Summary:
Adds an improved, replacement Bloom filter implementation (FastLocalBloom) for full and partitioned filters in the block-based table. This replacement is faster and more accurate, especially for high bits per key or millions of keys in a single filter.
Speed
The improved speed, at least on recent x86_64, comes from
* Using fastrange instead of modulo (%)
* Using our new hash function (XXH3 preview, added in a previous commit), which is much faster for large keys and only *slightly* slower on keys around 12 bytes if hashing the same size many thousands of times in a row.
* Optimizing the Bloom filter queries with AVX2 SIMD operations. (Added AVX2 to the USE_SSE=1 build.) Careful design was required to support (a) SIMD-optimized queries, (b) compatible non-SIMD code that's simple and efficient, (c) flexible choice of number of probes, and (d) essentially maximized accuracy for a cache-local Bloom filter. Probes are made eight at a time, so any number of probes up to 8 is the same speed, then up to 16, etc.
* Prefetching cache lines when building the filter. Although this optimization could be applied to the old structure as well, it seems to balance out the small added cost of accumulating 64 bit hashes for adding to the filter rather than 32 bit hashes.
Here's nominal speed data from filter_bench (200MB in filters, about 10k keys each, 10 bits filter data / key, 6 probes, avg key size 24 bytes, includes hashing time) on Skylake DE (relatively low clock speed):
$ ./filter_bench -quick -impl=2 -net_includes_hashing # New Bloom filter
Build avg ns/key: 47.7135
Mixed inside/outside queries...
Single filter net ns/op: 26.2825
Random filter net ns/op: 150.459
Average FP rate %: 0.954651
$ ./filter_bench -quick -impl=0 -net_includes_hashing # Old Bloom filter
Build avg ns/key: 47.2245
Mixed inside/outside queries...
Single filter net ns/op: 63.2978
Random filter net ns/op: 188.038
Average FP rate %: 1.13823
Similar build time but dramatically faster query times on hot data (63 ns to 26 ns), and somewhat faster on stale data (188 ns to 150 ns). Performance differences on batched and skewed query loads are between these extremes as expected.
The only other interesting thing about speed is "inside" (query key was added to filter) vs. "outside" (query key was not added to filter) query times. The non-SIMD implementations are substantially slower when most queries are "outside" vs. "inside". This goes against what one might expect or would have observed years ago, as "outside" queries only need about two probes on average, due to short-circuiting, while "inside" always have num_probes (say 6). The problem is probably the nastily unpredictable branch. The SIMD implementation has few branches (very predictable) and has pretty consistent running time regardless of query outcome.
Accuracy
The generally improved accuracy (re: Issue https://github.com/facebook/rocksdb/issues/5857) comes from a better design for probing indices
within a cache line (re: Issue https://github.com/facebook/rocksdb/issues/4120) and improved accuracy for millions of keys in a single filter from using a 64-bit hash function (XXH3p). Design details in code comments.
Accuracy data (generalizes, except old impl gets worse with millions of keys):
Memory bits per key: FP rate percent old impl -> FP rate percent new impl
6: 5.70953 -> 5.69888
8: 2.45766 -> 2.29709
10: 1.13977 -> 0.959254
12: 0.662498 -> 0.411593
16: 0.353023 -> 0.0873754
24: 0.261552 -> 0.0060971
50: 0.225453 -> ~0.00003 (less than 1 in a million queries are FP)
Fixes https://github.com/facebook/rocksdb/issues/5857
Fixes https://github.com/facebook/rocksdb/issues/4120
Unlike the old implementation, this implementation has a fixed cache line size (64 bytes). At 10 bits per key, the accuracy of this new implementation is very close to the old implementation with 128-byte cache line size. If there's sufficient demand, this implementation could be generalized.
Compatibility
Although old releases would see the new structure as corrupt filter data and read the table as if there's no filter, we've decided only to enable the new Bloom filter with new format_version=5. This provides a smooth path for automatic adoption over time, with an option for early opt-in.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6007
Test Plan: filter_bench has been used thoroughly to validate speed, accuracy, and correctness. Unit tests have been carefully updated to exercise new and old implementations, as well as the logic to select an implementation based on context (format_version).
Differential Revision: D18294749
Pulled By: pdillinger
fbshipit-source-id: d44c9db3696e4d0a17caaec47075b7755c262c5f
Summary:
Add a new API that allows a user to call MultiGet specifying multiple keys belonging to different column families. This is mainly useful for users who want to do a consistent read of keys across column families, with the added performance benefits of batching and returning values using PinnableSlice.
As part of this change, the code in the original multi-column family MultiGet for acquiring the super versions has been refactored into a separate function that can be used by both, the batching and the non-batching versions of MultiGet.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5816
Test Plan:
make check
make asan_check
asan_crash_test
Differential Revision: D18408676
Pulled By: anand1976
fbshipit-source-id: 933e7bec91dd70e7b633be4ff623a1116cc28c8d
Summary:
The calculation in BlockBasedTable::MultiGet for the required buffer length for reading in compressed blocks is incorrect. It needs to take the 5-byte block trailer into account.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6014
Test Plan: Add a unit test DBBasicTest.MultiGetBufferOverrun that fails in asan_check before the fix, and passes after.
Differential Revision: D18412753
Pulled By: anand1976
fbshipit-source-id: 754dfb66be1d5f161a7efdf87be872198c7e3b72
Summary:
When users use Level-Compaction-with-TTL by setting `cf_options.ttl`, the ttl-expired data could take n*ttl time to reach the bottom level (where n is the number of levels) due to how the `creation_time` table property was calculated for the newly created files during compaction. The creation time of new files was set to a max of all compaction-input-files-creation-times which essentially resulted in resetting the ttl as the key range moves across levels. This behavior is now fixed by changing the `creation_time` to be based on minimum of all compaction-input-files-creation-times; this will cause cascading compactions across levels for the ttl-expired data to move to the bottom level, resulting in getting rid of tombstones/deleted-data faster.
This will help start cascading compactions to move the expired key range to the bottom-most level faster.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5992
Test Plan: `make check`
Differential Revision: D18257883
Pulled By: sagar0
fbshipit-source-id: 00df0bb8d0b7e14d9fc239df2cba8559f3e54cbc
Summary:
Only if there is a crash, power failure, or I/O error in
DeleteBackup, shared or private files from the backup might be left
behind that are not cleaned up by PurgeOldBackups or DeleteBackup-- only
by GarbageCollect. This makes the BackupEngine API "leaky by default."
Even if it means a modest performance hit, I think we should make
Delete and Purge do as they say, with ongoing best effort: i.e. future
calls will attempt to finish any incomplete work from earlier calls.
This change does that by having DeleteBackup and PurgeOldBackups do a
GarbageCollect, unless (to minimize performance hit) this BackupEngine
has already done a GarbageCollect and there have been no
deletion-related I/O errors in that GarbageCollect or since then.
Rejected alternative 1: remove meta file last instead of first. This would in theory turn partially deleted backups into corrupted backups, but code changes would be needed to allow the missing files and consider it acceptably corrupt, rather than failing to open the BackupEngine. This might be a reasonable choice, but I mostly rejected it because it doesn't solve the legacy problem of cleaning up existing lingering files.
Rejected alternative 2: use a deletion marker file. If deletion started with creating a file that marks a backup as flagged for deletion, then we could reliably detect partially deleted backups and efficiently finish removing them. In addition to not solving the legacy problem, this could be precarious if there's a disk full situation, and we try to create a new file in order to delete some files. Ugh.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6015
Test Plan: Updated unit tests
Differential Revision: D18401333
Pulled By: pdillinger
fbshipit-source-id: 12944e372ce6809f3f5a4c416c3b321a8927d925
Summary:
The patch exposes the file numbers of the SSTs as well as the oldest blob
files they contain a reference to through the GetColumnFamilyMetaData/
GetLiveFilesMetaData interface.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6011
Test Plan:
Fixed and extended the existing unit tests. (The earlier ColumnFamilyMetaDataTest
wasn't really testing anything because the generated memtables were never
flushed, so the metadata structure was essentially empty.)
Differential Revision: D18361697
Pulled By: ltamasi
fbshipit-source-id: d5ed1d94ac70858b84393c48711441ddfe1251e9
Summary:
This PR fixes https://github.com/facebook/rocksdb/issues/5975. In ```BlockBasedTable::RetrieveMultipleBlocks()```, we were calling ```MaybeReadBlocksAndLoadToCache()```, which is a no-op if neither uncompressed nor compressed block cache are configured.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5991
Test Plan:
1. Add unit tests that fail with the old code and pass with the new
2. make check and asan_check
Cc spetrunia
Differential Revision: D18272744
Pulled By: anand1976
fbshipit-source-id: e62fa6090d1a6adf84fcd51dfd6859b03c6aebfe
Summary:
For upcoming new SST filter implementations, we will use a new
64-bit hash function (XXH3 preview, slightly modified). This change
updates hash.{h,cc} for that change, adds unit tests, and out-of-lines
the implementations to keep hash.h as clean/small as possible.
In developing the unit tests, I discovered that the XXH3 preview always
returns zero for the empty string. Zero is problematic for some
algorithms (including an upcoming SST filter implementation) if it
occurs more often than at the "natural" rate, so it should not be
returned from trivial values using trivial seeds. I modified our fork
of XXH3 to return a modest hash of the seed for the empty string.
With hash function details out-of-lines in hash.h, it makes sense to
enable XXH_INLINE_ALL, so that direct calls to XXH64/XXH32/XXH3p
are inlined. To fix array-bounds warnings on some inline calls, I
injected some casts to uintptr_t in xxhash.cc. (Issue reported to Yann.)
Revised: Reverted using XXH_INLINE_ALL for now. Some Facebook
checks are unhappy about #include on xxhash.cc file. I would
fix that by rename to xxhash_cc.h, but to best preserve history I want
to do that in a separate commit (PR) from the uintptr casts.
Also updated filter_bench for this change, improving the performance
predictability of dry run hashing and adding support for 64-bit hash
(for upcoming new SST filter implementations, minor dead code in the
tool for now).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5984
Differential Revision: D18246567
Pulled By: pdillinger
fbshipit-source-id: 6162fbf6381d63c8cc611dd7ec70e1ddc883fbb8
Summary:
Previously, periodic compaction is not supported in universal compaction. Add the support using following approach: if any file is marked as qualified for periodid compaction, trigger a full compaction. If a full compaction is prevented by files being compacted, try to compact the higher levels than files currently being compacted. If in this way we can only compact the last sorted run and none of the file to be compacted qualifies for periodic compaction, skip the compact. This is to prevent the same single level compaction from being executed again and again.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5970
Test Plan: Add several test cases.
Differential Revision: D18147097
fbshipit-source-id: 8ecc308154d9aca96fb192c51fbceba3947550c1
Summary:
Right now, by default FIFO compaction has no TTL. We believe that a default TTL of 30 days will be better. With this patch, the default will be changed to 30 days. Default of Options.periodic_compaction_seconds will mean the same as options.ttl. If Options.ttl and Options.periodic_compaction_seconds left default, a default 30 days TTL will be used. If both options are set, the stricter value of the two will be used.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5987
Test Plan: Add an option sanitize test to cover the case.
Differential Revision: D18237935
fbshipit-source-id: a6dcea1f36c3849e13c0a69e413d73ad8eab58c9
Summary:
- Periodic compactions are auto-enabled if a compaction filter or a compaction filter factory is set, in Level Compaction.
- The default value of `periodic_compaction_seconds` is changed to UINT64_MAX, which lets RocksDB auto-tune periodic compactions as needed. An explicit value of 0 will still work as before ie. to disable periodic compactions completely. For now, on seeing a compaction filter along with a UINT64_MAX value for `periodic_compaction_seconds`, RocksDB will make SST files older than 30 days to go through periodic copmactions.
Some RocksDB users make use of compaction filters to control when their data can be deleted, usually with a custom TTL logic. But it is occasionally possible that the compactions get delayed by considerable time due to factors like low writes to a key range, data reaching bottom level, etc before the TTL expiry. Periodic Compactions feature was originally built to help such cases. Now periodic compactions are auto enabled by default when compaction filters or compaction filter factories are used, as it is generally helpful to all cases to collect garbage.
`periodic_compaction_seconds` is set to a large value, 30 days, in `SanitizeOptions` when RocksDB sees that a `compaction_filter` or `compaction_filter_factory` is used.
This is done only for Level Compaction style.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5865
Test Plan:
- Added a new test `DBCompactionTest.LevelPeriodicCompactionWithCompactionFilters` to make sure that `periodic_compaction_seconds` is set if either `compaction_filter` or `compaction_filter_factory` options are set.
- `COMPILE_WITH_ASAN=1 make check`
Differential Revision: D17659180
Pulled By: sagar0
fbshipit-source-id: 4887b9cf2e53cf2dc93a7b658c6b15e1181217ee
Summary:
A bug occasionally shows up in crash test, and https://github.com/facebook/rocksdb/issues/5851 reproduces it.
The bug can surface in the following way.
1. Database has multiple column families.
2. Between one DB restart, the last log file is corrupted in the middle (not the tail)
3. During restart, DB crashes between flushing between two column families.
Then DB will fail to be opened again with error "SST file is ahead of WALs".
Solution is to update the log number associated with each column family altogether after flushing all column families' memtables. The version edits should be written to a new MANIFEST. Only after writing to all these version edits succeed does RocksDB (atomically) points the CURRENT file to the new MANIFEST.
Test plan (on devserver):
```
$make all && make check
```
Specifically
```
$make db_test2
$./db_test2 --gtest_filter=DBTest2.CrashInRecoveryMultipleCF
```
Also checked for compatibility as follows.
Use this branch, run DBTest2.CrashInRecoveryMultipleCF and preserve the db directory.
Then checkout 5.4, build ldb, and dump the MANIFEST.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5856
Differential Revision: D17620818
Pulled By: riversand963
fbshipit-source-id: b52ce5969c9a8052cacec2bd805fcfb373589039
Summary:
This patch adds a number of new information elements to the FlushJobInfo and
CompactionJobInfo structures that are passed to EventListeners via the
OnFlush{Begin, Completed} and OnCompaction{Begin, Completed} callbacks.
Namely, for flushes, the file numbers of the new SST and the oldest blob file it
references are propagated. For compactions, the new pieces of information are
the file number, level, and the oldest blob file referenced by each compaction
input and output file.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5962
Test Plan:
Extended the EventListener unit tests with logic that checks that these information
elements are correctly propagated from the corresponding FileMetaData.
Differential Revision: D18095568
Pulled By: ltamasi
fbshipit-source-id: 6874359a6aadb53366b5fe87adcb2f9bd27a0a56
Summary:
Several error paths in opening of a plain table would leak memory. PR https://github.com/facebook/rocksdb/issues/5940 opened the leak to one more error path, which happens to have been (mistakenly) exercised by CuckooTableDBTest.AdaptiveTable. That test has been fixed, and the exercising of
plain table error cases (more than before) has been added as BadOptions1 and BadOptions2
to PlainTableDBTest. This effectively moved the memory leak to plain_table_db_test.
Also here is a cheap fix for the memory leak, without (yet?) changing the signature of
ReadTableProperties. This fixes ASAN on unit tests.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5951
Test Plan: make COMPILE_WITH_ASAN=1 check
Differential Revision: D18051940
Pulled By: pdillinger
fbshipit-source-id: e2952930c09a2b46c4f1ff09818c5090426929de
Summary:
Right now, when LevelIterator::Seek() is called, when a file is filtered out by prefix bloom filter, the position is put to the beginning of the next file. This is a confusing internal interface because many keys in the levels are skipped. Avoid this behavior by checking the key of the next file against the seek key, and invalidate the whole iterator if the prefix doesn't match.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5861
Test Plan: Add a new unit test to validate the behavior; run all exsiting tests; run crash_test
Differential Revision: D17918213
fbshipit-source-id: f06b47d937c7cc8919001f18dcc3af5b28c9cdac
Summary:
When there are concurrent flush job on the same CF, `OnFlushCompleted` can be called before the flush result being install to LSM. Fixing the issue by passing `FlushJobInfo` through `MemTable`, and the thread who commit the flush result can fetch the `FlushJobInfo` and fire `OnFlushCompleted` on behave of the thread actually writing the SST.
Fix https://github.com/facebook/rocksdb/issues/5892
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5908
Test Plan: Add new test. The test will fail without the fix.
Differential Revision: D17916144
Pulled By: riversand963
fbshipit-source-id: e18df67d9533b5baee52ae3605026cdeb05cbe10
Summary:
Compaction can call OnTableFileCreationCompleted(). If file is empty, "(nil)"
is used as the file name.
Do the same for flush.
Test plan (dev server):
```
make all
make check
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5905
Differential Revision: D17883285
Pulled By: riversand963
fbshipit-source-id: 6565884adbb00e8023d88b17dfb3b6eb92220b59
Summary:
This PR allows for the creation of custom env when using sst_dump. If
the user does not set options.env or set options.env to nullptr, then sst_dump
will automatically try to create a custom env depending on the path to the sst
file or db directory. In order to use this feature, the user must call
ObjectRegistry::Register() beforehand.
Test Plan (on devserver):
```
$make all && make check
```
All tests must pass to ensure this change does not break anything.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5845
Differential Revision: D17678038
Pulled By: riversand963
fbshipit-source-id: 58ecb4b3f75246d52b07c4c924a63ee61c1ee626
Summary:
This reverts commit 9fad3e21eb.
Iterator verification in stress tests sometimes fail for assertion
table/block_based/block_based_table_reader.cc:2973: void rocksdb::BlockBasedTableIterator<TBlockIter, TValue>::FindBlockForward() [with TBlockIter = rocksdb::DataBlockIter; TValue = rocksdb::Slice]: Assertion `!next_block_is_out_of_bound || user_comparator_.Compare(*read_options_.iterate_upper_bound, index_iter_->user_key()) <= 0' failed.
It is likely to be linked to https://github.com/facebook/rocksdb/pull/5286 together with https://github.com/facebook/rocksdb/pull/5468 as the former PR makes some child iterator's seek being avoided, so that upper bound condition fails to be updated there. Strictly speaking, the former PR was merged before the latter one, but the latter one feels a more important improvement so I choose to revert the former one for now.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5871
Differential Revision: D17689196
fbshipit-source-id: 4ded5be68f67bee2782d31a29cb72ea68f59dd8c
Summary:
as title.
Test Plan (on devserver):
```
$make all && make check
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5855
Differential Revision: D17615125
Pulled By: riversand963
fbshipit-source-id: bd6ed8cf59eafff41f0d1fc044f39e8f3573172a
Summary:
Partitioned filters make use of a top-level index to find the partition in which the filter resides. The top-level index has a key per partition. The key is guaranteed to be larger or equal than any key in that partition. When used with format_version 3, which excludes the sequence number form index keys, the separator key in the index could be equal to the prefix of the keys in the next partition. In this way, when searching for the key, the top-level index will lead us to the previous partition, which has no key with that prefix. The prefix bloom test thus returns false, although the prefix exists in the bloom of the next partition.
The patch fixes that by a hack: It always adds the prefix of the first key of the next partition to the bloom of the current partition. In this way, in the corner cases that the index will lead us to the previous partition, we still can find the bloom filter there.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5835
Differential Revision: D17513585
Pulled By: maysamyabandeh
fbshipit-source-id: e2d1ff26c759e6e03875c4d57f4228316ecf50e9
Summary:
For our default block cache, each additional entry has extra memory overhead. It include LRUHandle (72 bytes currently) and the cache key (two varint64, file id and offset). The usage is not negligible. For example for block_size=4k, the overhead accounts for an extra 2% memory usage for the cache. The patch charging the cache for the extra usage, reducing untracked memory usage outside block cache. The feature is enabled by default and can be disabled by passing kDontChargeCacheMetadata to the cache constructor.
This PR builds up on https://github.com/facebook/rocksdb/issues/4258
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5797
Test Plan:
- Existing tests are updated to either disable the feature when the test has too much dependency on the old way of accounting the usage or increasing the cache capacity to account for the additional charge of metadata.
- The Usage tests in cache_test.cc are augmented to test the cache usage under kFullChargeCacheMetadata.
Differential Revision: D17396833
Pulled By: maysamyabandeh
fbshipit-source-id: 7684ccb9f8a40ca595e4f5efcdb03623afea0c6f
Summary:
Since DynamicBloom is now only used in-memory, we're free to
change it without schema compatibility issues. The new implementation
is drawn from (with manifest permission)
303542a767/bloom_simulation_tests/foo.cc (L613)
This has several speed advantages over the prior implementation:
* Uses fastrange instead of %
* Minimum logic to determine first (and all) probed memory addresses
* (Major) Two probes per 64-bit memory fetch/write.
* Very fast and effective (murmur-like) hash expansion/re-mixing. (At
least on recent CPUs, integer multiplication is very cheap.)
While a Bloom filter with 512-bit cache locality has about a 1.15x FP
rate penalty (e.g. 0.84% to 0.97%), further restricting to two probes
per 64 bits incurs an additional 1.12x FP rate penalty (e.g. 0.97% to
1.09%). Nevertheless, the unit tests show no "mediocre" FP rate samples,
unlike the old implementation with more erratic FP rates.
Especially for the memtable, we expect speed to outweigh somewhat higher
FP rates. For example, a negative table query would have to be 1000x
slower than a BF query to justify doubling BF query time to shave 10% off
FP rate (working assumption around 1% FP rate). While that seems likely
for SSTs, my data suggests a speed factor of roughly 50x for the memtable
(vs. BF; ~1.5% lower write throughput when enabling memtable Bloom
filter, after this change). Thus, it's probably not worth even 5% more
time in the Bloom filter to shave off 1/10th of the Bloom FP rate, or 0.1%
in absolute terms, and it's probably at least 20% slower to recoup that
much FP rate from this new implementation. Because of this, we do not see
a need for a 'locality' option that affects the MemTable Bloom filter
and have decoupled the MemTable Bloom filter from Options::bloom_locality.
Note that just 3% more memory to the Bloom filter (10.3 bits per key vs.
just 10) is able to make up for the ~12% FP rate drop in the new
implementation:
[] # Nearly "ideal" FP-wise but reasonably fast cache-local implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_WORM64_FROM32_any.out 10000000 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_WORM64_FROM32_any.out time: 3.29372 sampled_fp_rate: 0.00985956 ...
[] # Close match to this new implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out 10000000 6 10.3 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out time: 2.10072 sampled_fp_rate: 0.00985655 ...
[] # Old locality=1 implementation
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_ROCKSDB_DYNAMIC_any.out 10000000 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_ROCKSDB_DYNAMIC_any.out time: 3.95472 sampled_fp_rate: 0.00988943 ...
Also note the dramatic speed improvement vs. alternatives.
--
Performance unit test: DynamicBloomTest.concurrent_with_perf is updated
to report more precise timing data. (Measure running time of each
thread, not just longest running thread, etc.) Results averaged over
various sizes enabled with --enable_perf and 20 runs each; old dynamic
bloom refers to locality=1, the faster of the old:
old dynamic bloom, avg add latency = 65.6468
new dynamic bloom, avg add latency = 44.3809
old dynamic bloom, avg query latency = 50.6485
new dynamic bloom, avg query latency = 43.2186
old avg parallel add latency = 41.678
new avg parallel add latency = 24.5238
old avg parallel hit latency = 14.6322
new avg parallel hit latency = 12.3939
old avg parallel miss latency = 16.7289
new avg parallel miss latency = 12.2134
Tested on a dedicated 64-bit production machine at Facebook. Significant
improvement all around.
Despite now using std::atomic<uint64_t>, quick before-and-after test on
a 32-bit machine (Intel Atom N270, released 2008) shows no regression in
performance, in some cases modest improvement.
--
Performance integration test (synthetic): with DEBUG_LEVEL=0, used
TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=fillrandom,readmissing,readrandom,stats --num=2000000
and optionally with -memtable_whole_key_filtering -memtable_bloom_size_ratio=0.01
300 runs each configuration.
Write throughput change by enabling memtable bloom:
Old locality=0: -3.06%
Old locality=1: -2.37%
New: -1.50%
conclusion -> seems to substantially close the gap
Readmissing throughput change by enabling memtable bloom:
Old locality=0: +34.47%
Old locality=1: +34.80%
New: +33.25%
conclusion -> maybe a small new penalty from FP rate
Readrandom throughput change by enabling memtable bloom:
Old locality=0: +31.54%
Old locality=1: +31.13%
New: +30.60%
conclusion -> maybe also from FP rate (after memtable flush)
--
Another conclusion we can draw from this new implementation is that the
existing 32-bit hash function is not inherently crippling the Bloom
filter speed or accuracy, below about 5 million keys. For speed, the
implementation is essentially the same whether starting with 32-bits or
64-bits of hash; it just determines whether the first multiplication
after fastrange is a pseudorandom expansion or needed re-mix. Note that
this multiplication can occur while memory is fetching.
For accuracy, in a standard configuration, you need about 5 million
keys before you have about a 1.1x FP penalty due to using a
32-bit hash vs. 64-bit:
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out $((5 * 1000 * 1000 * 10)) 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_FROM32_any.out time: 2.52069 sampled_fp_rate: 0.0118267 ...
[~/wormhashing/bloom_simulation_tests] ./foo_gcc_IMPL_CACHE_MUL64_BLOCK_any.out $((5 * 1000 * 1000 * 10)) 6 10 $RANDOM 100000000
./foo_gcc_IMPL_CACHE_MUL64_BLOCK_any.out time: 2.43871 sampled_fp_rate: 0.0109059
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5762
Differential Revision: D17214194
Pulled By: pdillinger
fbshipit-source-id: ad9da031772e985fd6b62a0e1db8e81892520595
Summary:
Adding a light weight API to get last live WAL file name and size. Meant to be used as a helper for backup/restore tooling in a larger ecosystem such as MySQL with a MyRocks storage engine.
Specifically within MySQL's backup/restore mechanism, this call can be made with a write lock on the mysql db to get a transactionally consistent snapshot of the current WAL file position along with other non-rocksdb log/data files.
Without this, the alternative would be to take the aforementioned lock, scan the WAL dir for all files, find the last file and note its exact size as the rocksdb 'checkpoint'.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5765
Differential Revision: D17172717
Pulled By: affandar
fbshipit-source-id: f2fabafd4c0e6fc45f126670c8c88a9f84cb8a37
Summary:
Before this PR, when the number of column families involved in a file ingestion exceeds 2, a bug in the looping logic prevents correct file number being assigned to each ingestion job.
Also skip deleting non-existing hard links during cleanup-after-failure.
Test plan (devserver)
```
$COMPILE_WITH_ASAN=1 make all
$./external_sst_file_test --gtest_filter=ExternalSSTFileTest/ExternalSSTFileTest.IngestFilesIntoMultipleColumnFamilies_*/*
$makke check
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5760
Differential Revision: D17142982
Pulled By: riversand963
fbshipit-source-id: 06c1847a4e7a402647bcf28d124e70f2a0f9daf6
Summary:
Open-source users recently reported two occurrences of LSM-tree corruption (https://github.com/facebook/rocksdb/issues/5558 is one), which would be caught by options.force_consistency_checks = true. options.force_consistency_checks has a usability limitation because it crashes the service once inconsistency is detected. This makes the feature hard to use. Most users serve from multiple RocksDB shards per server and the impacts of crashing the service is higher than it should be.
Instead, we just pass the error back to users without killing the service, and ask them to deal with the problem accordingly.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5744
Differential Revision: D17096940
Pulled By: pdhandharia
fbshipit-source-id: b6780039044e265f26ed2ad03c51f4abbe8b603c
Summary:
MyRocks currently sets `max_write_buffer_number_to_maintain` in order to maintain enough history for transaction conflict checking. The effectiveness of this approach depends on the size of memtables. When memtables are small, it may not keep enough history; when memtables are large, this may consume too much memory.
We are proposing a new way to configure memtable list history: by limiting the memory usage of immutable memtables. The new option is `max_write_buffer_size_to_maintain` and it will take precedence over the old `max_write_buffer_number_to_maintain` if they are both set to non-zero values. The new option accounts for the total memory usage of flushed immutable memtables and mutable memtable. When the total usage exceeds the limit, RocksDB may start dropping immutable memtables (which is also called trimming history), starting from the oldest one.
The semantics of the old option actually works both as an upper bound and lower bound. History trimming will start if number of immutable memtables exceeds the limit, but it will never go below (limit-1) due to history trimming.
In order the mimic the behavior with the new option, history trimming will stop if dropping the next immutable memtable causes the total memory usage go below the size limit. For example, assuming the size limit is set to 64MB, and there are 3 immutable memtables with sizes of 20, 30, 30. Although the total memory usage is 80MB > 64MB, dropping the oldest memtable will reduce the memory usage to 60MB < 64MB, so in this case no memtable will be dropped.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5022
Differential Revision: D14394062
Pulled By: miasantreble
fbshipit-source-id: 60457a509c6af89d0993f988c9b5c2aa9e45f5c5
Summary:
The batched MultiGet() implementation was not correctly handling bloom filter lookups when whole_key_filtering is disabled. It was incorrectly skipping keys not in the prefix_extractor domain, and not calling transform for keys in domain. This PR fixes both problems by moving the domain check and transformation to the FilterBlockReader.
Tests:
Unit test (confirmed failed before the fix)
make check
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5665
Differential Revision: D16902380
Pulled By: anand1976
fbshipit-source-id: a6be81ad68a6e37134a65246aec7a2c590eccf00
Summary:
The comments of snap_refresh_nanos advertise that the snapshot refresh feature will be disabled when the option is set to 0. This contract is however not honored in the code: https://github.com/facebook/rocksdb/pull/5278
The patch fixes that and also adds an assert to ensure that the feature is not used when the option is zero.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5724
Differential Revision: D16918185
Pulled By: maysamyabandeh
fbshipit-source-id: fec167287df7d85093e087fc39c0eb243e3bbd7e
Summary:
Recently readahead is introduced for checksum verifying. However, users cannot override the setting for the checksum verifying before external SST file ingestion. Introduce a new option for the purpose.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5721
Test Plan: Add a new unit test for it.
Differential Revision: D16906896
fbshipit-source-id: 218ec37001ddcc05411cefddbe233d15ab308476
Summary:
Right now VerifyChecksum() doesn't do read-ahead. In some use cases, users won't be able to achieve good performance. With this change, by default, RocksDB will do a default readahead, and users will be able to overwrite the readahead size by passing in a ReadOptions.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5713
Test Plan: Add a new unit test.
Differential Revision: D16860874
fbshipit-source-id: 0cff0fe79ac855d3d068e6ccd770770854a68413
Summary:
Update HISTORY.md by removing a feature from "Unreleased" to 6.4.0 after cherry-picking related commits to 6.4.fb branch.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5714
Differential Revision: D16865334
Pulled By: riversand963
fbshipit-source-id: f17ede905a1dfbbcdf98806ca398c618cf54748a
Summary:
Add a command in ldb so that users can print out tombstones in SST files.
In order to test the code, change the interface of LDBCommandRunner::RunCommand() so that it doesn't return from the program, but return the status code.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5615
Test Plan: Add a new unit test
Differential Revision: D16550326
fbshipit-source-id: 88ddfe6984bdcbb3a528abdd115089df09eba52e
Summary:
Most existing RocksDB unit tests run on `Env::Default()`. It will be useful to port the unit tests to non-default environments, e.g. `HdfsEnv`, etc.
This pull request is one step towards this goal. If RocksDB unit tests are built with a static library exposing a function `RegisterCustomObjects()`, then it is possible to implement custom object registrar logic in the library. RocksDB unit test can call `RegisterCustomObjects()` at the beginning.
By default, `ROCKSDB_UNITTESTS_WITH_CUSTOM_OBJECTS_FROM_STATIC_LIBS` is not defined, thus this PR has no impact on existing RocksDB because `RegisterCustomObjects()` is a noop.
Test plan (on devserver):
```
$make clean && COMPILE_WITH_ASAN=1 make -j32 all
$make check
```
All unit tests must pass.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5676
Differential Revision: D16679157
Pulled By: riversand963
fbshipit-source-id: aca571af3fd0525277cdc674248d0fe06e060f9d
Summary:
In some cases, we don't have to get really accurate number. Something like 10% off is fine, we can create a new option for that use case. In this case, we can calculate size for full files first, and avoid estimation inside SST files if full files got us a huge number. For example, if we already covered 100GB of data, we should be able to skip partial dives into 10 SST files of 30MB.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5609
Differential Revision: D16433481
Pulled By: elipoz
fbshipit-source-id: 5830b31e1c656d0fd3a00d7fd2678ddc8f6e601b
Summary:
Master branch had been left at 6.2 and history of 6.3 and beyond were merged. Updated this to correct.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5652
Differential Revision: D16570498
Pulled By: gfosco
fbshipit-source-id: 79f62ec570539a3e3d7d7c84a6cf7b722395fafe
Summary:
The new DB::GetApproximateSizes with SizeApproximationOptions argument, which allows to add more options/knobs to the DB::GetApproximateSizes call (beyond only the include_flags)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5626
Differential Revision: D16496913
Pulled By: elipoz
fbshipit-source-id: ee8c6c182330a285fa056ecfc3905a592b451720
Summary:
The ObjectRegistry class replaces the Registrar and NewCustomObjects. Objects are registered with the registry by Type (the class must implement the static const char *Type() method).
This change is necessary for a few reasons:
- By having a class (rather than static template instances), the class can be passed between compilation units, meaning that objects could be registered and shared from a dynamic library with an executable.
- By having a class with instances, different units could have different objects registered. This could be useful if, for example, one Option allowed for a dynamic library and one did not.
When combined with some other PRs (being able to load shared libraries, a Configurable interface to configure objects to/from string), this code will allow objects in external shared libraries to be added to a RocksDB image at run-time, rather than requiring every new extension to be built into the main library and called explicitly by every program.
Test plan (on riversand963's devserver)
```
$COMPILE_WITH_ASAN=1 make -j32 all && sleep 1 && make check
```
All tests pass.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5293
Differential Revision: D16363396
Pulled By: riversand963
fbshipit-source-id: fbe4acb615bfc11103eef40a0b288845791c0180
Summary:
RocksDB has historically stored uncompression dictionary objects in the block
cache as opposed to storing just the block contents. This neccesitated
evicting the object upon table close. With the new code, only the raw blocks
are stored in the cache, eliminating the need for eviction.
In addition, the patch makes the following improvements:
1) Compression dictionary blocks are now prefetched/pinned similarly to
index/filter blocks.
2) A copy operation got eliminated when the uncompression dictionary is
retrieved.
3) Errors related to retrieving the uncompression dictionary are propagated as
opposed to silently ignored.
Note: the patch temporarily breaks the compression dictionary evicition stats.
They will be fixed in a separate phase.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5584
Test Plan: make asan_check
Differential Revision: D16344151
Pulled By: ltamasi
fbshipit-source-id: 2962b295f5b19628f9da88a3fcebbce5a5017a7b
