptions_.rate_limiter, for_compaction,
ioptions_.listeners));
s = ioptions_.table_factory->NewTableReader(
TableReaderOptions(ioptions_, prefix_extractor, env_options,
internal_comparator, skip_filters, immortal_tables_,
level, fd.largest_seqno),
std::move(file_reader), fd.GetFileSize(), table_reader,
prefetch_index_and_filter_in_cache);
TEST_SYNC_POINT("TableCache::GetTableReader:0");
}
return s;
}
void TableCache::EraseHandle(const FileDescriptor& fd, Cache::Handle* handle) {
ReleaseHandle(handle);
uint64_t number = fd.GetNumber();
Slice key = GetSliceForFileNumber(&number);
cache_->Erase(key);
}
Status TableCache::FindTable(const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator,
const FileDescriptor& fd, Cache::Handle** handle,
const SliceTransform* prefix_extractor,
const bool no_io, bool record_read_stats,
HistogramImpl* file_read_hist, bool skip_filters,
int level,
bool prefetch_index_and_filter_in_cache) {
PERF_TIMER_GUARD(find_table_nanos);
Status s;
uint64_t number = fd.GetNumber();
Slice key = GetSliceForFileNumber(&number);
*handle = cache_->Lookup(key);
TEST_SYNC_POINT_CALLBACK("TableCache::FindTable:0",
const_cast<bool*>(&no_io));
if (*handle == nullptr) {
if (no_io) { // Don't do IO and return a not-found status
return Status::Incomplete("Table not found in table_cache, no_io is set");
}
unique_ptr<TableReader> table_reader;
s = GetTableReader(env_options, internal_comparator, fd,
false /* sequential mode */, 0 /* readahead */,
record_read_stats, file_read_hist, &table_reader,
prefix_extractor, skip_filters, level,
prefetch_index_and_filter_in_cache);
if (!s.ok()) {
assert(table_reader == nullptr);
RecordTick(ioptions_.statistics, NO_FILE_ERRORS);
// We do not cache error results so that if the error is transient,
// or somebody repairs the file, we recover automatically.
} else {
s = cache_->Insert(key, table_reader.get(), 1, &DeleteEntry<TableReader>,
handle);
if (s.ok()) {
// Release ownership of table reader.
table_reader.release();
}
}
}
return s;
}
InternalIterator* TableCache::NewIterator(
const ReadOptions& options, const EnvOptions& env_options,
const InternalKeyComparator& icomparator, const FileMetaData& file_meta,
RangeDelAggregator* range_del_agg, const SliceTransform* prefix_extractor,
TableReader** table_reader_ptr, HistogramImpl* file_read_hist,
bool for_compaction, Arena* arena, bool skip_filters, int level,
const InternalKey* smallest_compaction_key,
const InternalKey* largest_compaction_key) {
PERF_TIMER_GUARD(new_table_iterator_nanos);
Status s;
bool create_new_table_reader = false;
TableReader* table_reader = nullptr;
Cache::Handle* handle = nullptr;
if (table_reader_ptr != nullptr) {
*table_reader_ptr = nullptr;
}
size_t readahead = 0;
if (for_compaction) {
#ifndef NDEBUG
bool use_direct_reads_for_compaction = env_options.use_direct_reads;
TEST_SYNC_POINT_CALLBACK("TableCache::NewIterator:for_compaction",
&use_direct_reads_for_compaction);
#endif // !NDEBUG
if (ioptions_.new_table_reader_for_compaction_inputs) {
// get compaction_readahead_size from env_options allows us to set the
// value dynamically
readahead = env_options.compaction_readahead_size;
create_new_table_reader = true;
}
} else {
readahead = options.readahead_size;
create_new_table_reader = readahead > 0;
}
auto& fd = file_meta.fd;
if (create_new_table_reader) {
unique_ptr<TableReader> table_reader_unique_ptr;
s = GetTableReader(
env_options, icomparator, fd, true /* sequential_mode */, readahead,
!for_compaction /* record stats */, nullptr, &table_reader_unique_ptr,
prefix_extractor, false /* skip_filters */, level,
true /* prefetch_index_and_filter_in_cache */, for_compaction);
if (s.ok()) {
table_reader = table_reader_unique_ptr.release();
}
} else {
table_reader = fd.table_reader;
if (table_reader == nullptr) {
s = FindTable(env_options, icomparator, fd, &handle, prefix_extractor,
options.read_tier == kBlockCacheTier /* no_io */,
!for_compaction /* record read_stats */, file_read_hist,
skip_filters, level);
if (s.ok()) {
table_reader = GetTableReaderFromHandle(handle);
}
}
}
InternalIterator* result = nullptr;
if (s.ok()) {
expose a hook to skip tables during iteration Summary: As discussed on the mailing list (["Skipping entire SSTs while iterating"](https://groups.google.com/forum/#!topic/rocksdb/ujHCJVLrHlU)), this patch adds a `table_filter` to `ReadOptions` that allows specifying a callback to be executed during iteration before each table in the database is scanned. The callback is passed the table's properties; the table is scanned iff the callback returns true. This can be used in conjunction with a `TablePropertiesCollector` to dramatically speed up scans by skipping tables that are known to contain irrelevant data for the scan at hand. We're using this [downstream in CockroachDB](https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/engine/db.cc#L2009-L2022) already. With this feature, under ideal conditions, we can reduce the time of an incremental backup in from hours to seconds. FYI, the first commit in this PR fixes a segfault that I unfortunately have not figured out how to reproduce outside of CockroachDB. I'm hoping you accept it on the grounds that it is not correct to return 8-byte aligned memory from a call to `malloc` on some 64-bit platforms; one correct approach is to infer the necessary alignment from `std::max_align_t`, as done here. As noted in the first commit message, the bug is tickled by having a`std::function` in `struct ReadOptions`. That is, the following patch alone is enough to cause RocksDB to segfault when run from CockroachDB on Darwin. ```diff --- a/include/rocksdb/options.h +++ b/include/rocksdb/options.h @@ -1546,6 +1546,13 @@ struct ReadOptions { // Default: false bool ignore_range_deletions; + // A callback to determine whether relevant keys for this scan exist in a + // given table based on the table's properties. The callback is passed the + // properties of each table during iteration. If the callback returns false, + // the table will not be scanned. + // Default: empty (every table will be scanned) + std::function<bool(const TableProperties&)> table_filter; + ReadOptions(); ReadOptions(bool cksum, bool cache); }; ``` /cc danhhz Closes https://github.com/facebook/rocksdb/pull/2265 Differential Revision: D5054262 Pulled By: yiwu-arbug fbshipit-source-id: dd6b28f2bba6cb8466250d8c5c542d3c92785476
7 years ago
if (options.table_filter &&
!options.table_filter(*table_reader->GetTableProperties())) {
result = NewEmptyInternalIterator<Slice>(arena);
expose a hook to skip tables during iteration Summary: As discussed on the mailing list (["Skipping entire SSTs while iterating"](https://groups.google.com/forum/#!topic/rocksdb/ujHCJVLrHlU)), this patch adds a `table_filter` to `ReadOptions` that allows specifying a callback to be executed during iteration before each table in the database is scanned. The callback is passed the table's properties; the table is scanned iff the callback returns true. This can be used in conjunction with a `TablePropertiesCollector` to dramatically speed up scans by skipping tables that are known to contain irrelevant data for the scan at hand. We're using this [downstream in CockroachDB](https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/engine/db.cc#L2009-L2022) already. With this feature, under ideal conditions, we can reduce the time of an incremental backup in from hours to seconds. FYI, the first commit in this PR fixes a segfault that I unfortunately have not figured out how to reproduce outside of CockroachDB. I'm hoping you accept it on the grounds that it is not correct to return 8-byte aligned memory from a call to `malloc` on some 64-bit platforms; one correct approach is to infer the necessary alignment from `std::max_align_t`, as done here. As noted in the first commit message, the bug is tickled by having a`std::function` in `struct ReadOptions`. That is, the following patch alone is enough to cause RocksDB to segfault when run from CockroachDB on Darwin. ```diff --- a/include/rocksdb/options.h +++ b/include/rocksdb/options.h @@ -1546,6 +1546,13 @@ struct ReadOptions { // Default: false bool ignore_range_deletions; + // A callback to determine whether relevant keys for this scan exist in a + // given table based on the table's properties. The callback is passed the + // properties of each table during iteration. If the callback returns false, + // the table will not be scanned. + // Default: empty (every table will be scanned) + std::function<bool(const TableProperties&)> table_filter; + ReadOptions(); ReadOptions(bool cksum, bool cache); }; ``` /cc danhhz Closes https://github.com/facebook/rocksdb/pull/2265 Differential Revision: D5054262 Pulled By: yiwu-arbug fbshipit-source-id: dd6b28f2bba6cb8466250d8c5c542d3c92785476
7 years ago
} else {
result = table_reader->NewIterator(options, prefix_extractor, arena,
skip_filters, for_compaction);
expose a hook to skip tables during iteration Summary: As discussed on the mailing list (["Skipping entire SSTs while iterating"](https://groups.google.com/forum/#!topic/rocksdb/ujHCJVLrHlU)), this patch adds a `table_filter` to `ReadOptions` that allows specifying a callback to be executed during iteration before each table in the database is scanned. The callback is passed the table's properties; the table is scanned iff the callback returns true. This can be used in conjunction with a `TablePropertiesCollector` to dramatically speed up scans by skipping tables that are known to contain irrelevant data for the scan at hand. We're using this [downstream in CockroachDB](https://github.com/cockroachdb/cockroach/blob/master/pkg/storage/engine/db.cc#L2009-L2022) already. With this feature, under ideal conditions, we can reduce the time of an incremental backup in from hours to seconds. FYI, the first commit in this PR fixes a segfault that I unfortunately have not figured out how to reproduce outside of CockroachDB. I'm hoping you accept it on the grounds that it is not correct to return 8-byte aligned memory from a call to `malloc` on some 64-bit platforms; one correct approach is to infer the necessary alignment from `std::max_align_t`, as done here. As noted in the first commit message, the bug is tickled by having a`std::function` in `struct ReadOptions`. That is, the following patch alone is enough to cause RocksDB to segfault when run from CockroachDB on Darwin. ```diff --- a/include/rocksdb/options.h +++ b/include/rocksdb/options.h @@ -1546,6 +1546,13 @@ struct ReadOptions { // Default: false bool ignore_range_deletions; + // A callback to determine whether relevant keys for this scan exist in a + // given table based on the table's properties. The callback is passed the + // properties of each table during iteration. If the callback returns false, + // the table will not be scanned. + // Default: empty (every table will be scanned) + std::function<bool(const TableProperties&)> table_filter; + ReadOptions(); ReadOptions(bool cksum, bool cache); }; ``` /cc danhhz Closes https://github.com/facebook/rocksdb/pull/2265 Differential Revision: D5054262 Pulled By: yiwu-arbug fbshipit-source-id: dd6b28f2bba6cb8466250d8c5c542d3c92785476
7 years ago
}
if (create_new_table_reader) {
assert(handle == nullptr);
result->RegisterCleanup(&DeleteTableReader, table_reader, nullptr);
} else if (handle != nullptr) {
result->RegisterCleanup(&UnrefEntry, cache_, handle);
handle = nullptr; // prevent from releasing below
}
if (for_compaction) {
table_reader->SetupForCompaction();
}
if (table_reader_ptr != nullptr) {
*table_reader_ptr = table_reader;
}
}
if (s.ok() && range_del_agg != nullptr && !options.ignore_range_deletions) {
Recommit "Avoid adding tombstones of the same file to RangeDelAggregator multiple times" Summary: The origin commit #3635 will hurt performance for users who aren't using range deletions, because unneeded std::set operations, so it was reverted by commit 44653c7b7aabe821e671946e732dda7ae6b43d1b. (see #3672) To fix this, move the set to and add a check in , i.e., file will be added only if is non-nullptr. The db_bench command which find the performance regression: > ./db_bench --benchmarks=fillrandom,seekrandomwhilewriting --threads=1 --num=1000000 --reads=150000 --key_size=66 > --value_size=1262 --statistics=0 --compression_ratio=0.5 --histogram=1 --seek_nexts=1 --stats_per_interval=1 > --stats_interval_seconds=600 --max_background_flushes=4 --num_multi_db=1 --max_background_compactions=16 --seed=1522388277 > -write_buffer_size=1048576 --level0_file_num_compaction_trigger=10000 --compression_type=none Before and after the modification, I re-run this command on the machine, the results of are as follows: **fillrandom** Table | P50 | P75 | P99 | P99.9 | P99.99 | ---- | --- | --- | --- | ----- | ------ | before commit | 5.92 | 8.57 | 19.63 | 980.97 | 12196.00 | after commit | 5.91 | 8.55 | 19.34 | 965.56 | 13513.56 | **seekrandomwhilewriting** Table | P50 | P75 | P99 | P99.9 | P99.99 | ---- | --- | --- | --- | ----- | ------ | before commit | 1418.62 | 1867.01 | 3823.28 | 4980.99 | 9240.00 | after commit | 1450.54 | 1880.61 | 3962.87 | 5429.60 | 7542.86 | Closes https://github.com/facebook/rocksdb/pull/3800 Differential Revision: D7874245 Pulled By: ajkr fbshipit-source-id: 2e8bec781b3f7399246babd66395c88619534a17
7 years ago
if (range_del_agg->AddFile(fd.GetNumber())) {
std::unique_ptr<InternalIterator> range_del_iter(
table_reader->NewRangeTombstoneIterator(options));
if (range_del_iter != nullptr) {
s = range_del_iter->status();
}
if (s.ok()) {
const InternalKey* smallest = &file_meta.smallest;
const InternalKey* largest = &file_meta.largest;
if (smallest_compaction_key != nullptr) {
smallest = smallest_compaction_key;
}
if (largest_compaction_key != nullptr) {
largest = largest_compaction_key;
}
s = range_del_agg->AddTombstones(std::move(range_del_iter), smallest,
largest);
Recommit "Avoid adding tombstones of the same file to RangeDelAggregator multiple times" Summary: The origin commit #3635 will hurt performance for users who aren't using range deletions, because unneeded std::set operations, so it was reverted by commit 44653c7b7aabe821e671946e732dda7ae6b43d1b. (see #3672) To fix this, move the set to and add a check in , i.e., file will be added only if is non-nullptr. The db_bench command which find the performance regression: > ./db_bench --benchmarks=fillrandom,seekrandomwhilewriting --threads=1 --num=1000000 --reads=150000 --key_size=66 > --value_size=1262 --statistics=0 --compression_ratio=0.5 --histogram=1 --seek_nexts=1 --stats_per_interval=1 > --stats_interval_seconds=600 --max_background_flushes=4 --num_multi_db=1 --max_background_compactions=16 --seed=1522388277 > -write_buffer_size=1048576 --level0_file_num_compaction_trigger=10000 --compression_type=none Before and after the modification, I re-run this command on the machine, the results of are as follows: **fillrandom** Table | P50 | P75 | P99 | P99.9 | P99.99 | ---- | --- | --- | --- | ----- | ------ | before commit | 5.92 | 8.57 | 19.63 | 980.97 | 12196.00 | after commit | 5.91 | 8.55 | 19.34 | 965.56 | 13513.56 | **seekrandomwhilewriting** Table | P50 | P75 | P99 | P99.9 | P99.99 | ---- | --- | --- | --- | ----- | ------ | before commit | 1418.62 | 1867.01 | 3823.28 | 4980.99 | 9240.00 | after commit | 1450.54 | 1880.61 | 3962.87 | 5429.60 | 7542.86 | Closes https://github.com/facebook/rocksdb/pull/3800 Differential Revision: D7874245 Pulled By: ajkr fbshipit-source-id: 2e8bec781b3f7399246babd66395c88619534a17
7 years ago
}
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
}
}
if (handle != nullptr) {
ReleaseHandle(handle);
}
if (!s.ok()) {
assert(result == nullptr);
result = NewErrorInternalIterator<Slice>(s, arena);
Compaction Support for Range Deletion Summary: This diff introduces RangeDelAggregator, which takes ownership of iterators provided to it via AddTombstones(). The tombstones are organized in a two-level map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data copy by holding Slices returned by the iterator, which remain valid thanks to pinning. For compaction, we create a hierarchical range tombstone iterator with structure matching the iterator over compaction input data. An aggregator based on that iterator is used by CompactionIterator to determine which keys are covered by range tombstones. In case of merge operand, the same aggregator is used by MergeHelper. Upon finishing each file in the compaction, relevant range tombstones are added to the output file's range tombstone metablock and file boundaries are updated accordingly. To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete() considers tombstones in the key's snapshot stripe. When this function is used outside of compaction, it also checks newer stripes, which can contain covering tombstones. Currently the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges within a stripe such that binary search can be used. RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range to a new table's range tombstone meta-block. Since range tombstones may fall in the gap between files, we may need to extend some files' key-ranges. The strategy is (1) first file extends as far left as possible and other files do not extend left, (2) all files extend right until either the start of the next file or the end of the last range tombstone in the gap, whichever comes first. One other notable change is adding release/move semantics to ScopedArenaIterator such that it can be used to transfer ownership of an arena-allocated iterator, similar to how unique_ptr is used for malloc'd data. Depends on D61473 Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927 Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark Reviewed By: lightmark Subscribers: andrewkr, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D62205
8 years ago
}
return result;
}
Status TableCache::Get(const ReadOptions& options,
const InternalKeyComparator& internal_comparator,
const FileMetaData& file_meta, const Slice& k,
GetContext* get_context,
const SliceTransform* prefix_extractor,
HistogramImpl* file_read_hist, bool skip_filters,
int level) {
auto& fd = file_meta.fd;
std::string* row_cache_entry = nullptr;
bool done = false;
#ifndef ROCKSDB_LITE
IterKey row_cache_key;
std::string row_cache_entry_buffer;
// Check row cache if enabled. Since row cache does not currently store
// sequence numbers, we cannot use it if we need to fetch the sequence.
if (ioptions_.row_cache && !get_context->NeedToReadSequence()) {
uint64_t fd_number = fd.GetNumber();
auto user_key = ExtractUserKey(k);
// We use the user key as cache key instead of the internal key,
// otherwise the whole cache would be invalidated every time the
// sequence key increases. However, to support caching snapshot
// reads, we append the sequence number (incremented by 1 to
// distinguish from 0) only in this case.
uint64_t seq_no =
options.snapshot == nullptr ? 0 : 1 + GetInternalKeySeqno(k);
// Compute row cache key.
row_cache_key.TrimAppend(row_cache_key.Size(), row_cache_id_.data(),
row_cache_id_.size());
AppendVarint64(&row_cache_key, fd_number);
AppendVarint64(&row_cache_key, seq_no);
row_cache_key.TrimAppend(row_cache_key.Size(), user_key.data(),
user_key.size());
if (auto row_handle =
ioptions_.row_cache->Lookup(row_cache_key.GetUserKey())) {
// Cleanable routine to release the cache entry
Cleanable value_pinner;
auto release_cache_entry_func = [](void* cache_to_clean,
void* cache_handle) {
((Cache*)cache_to_clean)->Release((Cache::Handle*)cache_handle);
};
auto found_row_cache_entry = static_cast<const std::string*>(
ioptions_.row_cache->Value(row_handle));
// If it comes here value is located on the cache.
// found_row_cache_entry points to the value on cache,
// and value_pinner has cleanup procedure for the cached entry.
// After replayGetContextLog() returns, get_context.pinnable_slice_
// will point to cache entry buffer (or a copy based on that) and
// cleanup routine under value_pinner will be delegated to
// get_context.pinnable_slice_. Cache entry is released when
// get_context.pinnable_slice_ is reset.
value_pinner.RegisterCleanup(release_cache_entry_func,
ioptions_.row_cache.get(), row_handle);
replayGetContextLog(*found_row_cache_entry, user_key, get_context,
&value_pinner);
RecordTick(ioptions_.statistics, ROW_CACHE_HIT);
done = true;
} else {
// Not found, setting up the replay log.
RecordTick(ioptions_.statistics, ROW_CACHE_MISS);
row_cache_entry = &row_cache_entry_buffer;
}
}
#endif // ROCKSDB_LITE
Status s;
TableReader* t = fd.table_reader;
Cache::Handle* handle = nullptr;
if (!done && s.ok()) {
if (t == nullptr) {
s = FindTable(
env_options_, internal_comparator, fd, &handle, prefix_extractor,
options.read_tier == kBlockCacheTier /* no_io */,
true /* record_read_stats */, file_read_hist, skip_filters, level);
if (s.ok()) {
t = GetTableReaderFromHandle(handle);
}
}
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 =
get_context->max_covering_tombstone_seq();
if (s.ok() && max_covering_tombstone_seq != nullptr &&
!options.ignore_range_deletions) {
std::unique_ptr<InternalIterator> range_del_iter(
t->NewRangeTombstoneIterator(options));
Cache fragmented range tombstones in BlockBasedTableReader (#4493) Summary: This allows tombstone fragmenting to only be performed when the table is opened, and cached for subsequent accesses. On the same DB used in #4449, running `readrandom` results in the following: ``` readrandom : 0.983 micros/op 1017076 ops/sec; 78.3 MB/s (63103 of 100000 found) ``` Now that Get performance in the presence of range tombstones is reasonable, I also compared the performance between a DB with range tombstones, "expanded" range tombstones (several point tombstones that cover the same keys the equivalent range tombstone would cover, a common workaround for DeleteRange), and no range tombstones. The created DBs had 5 million keys each, and DeleteRange was called at regular intervals (depending on the total number of range tombstones being written) after 4.5 million Puts. The table below summarizes the results of a `readwhilewriting` benchmark (in order to provide somewhat more realistic results): ``` Tombstones? | avg micros/op | stddev micros/op | avg ops/s | stddev ops/s ----------------- | ------------- | ---------------- | ------------ | ------------ None | 0.6186 | 0.04637 | 1,625,252.90 | 124,679.41 500 Expanded | 0.6019 | 0.03628 | 1,666,670.40 | 101,142.65 500 Unexpanded | 0.6435 | 0.03994 | 1,559,979.40 | 104,090.52 1k Expanded | 0.6034 | 0.04349 | 1,665,128.10 | 125,144.57 1k Unexpanded | 0.6261 | 0.03093 | 1,600,457.50 | 79,024.94 5k Expanded | 0.6163 | 0.05926 | 1,636,668.80 | 154,888.85 5k Unexpanded | 0.6402 | 0.04002 | 1,567,804.70 | 100,965.55 10k Expanded | 0.6036 | 0.05105 | 1,667,237.70 | 142,830.36 10k Unexpanded | 0.6128 | 0.02598 | 1,634,633.40 | 72,161.82 25k Expanded | 0.6198 | 0.04542 | 1,620,980.50 | 116,662.93 25k Unexpanded | 0.5478 | 0.0362 | 1,833,059.10 | 121,233.81 50k Expanded | 0.5104 | 0.04347 | 1,973,107.90 | 184,073.49 50k Unexpanded | 0.4528 | 0.03387 | 2,219,034.50 | 170,984.32 ``` After a large enough quantity of range tombstones are written, range tombstone Gets can become faster than reading from an equivalent DB with several point tombstones. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4493 Differential Revision: D10842844 Pulled By: abhimadan fbshipit-source-id: a7d44534f8120e6aabb65779d26c6b9df954c509
6 years ago
*max_covering_tombstone_seq =
std::max(*max_covering_tombstone_seq,
MaxCoveringTombstoneSeqnum(
static_cast<FragmentedRangeTombstoneIterator*>(
range_del_iter.get()),
k, internal_comparator.user_comparator()));
}
if (s.ok()) {
get_context->SetReplayLog(row_cache_entry); // nullptr if no cache.
s = t->Get(options, k, get_context, prefix_extractor, skip_filters);
get_context->SetReplayLog(nullptr);
} else if (options.read_tier == kBlockCacheTier && s.IsIncomplete()) {
// Couldn't find Table in cache but treat as kFound if no_io set
get_context->MarkKeyMayExist();
s = Status::OK();
done = true;
}
}
#ifndef ROCKSDB_LITE
// Put the replay log in row cache only if something was found.
if (!done && s.ok() && row_cache_entry && !row_cache_entry->empty()) {
size_t charge =
row_cache_key.Size() + row_cache_entry->size() + sizeof(std::string);
void* row_ptr = new std::string(std::move(*row_cache_entry));
ioptions_.row_cache->Insert(row_cache_key.GetUserKey(), row_ptr, charge,
&DeleteEntry<std::string>);
}
#endif // ROCKSDB_LITE
if (handle != nullptr) {
ReleaseHandle(handle);
}
return s;
}
Status TableCache::GetTableProperties(
const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
std::shared_ptr<const TableProperties>* properties,
const SliceTransform* prefix_extractor, bool no_io) {
Status s;
auto table_reader = fd.table_reader;
// table already been pre-loaded?
if (table_reader) {
*properties = table_reader->GetTableProperties();
return s;
}
Cache::Handle* table_handle = nullptr;
s = FindTable(env_options, internal_comparator, fd, &table_handle,
prefix_extractor, no_io);
if (!s.ok()) {
return s;
}
assert(table_handle);
auto table = GetTableReaderFromHandle(table_handle);
*properties = table->GetTableProperties();
ReleaseHandle(table_handle);
return s;
}
size_t TableCache::GetMemoryUsageByTableReader(
const EnvOptions& env_options,
const InternalKeyComparator& internal_comparator, const FileDescriptor& fd,
const SliceTransform* prefix_extractor) {
Status s;
auto table_reader = fd.table_reader;
// table already been pre-loaded?
if (table_reader) {
return table_reader->ApproximateMemoryUsage();
}
Cache::Handle* table_handle = nullptr;
s = FindTable(env_options, internal_comparator, fd, &table_handle,
prefix_extractor, true);
if (!s.ok()) {
return 0;
}
assert(table_handle);
auto table = GetTableReaderFromHandle(table_handle);
auto ret = table->ApproximateMemoryUsage();
ReleaseHandle(table_handle);
return ret;
}
void TableCache::Evict(Cache* cache, uint64_t file_number) {
cache->Erase(GetSliceForFileNumber(&file_number));
}
} // namespace rocksdb