Summary:
Apply the formatting changes suggested by `clang-format`, except
where they would ruin the ASCII art in `blob_log_format.h`.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10856
Test Plan: `make check`
Reviewed By: siying
Differential Revision: D40652224
Pulled By: ltamasi
fbshipit-source-id: 8c1f5757b758474ea3e8102a7c5a1cf9e6dc1402
Summary:
Allow the last level only compaction able to output result to penultimate level if the penultimate level is empty. Which will also block the other compaction output to the penultimate level.
(it includes the PR https://github.com/facebook/rocksdb/issues/10829)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10822
Reviewed By: siying
Differential Revision: D40389180
Pulled By: jay-zhuang
fbshipit-source-id: 4e5dcdce307795b5e07b5dd1fa29dd75bb093bad
Summary:
Right now UserComparatorWrapper is a Customizable object, although it is not, which introduces some intialization overhead for the object. In some benchmarks, it shows up in CPU profiling. Make it not configurable by defining most functions needed by UserComparatorWrapper to an interface and implement the interface.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10837
Test Plan: Make sure existing tests pass
Reviewed By: pdillinger
Differential Revision: D40528511
fbshipit-source-id: 70eaac89ecd55401a26e8ed32abbc413a9617c62
Summary:
Refactor the classes, APIs and data structures for block cache tracing to allow a user provided trace writer to be used. Currently, only a TraceWriter is supported, with a default built-in implementation of FileTraceWriter. The TraceWriter, however, takes a flat trace record and is thus only suitable for file tracing. This PR introduces an abstract BlockCacheTraceWriter class that takes a structured BlockCacheTraceRecord. The BlockCacheTraceWriter implementation can then format and log the record in whatever way it sees fit. The default BlockCacheTraceWriterImpl does file tracing using a user provided TraceWriter.
`DB::StartBlockTrace` will internally redirect to changed `BlockCacheTrace::StartBlockCacheTrace`.
New API `DB::StartBlockTrace` is also added that directly takes `BlockCacheTraceWriter` pointer.
This same philosophy can be applied to KV and IO tracing as well.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10811
Test Plan:
existing unit tests
Old API DB::StartBlockTrace checked with db_bench tool
create database
```
./db_bench --benchmarks="fillseq" \
--key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 \
--cache_index_and_filter_blocks --cache_size=1048576 \
--disable_auto_compactions=1 --disable_wal=1 --compression_type=none \
--min_level_to_compress=-1 --compression_ratio=1 --num=10000000
```
To trace block cache accesses when running readrandom benchmark:
```
./db_bench --benchmarks="readrandom" --use_existing_db --duration=60 \
--key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 \
--cache_index_and_filter_blocks --cache_size=1048576 \
--disable_auto_compactions=1 --disable_wal=1 --compression_type=none \
--min_level_to_compress=-1 --compression_ratio=1 --num=10000000 \
--threads=16 \
-block_cache_trace_file="/tmp/binary_trace_test_example" \
-block_cache_trace_max_trace_file_size_in_bytes=1073741824 \
-block_cache_trace_sampling_frequency=1
```
Reviewed By: anand1976
Differential Revision: D40435289
Pulled By: akankshamahajan15
fbshipit-source-id: fa2755f4788185e19f4605e731641cfd21ab3282
Summary:
Currently, the code in `SaveValue` that handles `kTypeValue` and
`kTypeBlobIndex` (and more recently, `kTypeWideColumnEntity`) is
mostly shared. This made sense originally; however, by now the
handling of these three value types has diverged significantly. The
patch makes the logic cleaner and also eliminates quite a bit of branching
by giving each value type its own `case` and removing a fall-through.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10840
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D40568420
Pulled By: ltamasi
fbshipit-source-id: 2e614606afd1c3d9c76d9b5f1efa0959fc174103
Summary:
When the `preclude_last_level_data_seconds` or
`preserve_internal_time_seconds` is smaller than 100 (seconds), no seqno->time information was recorded.
Also make sure all data will be compacted to the last level even if there's no write to record the time information.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10829
Test Plan: added unittest
Reviewed By: siying
Differential Revision: D40443934
Pulled By: jay-zhuang
fbshipit-source-id: 2ecf1361daf9f3e5c3385aee6dc924fa59e2813a
Summary:
The patch makes it possible to provide the value of the default column
separately when calling `WideColumnSerialization::Serialize`. This eliminates
the need to construct a new `WideColumns` vector in certain cases
(for example, it will come in handy when implementing `Merge`).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10839
Test Plan: `make check`
Reviewed By: riversand963
Differential Revision: D40561448
Pulled By: ltamasi
fbshipit-source-id: 69becdd510e6a83ab1feb956c12772110e1040d6
Summary:
This new property allows users to trigger the background block cache stats collection mode through the `GetProperty()` and `GetMapProperty()` APIs. The background mode has much lower overhead at the expense of returning stale values in more cases.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10832
Test Plan: updated unit test
Reviewed By: pdillinger
Differential Revision: D40497883
Pulled By: ajkr
fbshipit-source-id: bdcc93402f426463abb2153756aad9e295447343
Summary:
FIFO compaction can theoretically open a DB with any compaction style.
However, the current code only allows FIFO compaction to open a DB with
a single level.
This PR relaxes the limitation of FIFO compaction and allows it to open a
DB with multiple levels. Below is the read / write / compaction behavior:
* The read behavior is untouched, and it works like a regular rocksdb instance.
* The write behavior is untouched as well. When a FIFO compacted DB
is opened with multiple levels, all new files will still be in level 0, and no files
will be moved to a different level.
* Compaction logic is extended. It will first identify the bottom-most non-empty level.
Then, it will delete the oldest file in that level.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10348
Test Plan:
Added a new test to verify the migration from level to FIFO where the db has multiple levels.
Extended existing test cases in db_test and db_basic_test to also verify
all entries of a key after reopening the DB with FIFO compaction.
Reviewed By: jay-zhuang
Differential Revision: D40233744
fbshipit-source-id: 6cc011d6c3467e6bfb9b6a4054b87619e69815e1
Summary:
Instead of existing calls to ps from gnu_parallel, call a new wrapper that does ps, looks for unit test like processes, and uses pstack or gdb to print thread stack traces. Also, using `ps -wwf` instead of `ps -wf` ensures output is not cut off.
For security, CircleCI runs with security restrictions on ptrace (/proc/sys/kernel/yama/ptrace_scope = 1), and this change adds a work-around to `InstallStackTraceHandler()` (only used by testing tools) to allow any process from the same user to debug it. (I've also touched >100 files to ensure all the unit tests call this function.)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10828
Test Plan: local manual + temporary infinite loop in a unit test to observe in CircleCI
Reviewed By: hx235
Differential Revision: D40447634
Pulled By: pdillinger
fbshipit-source-id: 718a4c4a5b54fa0f9af2d01a446162b45e5e84e1
Summary:
The motivation for this change is a planned feature (related to HyperClockCache) that will depend on a large array that can essentially grow automatically, up to some bound, without the pointer address changing and with guaranteed zero-initialization of the data. Anonymous mmaps provide such functionality, and this change provides an internal API for that.
The other existing use of anonymous mmap in RocksDB is for allocating in huge pages. That code and other related Arena code used some awkward non-RAII and pre-C++11 idioms, so I cleaned up much of that as well, with RAII, move semantics, constexpr, etc.
More specifcs:
* Minimize conditional compilation
* Add Windows support for anonymous mmaps
* Use std::deque instead of std::vector for more efficient bag
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10810
Test Plan: unit test added for new functionality
Reviewed By: riversand963
Differential Revision: D40347204
Pulled By: pdillinger
fbshipit-source-id: ca83fcc47e50fabf7595069380edd2954f4f879c
Summary:
This reverts https://github.com/facebook/rocksdb/issues/10792 and uses a different strategy to stabilize the test: remove the unnecessary randomness by providing a constant seed for shuffling keys.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10816
Test Plan: `gtest-parallel ./db_bloom_filter_test -r1000 --gtest_filter=*ForHits*`
Reviewed By: jay-zhuang
Differential Revision: D40347957
Pulled By: pdillinger
fbshipit-source-id: a270e157485cbd94ed03b80cdd21b954ebd57d57
Summary:
Lock the penultimate level for the whole compaction inputs range, so any
key in that compaction is safe to move up from the last level to
penultimate level.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10782
Reviewed By: siying
Differential Revision: D40231540
Pulled By: siying
fbshipit-source-id: ca115cc8b4018b35d797329fa85a19b06cc8c13e
Summary:
This change is motivated by ensuring that `ldb update_manifest` or `UpdateManifestForFilesState` can run without expecting files to open when the old temperature is provided (in case the FileSystem strictly interprets non-kUnknown), but ended up fixing a problem in `OfflineManifestWriter` (used by `ldb unsafe_remove_sst_file`) where it would open some SST files during recovery and expect them to match the prior manifest state, even if not required by the intended new state.
Also update BackupEngine to retry with Temperature kUnknown when reading file with potentially "wrong" temperature.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10796
Test Plan: tests added/updated, that fail before the change(s) and now pass
Reviewed By: jay-zhuang
Differential Revision: D40232645
Pulled By: jay-zhuang
fbshipit-source-id: b5aa2688aecfe0c320b80a7da689b315414c20be
Summary:
Provide support for async_io if ReadOptions.tailing is set true.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10781
Test Plan:
- Update unit tests
- Ran db_bench: ./db_bench --benchmarks="readrandom" --use_existing_db --use_tailing_iterator=1 --async_io=1
Reviewed By: anand1976
Differential Revision: D40128882
Pulled By: anand1976
fbshipit-source-id: 55e17855536871a5c47e2de92d238ae005c32d01
Summary:
The test may fail because the L5 files may only cover small portion of the whole key range.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10792
Test Plan:
```
gtest-parallel ./db_bloom_filter_test --gtest_filter=DBBloomFilterTest.OptimizeFiltersForHits -r 1000 -w 100
```
Reviewed By: siying
Differential Revision: D40217600
Pulled By: siying
fbshipit-source-id: 18db549184bccf5e513eaa7e31ab17385b71ef71
Summary:
Add option `preserve_internal_time_seconds` to preserve the internal
time information.
It's mostly for the migration of the existing data to tiered storage (
`preclude_last_level_data_seconds`). When the tiering feature is just
enabled, the existing data won't have the time information to decide if
it's hot or cold. Enabling this feature will start collect and preserve
the time information for the new data.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10747
Reviewed By: siying
Differential Revision: D39910141
Pulled By: siying
fbshipit-source-id: 25c21638e37b1a7c44006f636b7d714fe7242138
Summary:
When checking if a range [start, end) overlaps with a compaction whose range is [start1, end1), always exclude timestamp from start, end, start1 and end1, otherwise some versions of one user key may be compacted to bottommost layer while others remain in the original level.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10787
Test Plan: make check
Reviewed By: ltamasi
Differential Revision: D40187672
Pulled By: ltamasi
fbshipit-source-id: 81226267fd3e33ffa79665c62abadf2ebec45496
Summary:
To make it consistent with the compaction picker which uses the `sstableKeyCompare()` to pick the overlap files. For example, without this change, it may cut L1 files like:
```
L1: [2-21] [22-30]
L2: [1-10] [21-30]
```
Because "21" on L1 is smaller than "21" on L2. But for compaction, these 2 files are overlapped.
`sstableKeyCompare()` also take range delete into consideration which may cut file for the same key.
It also makes the `max_compaction_bytes` calculation more accurate for cases like above, the overlapped bytes was under estimated. Also make sure the 2 keys won't be splitted to 2 files because of reaching `max_compaction_bytes`.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10763
Reviewed By: cbi42
Differential Revision: D39971904
Pulled By: cbi42
fbshipit-source-id: bcc309e9c3dc61a8f50667a6f633e6132c0154a8
Summary:
As the first step of covering the wide-column functionality of iterators
in our stress tests, the patch adds verification logic to
`NonBatchedOpsStressTest::VerifyDb` that checks whether the
iterator's value and columns are in sync. Note: I plan to update the other
types of stress tests and add similar verification for prefix scans etc.
in separate PRs.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10783
Test Plan: Ran some simple blackbox crash tests.
Reviewed By: riversand963
Differential Revision: D40152370
Pulled By: riversand963
fbshipit-source-id: 8f9d17d7af5da58ccf1bd2057cab53cc9645ac35
Summary:
With current implementation, within the same RocksDB instance, all column families with non-empty memtables will be scheduled for flush if RocksDB determines that any column family needs to be flushed, e.g. memtable full, write buffer manager, etc., if atomic flush is enabled. Not doing so can lead to data loss and inconsistency when WAL is disabled, which is a common setting when atomic flush is enabled. Therefore, setting a per-column-family knob, min_write_buffer_number_to_merge to a value greater than 1 is not compatible with atomic flush, and should be sanitized during column family creation and db open.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10773
Test Plan:
Reproduce: D39993203 has detailed steps.
Run the test with and without the fix.
Reviewed By: cbi42
Differential Revision: D40077955
Pulled By: cbi42
fbshipit-source-id: 451a9179eb531ac42eaccf40b451b9dec4085240
Summary:
fix for https://github.com/facebook/rocksdb/issues/10752 where RocksDB could be in an infinite compaction loop (with compaction reason kBottommostFiles) if allow_ingest_behind is enabled and the bottommost level is unfilled.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10767
Test Plan: Added a unit test to reproduce the compaction loop.
Reviewed By: ajkr
Differential Revision: D40031861
Pulled By: ajkr
fbshipit-source-id: 71c4b02931fbe507a847632905404c9b8fa8c96b
Summary:
Fix a bug in Iterator::Refresh() where the local SV it obtained could be obsolete upon return, and should be cleaned up.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10770
Test Plan: added a unit test to reproduce the issue.
Reviewed By: ajkr
Differential Revision: D40063809
Pulled By: ajkr
fbshipit-source-id: 619e728eb0f1ac9540b4d0ad38e43acc37a514b2
Summary:
When `FlushOptions::wait` is set to false, manual flush should not stall forever.
If the database has already stopped writes, then the thread calling `DB::Flush()` with
`FlushOptions::wait=false` should not enter the `DBImpl::write_thread_`.
To prevent this, we should do a check at the beginning and return `TryAgain()`
Resolves: https://github.com/facebook/rocksdb/issues/9892
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10001
Reviewed By: siying
Differential Revision: D36422303
Pulled By: siying
fbshipit-source-id: 723bd3065e8edc4f17c82449d0d6b95a2381ac0a
Summary:
For RoundRobin compaction, the data should be mostly sorted per level and within level. Use normal compaction picker for RR until all expired data is compacted.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10725
Reviewed By: ajkr
Differential Revision: D39771069
Pulled By: jay-zhuang
fbshipit-source-id: 7ccf88d7c093fad5673bda73a7b08cc4757780cd
Summary:
This is intended as a step toward possibly separating secondary cache integration from the
Cache implementation as much as possible, to (hopefully) minimize code duplication in
adding secondary cache support to HyperClockCache.
* Major clarifications to API docs of secondary cache compatible parts of Cache. For example, previously the docs seemed to suggest that Wait() was not needed if IsReady()==true. And it wasn't clear what operations were actually supported on pending handles.
* Add some assertions related to these requirements, such as that we don't Release() before Wait() (which would leak a secondary cache handle).
* Fix a leaky abstraction with dummy handles, which are supposed to be internal to the Cache. Previously, these just used value=nullptr to indicate dummy handle, which meant that they could be confused with legitimate value=nullptr cases like cache reservations. Also fixed blob_source_test which was relying on this leaky abstraction.
* Drop "incomplete" terminology, which was another name for "pending".
* Split handle flags into "mutable" ones requiring mutex and "immutable" ones which do not. Because of single-threaded access to pending handles, the "Is Pending" flag can be in the "immutable" set. This allows removal of a TSAN work-around and removing a mutex acquire-release in IsReady().
* Remove some unnecessary handling of charges on handles of failed lookups. Keeping total_charge=0 means no special handling needed. (Removed one unnecessary mutex acquire/release.)
* Simplify handling of dummy handle in Lookup(). There is no need to explicitly Ref & Release w/Erase if we generally overwrite the dummy anyway. (Removed one mutex acquire/release, a call to Release().)
Intended follow-up:
* Clarify APIs in secondary_cache.h
* Doesn't SecondaryCacheResultHandle transfer ownership of the Value() on success (implementations should not release the value in destructor)?
* Does Wait() need to be called if IsReady() == true? (This would be different from Cache.)
* Do Value() and Size() have undefined behavior if IsReady() == false?
* Why have a custom API for what is essentially a std::future<std::pair<void*, size_t>>?
* Improve unit testing of standalone handle case
* Apparent null `e` bug in `free_standalone_handle` case
* Clean up secondary cache testing in lru_cache_test
* Why does TestSecondaryCacheResultHandle hold on to a Cache::Handle?
* Why does TestSecondaryCacheResultHandle::Wait() do nothing? Shouldn't it establish the post-condition IsReady() == true?
* (Assuming that is sorted out...) Shouldn't TestSecondaryCache::WaitAll simply wait on each handle in order (no casting required)? How about making that the default implementation?
* Why does TestSecondaryCacheResultHandle::Size() check Value() first? If the API is intended to be returning 0 before IsReady(), then that is weird but should at least be documented. Otherwise, if it's intended to be undefined behavior, we should assert IsReady().
* Consider replacing "standalone" and "dummy" entries with a single kind of "weak" entry that deletes its value when it reaches zero refs. Suppose you are using compressed secondary cache and have two iterators at similar places. It will probably common for one iterator to have standalone results pinned (out of cache) when the second iterator needs those same blocks and has to re-load them from secondary cache and duplicate the memory. Combining the dummy and the standalone should fix this.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10730
Test Plan:
existing tests (minor update), and crash test with sanitizers and secondary cache
Performance test for any regressions in LRUCache (primary only):
Create DB with
```
TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=30000000 -disable_wal=1 -bloom_bits=16
```
Test before & after (run at same time) with
```
TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=readrandom[-X100] -readonly -num=30000000 -bloom_bits=16 -cache_index_and_filter_blocks=1 -cache_size=233000000 -duration 30 -threads=16
```
Before: readrandom [AVG 100 runs] : 22234 (± 63) ops/sec; 1.6 (± 0.0) MB/sec
After: readrandom [AVG 100 runs] : 22197 (± 64) ops/sec; 1.6 (± 0.0) MB/sec
That's within 0.2%, which is not significant by the confidence intervals.
Reviewed By: anand1976
Differential Revision: D39826010
Pulled By: anand1976
fbshipit-source-id: 3202b4a91f673231c97648ae070e502ae16b0f44
Summary:
Add new property "do_not_recurse" in IOOptions for underlying file system to skip iteration of directories during DB::Open if there are no sub directories and list only files.
By default this property is set to false. This property is set true currently in the code where RocksDB is sure only files are needed during DB::Open.
Provided support in PosixFileSystem to use "do_not_recurse".
TestPlan:
- Existing tests
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10668
Reviewed By: anand1976
Differential Revision: D39471683
Pulled By: akankshamahajan15
fbshipit-source-id: 90e32f0b86d5346d53bc2714d3a0e7002590527f
Summary:
Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are
- internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps.
- Garbage collection of range tombstones and range tombstone covered keys during compaction.
- Get()/MultiGet() to return the timestamp of a range tombstone when needed.
- Get/Iterator with range tombstones bounded by readoptions.timestamp.
- timestamp crash test now issues DeleteRange by default.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10661
Test Plan:
- Added unit test: `make check`
- Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4`
- Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`. Did not see consistent regression in no timestamp case.
| micros/op | fillrandom | seekrandom |
| --- | --- | --- |
|main| 2.58 |10.96|
|PR 10661| 2.68 |10.63|
Reviewed By: riversand963
Differential Revision: D39441192
Pulled By: cbi42
fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2
Summary:
**Context/Summary:**
Introduce `manual_wal_flush_one_in` as titled.
- When `manual_wal_flush_one_in > 0`, we also need tracing to correctly verify recovery because WAL data can be lost in this case when `FlushWAL()` is not explicitly called by users of RocksDB (in our case, db stress) and the recovery from such potential WAL data loss is a prefix recovery that requires tracing to verify. As another consequence, we need to disable features can't run under unsync data loss with `manual_wal_flush_one_in`
Incompatibilities fixed along the way:
```
db_stress: db/db_impl/db_impl_open.cc:2063: static rocksdb::Status rocksdb::DBImpl::Open(const rocksdb::DBOptions&, const string&, const std::vector<rocksdb::ColumnFamilyDescriptor>&, std::vector<rocksdb::ColumnFamilyHandle*>*, rocksdb::DB**, bool, bool): Assertion `impl->TEST_WALBufferIsEmpty()' failed.
```
- It turns out that `Writer::AddCompressionTypeRecord` before this assertion `EmitPhysicalRecord(kSetCompressionType, encode.data(), encode.size());` but do not trigger flush if `manual_wal_flush` is set . This leads to `impl->TEST_WALBufferIsEmpty()' is false.
- As suggested, assertion is removed and violation case is handled by `FlushWAL(sync=true)` along with refactoring `TEST_WALBufferIsEmpty()` to be `WALBufferIsEmpty()` since it is used in prod code now.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10698
Test Plan:
- Locally running `python3 tools/db_crashtest.py blackbox --manual_wal_flush_one_in=1 --manual_wal_flush=1 --sync_wal_one_in=100 --atomic_flush=1 --flush_one_in=100 --column_families=3`
- Joined https://github.com/facebook/rocksdb/pull/10624 in auto CI testings with all RocksDB stress/crash test jobs
Reviewed By: ajkr
Differential Revision: D39593752
Pulled By: ajkr
fbshipit-source-id: 3a2135bb792c52d2ffa60257d4fbc557fb04d2ce
Summary:
currently, there are places in compaction_picker where we add up `compensated_file_size` of files being compacted and limit the sum to be under `max_compaction_bytes`. `compensated_file_size` contains booster for point tombstones and should be used only for determining file's compaction priority. This PR replaces `compensated_file_size` with actual file size in such places.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10728
Test Plan: CI
Reviewed By: ajkr
Differential Revision: D39789427
Pulled By: cbi42
fbshipit-source-id: 1f89fb6c0159c53bf01d8dc783f465959f442c81
Summary:
Try to align the compaction output file boundaries to the next level ones
(grandparent level), to reduce the level compaction write-amplification.
In level compaction, there are "wasted" data at the beginning and end of the
output level files. Align the file boundary can avoid such "wasted" compaction.
With this PR, it tries to align the non-bottommost level file boundaries to its
next level ones. It may cut file when the file size is large enough (at least
50% of target_file_size) and not too large (2x target_file_size).
db_bench shows about 12.56% compaction reduction:
```
TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432
# baseline:
Flush(GB): cumulative 25.882, interval 7.216
Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds
# with this change:
Flush(GB): cumulative 25.882, interval 7.753
Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds
```
The compaction simulator shows a similar result (14% with 100G random data).
As a side effect, with this PR, the SST file size can exceed the
target_file_size, but is capped at 2x target_file_size. And there will be
smaller files. Here are file size statistics when loading 100GB with the target
file size 32MB:
```
baseline this_PR
count 1.656000e+03 1.705000e+03
mean 3.116062e+07 3.028076e+07
std 7.145242e+06 8.046139e+06
```
The feature is enabled by default, to revert to the old behavior disable it
with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false`
Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for
use case like user adding 2 or more non-overlapping data range at the same
time, it can reduce the overlapping of 2 datasets in the lower levels.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655
Reviewed By: cbi42
Differential Revision: D39552321
Pulled By: jay-zhuang
fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
Summary:
when a new internal iterator is constructed during iterator refresh, pointer to the previous memtable range tombstone iterator was not cleared. This could cause segfault for future `Refresh()` calls when they try to free the memtable range tombstones. This PR fixes this issue.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10739
Test Plan: added a unit test in db_range_del_test.cc to reproduce this issue.
Reviewed By: ajkr, riversand963
Differential Revision: D39825283
Pulled By: cbi42
fbshipit-source-id: 3b59a2b73865aed39e28cdd5c1b57eed7991b94c
Summary:
Currently, without this fix, DBImpl::GetLatestSequenceForKey() may not return the latest sequence number for merge operands of the key. This can cause conflict checking during optimistic transaction commit phase to fail. Fix it by always returning the latest sequence number of the key, also considering range tombstones.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10724
Test Plan: make check
Reviewed By: cbi42
Differential Revision: D39756847
Pulled By: riversand963
fbshipit-source-id: 0764c3dd4cb24960b37e18adccc6e7feed0e6876
Summary:
We have a lot of confusing code because of mixed, sometimes
completely opposite uses of of the term "raw block" or "raw contents",
sometimes within the same source file. For example, in `BlockBasedTableBuilder`,
`raw_block_contents` and `raw_size` generally referred to uncompressed block
contents and size, while `WriteRawBlock` referred to writing a block that
is already compressed if it is going to be. Meanwhile, in
`BlockBasedTable`, `raw_block_contents` either referred to a (maybe
compressed) block with trailer, or a maybe compressed block maybe
without trailer. (Note: left as follow-up work to use C++ typing to
better sort out the various kinds of BlockContents.)
This change primarily tries to apply some consistent terminology around
the kinds of block representations, avoiding the unclear "raw". (Any
meaning of "raw" assumes some bias toward the storage layer or toward
the logical data layer.) Preferred terminology:
* **Serialized block** - bytes that go into storage. For block-based table
(usually the case) this includes the block trailer. WART: block `size` may or
may not include the trailer; need to be clear about whether it does or not.
* **Maybe compressed block** - like a serialized block, but without the
trailer (or no promise of including a trailer). Must be accompanied by a
CompressionType.
* **Uncompressed block** - "payload" bytes that are either stored with no
compression, used as input to compression function, or result of
decompression function.
* **Parsed block** - an in-memory form of a block in block cache, as it is
used by the table reader. Different C++ types are used depending on the
block type (see block_like_traits.h).
Other refactorings:
* Misc corrections/improvements of internal API comments
* Remove a few misleading / unhelpful / redundant comments.
* Use move semantics in some places to simplify contracts
* Use better parameter names to indicate which parameters are used for
outputs
* Remove some extraneous `extern`
* Various clean-ups to `CacheDumperImpl` (mostly unnecessary code)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10408
Test Plan: existing tests
Reviewed By: akankshamahajan15
Differential Revision: D38172617
Pulled By: pdillinger
fbshipit-source-id: ccb99299f324ac5ca46996d34c5089621a4f260c
Summary:
Change the library order in PLATFORM_LDFLAGS to enable fbcode platform 10 build with folly. This PR also has a few fixes for platform 10 compiler errors.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10708
Test Plan:
ROCKSDB_FBCODE_BUILD_WITH_PLATFORM010=1 USE_COROUTINES=1 make -j64 check
ROCKSDB_FBCODE_BUILD_WITH_PLATFORM010=1 USE_FOLLY=1 make -j64 check
Reviewed By: ajkr
Differential Revision: D39666590
Pulled By: anand1976
fbshipit-source-id: 256a1127ef561399cd6299a6a392ca29bd68ca44
Summary:
when there is a single memtable without range tombstones and no SST files in the database, DBIter should wrap memtable iterator directly. Currently we create a merging iterator on top of the memtable iterator, and have DBIter wrap around it. This causes iterator regression and this PR fixes this issue.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10705
Test Plan:
- `make check`
- Performance:
- Set up: `./db_bench -benchmarks=filluniquerandom -write_buffer_size=$((1 << 30)) -num=10000`
- Benchmark: `./db_bench -benchmarks=seekrandom -use_existing_db=true -avoid_flush_during_recovery=true -write_buffer_size=$((1 << 30)) -num=10000 -threads=16 -duration=60 -seek_nexts=$seek_nexts`
```
seek_nexts main op/sec https://github.com/facebook/rocksdb/issues/10705 RocksDB v7.6
0 5746568 5749033 5786180
30 2411690 3006466 2837699
1000 102556 128902 124667
```
Reviewed By: ajkr
Differential Revision: D39644221
Pulled By: cbi42
fbshipit-source-id: 8063ff611ba31b0e5670041da3927c8c54b2097d
Summary:
The background compaction may still running while the test end, which would cause ASAN stack-use-after-scope error.
Explicitly close the DB before test end.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10697
Test Plan:
able to reproduce with:
```
gtest-parallel ./compaction_service_test --gtest_filter=CompactionServiceTest.BasicCompactions -r 10000 -w 100
```
Reviewed By: gitbw95
Differential Revision: D39590974
Pulled By: jay-zhuang
fbshipit-source-id: da264b2e6a276afbda7d5ff7adb9d7b8d4213d90
Summary:
Fix invalid reference in MultiGet due to resizing of the ```batches``` autovector.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10702
Test Plan: Run asan crash test
Reviewed By: riversand963
Differential Revision: D39608753
Pulled By: anand1976
fbshipit-source-id: 7a9e7fc6f436f08eb22003d0e6b0e1e4dcdc1a2a
Summary:
The assertion in ```FilePickerMultiGet::ReplaceRange()``` was incorrect. The function should only be called to replace the range after finishing the search in the current level, which is indicated by ```hit_file_ == nullptr``` i.e no more overlapping files in this level.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10695
Reviewed By: gitbw95
Differential Revision: D39583217
Pulled By: anand1976
fbshipit-source-id: d4cedfb2b62fb9f3a083e9848a403ae6342f0519
Summary:
This change establishes a distinctive name for the experimental new lock-free clock cache (originally developed by guidotag and revamped in PR https://github.com/facebook/rocksdb/issues/10626). A few reasons:
* We want to make it clear that this is a fundamentally different implementation vs. the old clock cache, to avoid people saying "I already tried clock cache."
* We want to highlight the key feature: it's fast (especially under parallel load)
* Because it requires an estimated charge per entry, it is not drop-in API compatible with old clock cache. This estimate might always be required for highest performance, and giving it a distinct name should reduce confusion about the distinct API requirements.
* We might develop a variant requiring the same estimate parameter but with LRU eviction. In that case, using the name HyperLRUCache should make things more clear. (FastLRUCache is just a prototype that might soon be removed.)
Some API detail:
* To reduce copy-pasting parameter lists, etc. as in LRUCache construction, I have a `MakeSharedCache()` function on `HyperClockCacheOptions` instead of `NewHyperClockCache()`.
* Changes -cache_type=clock_cache to -cache_type=hyper_clock_cache for applicable tools. I think this is more consistent / sustainable for reasons already stated.
For performance tests see https://github.com/facebook/rocksdb/pull/10626
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10684
Test Plan: no interesting functional changes; tests updated
Reviewed By: anand1976
Differential Revision: D39547800
Pulled By: pdillinger
fbshipit-source-id: 5c0fe1b5cf3cb680ab369b928c8569682b9795bf
Summary:
* Consolidates most metadata into a single word per slot so that more
can be accomplished with a single atomic update. In the common case,
Lookup was previously about 4 atomic updates, now just 1 atomic update.
Common case Release was previously 1 atomic read + 1 atomic update,
now just 1 atomic update.
* Eliminate spins / waits / yields, which likely threaten some "lock free"
benefits. Compare-exchange loops are only used in explicit Erase, and
strict_capacity_limit=true Insert. Eviction uses opportunistic compare-
exchange.
* Relaxes some aggressiveness and guarantees. For example,
* Duplicate Inserts will sometimes go undetected and the shadow duplicate
will age out with eviction.
* In many cases, the older Inserted value for a given cache key will be kept
(i.e. Insert does not support overwrite).
* Entries explicitly erased (rather than evicted) might not be freed
immediately in some rare cases.
* With strict_capacity_limit=false, capacity limit is not tracked/enforced as
precisely as LRUCache, but is self-correcting and should only deviate by a
very small number of extra or fewer entries.
* Use smaller "computed default" number of cache shards in many cases,
because benefits to larger usage tracking / eviction pools outweigh the small
cost of more lock-free atomic contention. The improvement in CPU and I/O
is dramatic in some limit-memory cases.
* Even without the sharding change, the eviction algorithm is likely more
effective than LRU overall because it's more stateful, even though the
"hot path" state tracking for it is essentially free with ref counting. It
is like a generalized CLOCK with aging (see code comments). I don't have
performance numbers showing a specific improvement, but in theory, for a
Poisson access pattern to each block, keeping some state allows better
estimation of time to next access (Poisson interval) than strict LRU. The
bounded randomness in CLOCK can also reduce "cliff" effect for repeated
range scans approaching and exceeding cache size.
## Hot path algorithm comparison
Rough descriptions, focusing on number and kind of atomic operations:
* Old `Lookup()` (2-5 atomic updates per probe):
```
Loop:
Increment internal ref count at slot
If possible hit:
Check flags atomic (and non-atomic fields)
If cache hit:
Three distinct updates to 'flags' atomic
Increment refs for internal-to-external
Return
Decrement internal ref count
while atomic read 'displacements' > 0
```
* New `Lookup()` (1-2 atomic updates per probe):
```
Loop:
Increment acquire counter in meta word (optimistic)
If visible entry (already read meta word):
If match (read non-atomic fields):
Return
Else:
Decrement acquire counter in meta word
Else if invisible entry (rare, already read meta word):
Decrement acquire counter in meta word
while atomic read 'displacements' > 0
```
* Old `Release()` (1 atomic update, conditional on atomic read, rarely more):
```
Read atomic ref count
If last reference and invisible (rare):
Use CAS etc. to remove
Return
Else:
Decrement ref count
```
* New `Release()` (1 unconditional atomic update, rarely more):
```
Increment release counter in meta word
If last reference and invisible (rare):
Use CAS etc. to remove
Return
```
## Performance test setup
Build DB with
```
TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=fillrandom -num=30000000 -disable_wal=1 -bloom_bits=16
```
Test with
```
TEST_TMPDIR=/dev/shm ./db_bench -benchmarks=readrandom -readonly -num=30000000 -bloom_bits=16 -cache_index_and_filter_blocks=1 -cache_size=${CACHE_MB}000000 -duration 60 -threads=$THREADS -statistics
```
Numbers on a single socket Skylake Xeon system with 48 hardware threads, DEBUG_LEVEL=0 PORTABLE=0. Very similar story on a dual socket system with 80 hardware threads. Using (every 2nd) Fibonacci MB cache sizes to sample the territory between powers of two. Configurations:
base: LRUCache before this change, but with db_bench change to default cache_numshardbits=-1 (instead of fixed at 6)
folly: LRUCache before this change, with folly enabled (distributed mutex) but on an old compiler (sorry)
gt_clock: experimental ClockCache before this change
new_clock: experimental ClockCache with this change
## Performance test results
First test "hot path" read performance, with block cache large enough for whole DB:
4181MB 1thread base -> kops/s: 47.761
4181MB 1thread folly -> kops/s: 45.877
4181MB 1thread gt_clock -> kops/s: 51.092
4181MB 1thread new_clock -> kops/s: 53.944
4181MB 16thread base -> kops/s: 284.567
4181MB 16thread folly -> kops/s: 249.015
4181MB 16thread gt_clock -> kops/s: 743.762
4181MB 16thread new_clock -> kops/s: 861.821
4181MB 24thread base -> kops/s: 303.415
4181MB 24thread folly -> kops/s: 266.548
4181MB 24thread gt_clock -> kops/s: 975.706
4181MB 24thread new_clock -> kops/s: 1205.64 (~= 24 * 53.944)
4181MB 32thread base -> kops/s: 311.251
4181MB 32thread folly -> kops/s: 274.952
4181MB 32thread gt_clock -> kops/s: 1045.98
4181MB 32thread new_clock -> kops/s: 1370.38
4181MB 48thread base -> kops/s: 310.504
4181MB 48thread folly -> kops/s: 268.322
4181MB 48thread gt_clock -> kops/s: 1195.65
4181MB 48thread new_clock -> kops/s: 1604.85 (~= 24 * 1.25 * 53.944)
4181MB 64thread base -> kops/s: 307.839
4181MB 64thread folly -> kops/s: 272.172
4181MB 64thread gt_clock -> kops/s: 1204.47
4181MB 64thread new_clock -> kops/s: 1615.37
4181MB 128thread base -> kops/s: 310.934
4181MB 128thread folly -> kops/s: 267.468
4181MB 128thread gt_clock -> kops/s: 1188.75
4181MB 128thread new_clock -> kops/s: 1595.46
Whether we have just one thread on a quiet system or an overload of threads, the new version wins every time in thousand-ops per second, sometimes dramatically so. Mutex-based implementation quickly becomes contention-limited. New clock cache shows essentially perfect scaling up to number of physical cores (24), and then each hyperthreaded core adding about 1/4 the throughput of an additional physical core (see 48 thread case). Block cache miss rates (omitted above) are negligible across the board. With partitioned instead of full filters, the maximum speed-up vs. base is more like 2.5x rather than 5x.
Now test a large block cache with low miss ratio, but some eviction is required:
1597MB 1thread base -> kops/s: 46.603 io_bytes/op: 1584.63 miss_ratio: 0.0201066 max_rss_mb: 1589.23
1597MB 1thread folly -> kops/s: 45.079 io_bytes/op: 1530.03 miss_ratio: 0.019872 max_rss_mb: 1550.43
1597MB 1thread gt_clock -> kops/s: 48.711 io_bytes/op: 1566.63 miss_ratio: 0.0198923 max_rss_mb: 1691.4
1597MB 1thread new_clock -> kops/s: 51.531 io_bytes/op: 1589.07 miss_ratio: 0.0201969 max_rss_mb: 1583.56
1597MB 32thread base -> kops/s: 301.174 io_bytes/op: 1439.52 miss_ratio: 0.0184218 max_rss_mb: 1656.59
1597MB 32thread folly -> kops/s: 273.09 io_bytes/op: 1375.12 miss_ratio: 0.0180002 max_rss_mb: 1586.8
1597MB 32thread gt_clock -> kops/s: 904.497 io_bytes/op: 1411.29 miss_ratio: 0.0179934 max_rss_mb: 1775.89
1597MB 32thread new_clock -> kops/s: 1182.59 io_bytes/op: 1440.77 miss_ratio: 0.0185449 max_rss_mb: 1636.45
1597MB 128thread base -> kops/s: 309.91 io_bytes/op: 1438.25 miss_ratio: 0.018399 max_rss_mb: 1689.98
1597MB 128thread folly -> kops/s: 267.605 io_bytes/op: 1394.16 miss_ratio: 0.0180286 max_rss_mb: 1631.91
1597MB 128thread gt_clock -> kops/s: 691.518 io_bytes/op: 9056.73 miss_ratio: 0.0186572 max_rss_mb: 1982.26
1597MB 128thread new_clock -> kops/s: 1406.12 io_bytes/op: 1440.82 miss_ratio: 0.0185463 max_rss_mb: 1685.63
610MB 1thread base -> kops/s: 45.511 io_bytes/op: 2279.61 miss_ratio: 0.0290528 max_rss_mb: 615.137
610MB 1thread folly -> kops/s: 43.386 io_bytes/op: 2217.29 miss_ratio: 0.0289282 max_rss_mb: 600.996
610MB 1thread gt_clock -> kops/s: 46.207 io_bytes/op: 2275.51 miss_ratio: 0.0290057 max_rss_mb: 637.934
610MB 1thread new_clock -> kops/s: 48.879 io_bytes/op: 2283.1 miss_ratio: 0.0291253 max_rss_mb: 613.5
610MB 32thread base -> kops/s: 306.59 io_bytes/op: 2250 miss_ratio: 0.0288721 max_rss_mb: 683.402
610MB 32thread folly -> kops/s: 269.176 io_bytes/op: 2187.86 miss_ratio: 0.0286938 max_rss_mb: 628.742
610MB 32thread gt_clock -> kops/s: 855.097 io_bytes/op: 2279.26 miss_ratio: 0.0288009 max_rss_mb: 733.062
610MB 32thread new_clock -> kops/s: 1121.47 io_bytes/op: 2244.29 miss_ratio: 0.0289046 max_rss_mb: 666.453
610MB 128thread base -> kops/s: 305.079 io_bytes/op: 2252.43 miss_ratio: 0.0288884 max_rss_mb: 723.457
610MB 128thread folly -> kops/s: 269.583 io_bytes/op: 2204.58 miss_ratio: 0.0287001 max_rss_mb: 676.426
610MB 128thread gt_clock -> kops/s: 53.298 io_bytes/op: 8128.98 miss_ratio: 0.0292452 max_rss_mb: 956.273
610MB 128thread new_clock -> kops/s: 1301.09 io_bytes/op: 2246.04 miss_ratio: 0.0289171 max_rss_mb: 788.812
The new version is still winning every time, sometimes dramatically so, and we can tell from the maximum resident memory numbers (which contain some noise, by the way) that the new cache is not cheating on memory usage. IMPORTANT: The previous generation experimental clock cache appears to hit a serious bottleneck in the higher thread count configurations, presumably due to some of its waiting functionality. (The same bottleneck is not seen with partitioned index+filters.)
Now we consider even smaller cache sizes, with higher miss ratios, eviction work, etc.
233MB 1thread base -> kops/s: 10.557 io_bytes/op: 227040 miss_ratio: 0.0403105 max_rss_mb: 247.371
233MB 1thread folly -> kops/s: 15.348 io_bytes/op: 112007 miss_ratio: 0.0372238 max_rss_mb: 245.293
233MB 1thread gt_clock -> kops/s: 6.365 io_bytes/op: 244854 miss_ratio: 0.0413873 max_rss_mb: 259.844
233MB 1thread new_clock -> kops/s: 47.501 io_bytes/op: 2591.93 miss_ratio: 0.0330989 max_rss_mb: 242.461
233MB 32thread base -> kops/s: 96.498 io_bytes/op: 363379 miss_ratio: 0.0459966 max_rss_mb: 479.227
233MB 32thread folly -> kops/s: 109.95 io_bytes/op: 314799 miss_ratio: 0.0450032 max_rss_mb: 400.738
233MB 32thread gt_clock -> kops/s: 2.353 io_bytes/op: 385397 miss_ratio: 0.048445 max_rss_mb: 500.688
233MB 32thread new_clock -> kops/s: 1088.95 io_bytes/op: 2567.02 miss_ratio: 0.0330593 max_rss_mb: 303.402
233MB 128thread base -> kops/s: 84.302 io_bytes/op: 378020 miss_ratio: 0.0466558 max_rss_mb: 1051.84
233MB 128thread folly -> kops/s: 89.921 io_bytes/op: 338242 miss_ratio: 0.0460309 max_rss_mb: 812.785
233MB 128thread gt_clock -> kops/s: 2.588 io_bytes/op: 462833 miss_ratio: 0.0509158 max_rss_mb: 1109.94
233MB 128thread new_clock -> kops/s: 1299.26 io_bytes/op: 2565.94 miss_ratio: 0.0330531 max_rss_mb: 361.016
89MB 1thread base -> kops/s: 0.574 io_bytes/op: 5.35977e+06 miss_ratio: 0.274427 max_rss_mb: 91.3086
89MB 1thread folly -> kops/s: 0.578 io_bytes/op: 5.16549e+06 miss_ratio: 0.27276 max_rss_mb: 96.8984
89MB 1thread gt_clock -> kops/s: 0.512 io_bytes/op: 4.13111e+06 miss_ratio: 0.242817 max_rss_mb: 119.441
89MB 1thread new_clock -> kops/s: 48.172 io_bytes/op: 2709.76 miss_ratio: 0.0346162 max_rss_mb: 100.754
89MB 32thread base -> kops/s: 5.779 io_bytes/op: 6.14192e+06 miss_ratio: 0.320399 max_rss_mb: 311.812
89MB 32thread folly -> kops/s: 5.601 io_bytes/op: 5.83838e+06 miss_ratio: 0.313123 max_rss_mb: 252.418
89MB 32thread gt_clock -> kops/s: 0.77 io_bytes/op: 3.99236e+06 miss_ratio: 0.236296 max_rss_mb: 396.422
89MB 32thread new_clock -> kops/s: 1064.97 io_bytes/op: 2687.23 miss_ratio: 0.0346134 max_rss_mb: 155.293
89MB 128thread base -> kops/s: 4.959 io_bytes/op: 6.20297e+06 miss_ratio: 0.323945 max_rss_mb: 823.43
89MB 128thread folly -> kops/s: 4.962 io_bytes/op: 5.9601e+06 miss_ratio: 0.319857 max_rss_mb: 626.824
89MB 128thread gt_clock -> kops/s: 1.009 io_bytes/op: 4.1083e+06 miss_ratio: 0.242512 max_rss_mb: 1095.32
89MB 128thread new_clock -> kops/s: 1224.39 io_bytes/op: 2688.2 miss_ratio: 0.0346207 max_rss_mb: 218.223
^ Now something interesting has happened: the new clock cache has gained a dramatic lead in the single-threaded case, and this is because the cache is so small, and full filters are so big, that dividing the cache into 64 shards leads to significant (random) imbalances in cache shards and excessive churn in imbalanced shards. This new clock cache only uses two shards for this configuration, and that helps to ensure that entries are part of a sufficiently big pool that their eviction order resembles the single-shard order. (This effect is not seen with partitioned index+filters.)
Even smaller cache size:
34MB 1thread base -> kops/s: 0.198 io_bytes/op: 1.65342e+07 miss_ratio: 0.939466 max_rss_mb: 48.6914
34MB 1thread folly -> kops/s: 0.201 io_bytes/op: 1.63416e+07 miss_ratio: 0.939081 max_rss_mb: 45.3281
34MB 1thread gt_clock -> kops/s: 0.448 io_bytes/op: 4.43957e+06 miss_ratio: 0.266749 max_rss_mb: 100.523
34MB 1thread new_clock -> kops/s: 1.055 io_bytes/op: 1.85439e+06 miss_ratio: 0.107512 max_rss_mb: 75.3125
34MB 32thread base -> kops/s: 3.346 io_bytes/op: 1.64852e+07 miss_ratio: 0.93596 max_rss_mb: 180.48
34MB 32thread folly -> kops/s: 3.431 io_bytes/op: 1.62857e+07 miss_ratio: 0.935693 max_rss_mb: 137.531
34MB 32thread gt_clock -> kops/s: 1.47 io_bytes/op: 4.89704e+06 miss_ratio: 0.295081 max_rss_mb: 392.465
34MB 32thread new_clock -> kops/s: 8.19 io_bytes/op: 3.70456e+06 miss_ratio: 0.20826 max_rss_mb: 519.793
34MB 128thread base -> kops/s: 2.293 io_bytes/op: 1.64351e+07 miss_ratio: 0.931866 max_rss_mb: 449.484
34MB 128thread folly -> kops/s: 2.34 io_bytes/op: 1.6219e+07 miss_ratio: 0.932023 max_rss_mb: 396.457
34MB 128thread gt_clock -> kops/s: 1.798 io_bytes/op: 5.4241e+06 miss_ratio: 0.324881 max_rss_mb: 1104.41
34MB 128thread new_clock -> kops/s: 10.519 io_bytes/op: 2.39354e+06 miss_ratio: 0.136147 max_rss_mb: 1050.52
As the miss ratio gets higher (say, above 10%), the CPU time spent in eviction starts to erode the advantage of using fewer shards (13% miss rate much lower than 94%). LRU's O(1) eviction time can eventually pay off when there's enough block cache churn:
13MB 1thread base -> kops/s: 0.195 io_bytes/op: 1.65732e+07 miss_ratio: 0.946604 max_rss_mb: 45.6328
13MB 1thread folly -> kops/s: 0.197 io_bytes/op: 1.63793e+07 miss_ratio: 0.94661 max_rss_mb: 33.8633
13MB 1thread gt_clock -> kops/s: 0.519 io_bytes/op: 4.43316e+06 miss_ratio: 0.269379 max_rss_mb: 100.684
13MB 1thread new_clock -> kops/s: 0.176 io_bytes/op: 1.54148e+07 miss_ratio: 0.91545 max_rss_mb: 66.2383
13MB 32thread base -> kops/s: 3.266 io_bytes/op: 1.65544e+07 miss_ratio: 0.943386 max_rss_mb: 132.492
13MB 32thread folly -> kops/s: 3.396 io_bytes/op: 1.63142e+07 miss_ratio: 0.943243 max_rss_mb: 101.863
13MB 32thread gt_clock -> kops/s: 2.758 io_bytes/op: 5.13714e+06 miss_ratio: 0.310652 max_rss_mb: 396.121
13MB 32thread new_clock -> kops/s: 3.11 io_bytes/op: 1.23419e+07 miss_ratio: 0.708425 max_rss_mb: 321.758
13MB 128thread base -> kops/s: 2.31 io_bytes/op: 1.64823e+07 miss_ratio: 0.939543 max_rss_mb: 425.539
13MB 128thread folly -> kops/s: 2.339 io_bytes/op: 1.6242e+07 miss_ratio: 0.939966 max_rss_mb: 346.098
13MB 128thread gt_clock -> kops/s: 3.223 io_bytes/op: 5.76928e+06 miss_ratio: 0.345899 max_rss_mb: 1087.77
13MB 128thread new_clock -> kops/s: 2.984 io_bytes/op: 1.05341e+07 miss_ratio: 0.606198 max_rss_mb: 898.27
gt_clock is clearly blowing way past its memory budget for lower miss rates and best throughput. new_clock also seems to be exceeding budgets, and this warrants more investigation but is not the use case we are targeting with the new cache. With partitioned index+filter, the miss ratio is much better, and although still high enough that the eviction CPU time is definitely offsetting mutex contention:
13MB 1thread base -> kops/s: 16.326 io_bytes/op: 23743.9 miss_ratio: 0.205362 max_rss_mb: 65.2852
13MB 1thread folly -> kops/s: 15.574 io_bytes/op: 19415 miss_ratio: 0.184157 max_rss_mb: 56.3516
13MB 1thread gt_clock -> kops/s: 14.459 io_bytes/op: 22873 miss_ratio: 0.198355 max_rss_mb: 63.9688
13MB 1thread new_clock -> kops/s: 16.34 io_bytes/op: 24386.5 miss_ratio: 0.210512 max_rss_mb: 61.707
13MB 128thread base -> kops/s: 289.786 io_bytes/op: 23710.9 miss_ratio: 0.205056 max_rss_mb: 103.57
13MB 128thread folly -> kops/s: 185.282 io_bytes/op: 19433.1 miss_ratio: 0.184275 max_rss_mb: 116.219
13MB 128thread gt_clock -> kops/s: 354.451 io_bytes/op: 23150.6 miss_ratio: 0.200495 max_rss_mb: 102.871
13MB 128thread new_clock -> kops/s: 295.359 io_bytes/op: 24626.4 miss_ratio: 0.212452 max_rss_mb: 121.109
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10626
Test Plan: updated unit tests, stress/crash test runs including with TSAN, ASAN, UBSAN
Reviewed By: anand1976
Differential Revision: D39368406
Pulled By: pdillinger
fbshipit-source-id: 5afc44da4c656f8f751b44552bbf27bd3ca6fef9
Summary:
The stats were not accurate for the coroutine version of MultiGet. This PR fixes it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10673
Reviewed By: akankshamahajan15
Differential Revision: D39492615
Pulled By: anand1976
fbshipit-source-id: b46c04e15ea27e66f4c31f00c66497aa283bf9d3
Summary:
Fix a bug in the async IO/coroutine version of MultiGet that may cause a segfault or assertion failure due to accessing an invalid file index in a LevelFilesBrief. The bug is that when a MultiGetRange is split into two, we may re-process keys in the original range that were already marked to be skipped (in ```current_level_range_```) due to not overlapping the level.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/10688
Reviewed By: gitbw95
Differential Revision: D39556131
Pulled By: anand1976
fbshipit-source-id: 65e79438508a283cb19e64eca5c91d0714b81458
Levi Tamasi