Summary:
There's no technological impediment to allowing the Bloom
filter bits/key to be non-integer (fractional/decimal) values, and it
provides finer control over the memory vs. accuracy trade-off. This is
especially handy in using the format_version=5 Bloom filter in place
of the old one, because bits_per_key=9.55 provides the same accuracy as
the old bits_per_key=10.
This change not only requires refining the logic for choosing the best
num_probes for a given bits/key setting, it revealed a flaw in that logic.
As bits/key gets higher, the best num_probes for a cache-local Bloom
filter is closer to bpk / 2 than to bpk * 0.69, the best choice for a
standard Bloom filter. For example, at 16 bits per key, the best
num_probes is 9 (FP rate = 0.0843%) not 11 (FP rate = 0.0884%).
This change fixes and refines that logic (for the format_version=5
Bloom filter only, just in case) based on empirical tests to find
accuracy inflection points between each num_probes.
Although bits_per_key is now specified as a double, the new Bloom
filter converts/rounds this to "millibits / key" for predictable/precise
internal computations. Just in case of unforeseen compatibility
issues, we round to the nearest whole number bits / key for the
legacy Bloom filter, so as not to unlock new behaviors for it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6092
Test Plan: unit tests included
Differential Revision: D18711313
Pulled By: pdillinger
fbshipit-source-id: 1aa73295f152a995328cb846ef9157ae8a05522a
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:
Bug in PR https://github.com/facebook/rocksdb/issues/5941 when char is unsigned that should only affect
assertion on unused/invalid filter metadata.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6024
Test Plan: on ARM: ./bloom_test && ./db_bloom_filter_test && ./block_based_filter_block_test && ./full_filter_block_test && ./partitioned_filter_block_test
Differential Revision: D18461206
Pulled By: pdillinger
fbshipit-source-id: 68a7c813a0b5791c05265edc03cdf52c78880e9a
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:
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:
According to
https://github.com/facebook/rocksdb/wiki/Rocksdb-BlockBasedTable-Format,
the block read by BlockBasedTable::ReadMetaBlock is actually the meta index
block. Therefore, it is better to rename the function to ReadMetaIndexBlock.
This PR also applies some format change to existing code.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6009
Test Plan: make check
Differential Revision: D18333238
Pulled By: riversand963
fbshipit-source-id: 2c4340a29b3edba53d19c132cbfd04caf6242aed
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:
This change sets up for alternate implementations underlying
BloomFilterPolicy:
* Refactor BloomFilterPolicy and expose in internal .h file so that it's easy to iterate over / select implementations for testing, regardless of what the best public interface will look like. Most notably updated db_bloom_filter_test to use this.
* Hide FullFilterBitsBuilder from unit tests (alternate derived classes planned); expose the part important for testing (CalculateSpace), as abstract class BuiltinFilterBitsBuilder. (Also cleaned up internally exposed interface to CalculateSpace.)
* Rename BloomTest -> BlockBasedBloomTest for clarity (despite ongoing confusion between block-based table and block-based filter)
* Assert that block-based filter construction interface is only used on BloomFilterPolicy appropriately constructed. (A couple of tests updated to add ", true".)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5967
Test Plan: make check
Differential Revision: D18138704
Pulled By: pdillinger
fbshipit-source-id: 55ef9273423b0696309e251f50b8c1b5e9ec7597
Summary:
Some filtering tests were unfriendly to new implementations of
FilterBitsBuilder because of dynamic_cast to FullFilterBitsBuilder. Most
of those have now been cleaned up, worked around, or at least changed
from crash on dynamic_cast failure to individual test failure.
Also put some clarifying comments on filter-related APIs.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5960
Test Plan: make check
Differential Revision: D18121223
Pulled By: pdillinger
fbshipit-source-id: e83827d9d5d96315d96f8e25a99cd70f497d802c
Summary:
- Updated our included xxhash implementation to version 0.7.2 (== the latest dev version as of 2019-10-09).
- Using XXH_NAMESPACE (like other fb projects) to avoid potential name collisions.
- Added fastrange64, and unit tests for it and fastrange32. These are faster alternatives to hash % range.
- Use preview version of XXH3 instead of MurmurHash64A for NPHash64
-- Had to update cache_test to increase probability of passing for any given hash function.
- Use fastrange64 instead of % with uses of NPHash64
-- Had to fix WritePreparedTransactionTest.CommitOfDelayedPrepared to avoid deadlock apparently caused by new hash collision.
- Set default seed for NPHash64 because specifying a seed rarely makes sense for it.
- Removed unnecessary include xxhash.h in a popular .h file
- Rename preview version of XXH3 to XXH3p for clarity and to ease backward compatibility in case final version of XXH3 is integrated.
Relying on existing unit tests for NPHash64-related changes. Each new implementation of fastrange64 passed unit tests when manipulating my local build to select it. I haven't done any integration performance tests, but I consider the improved performance of the pieces being swapped in to be well established.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5909
Differential Revision: D18125196
Pulled By: pdillinger
fbshipit-source-id: f6bf83d49d20cbb2549926adf454fd035f0ecc0d
Summary:
The parts that are used to implement FilterPolicy /
NewBloomFilterPolicy and not used other than for the block-based table
should be consolidated under table/block_based/filter_policy*.
This change is step 2 of 2:
mv util/bloom.cc table/block_based/filter_policy.cc
This gets its own PR so that git has the best chance of following the
rename for blame purposes. Note that low-level shared implementation
details of Bloom filters remain in util/bloom_impl.h, and
util/bloom_test.cc remains where it is for now.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5966
Test Plan: make check
Differential Revision: D18124930
Pulled By: pdillinger
fbshipit-source-id: 823bc09025b3395f092ef46a46aa5ba92a914d84
Summary:
The parts that are used to implement FilterPolicy /
NewBloomFilterPolicy and not used other than for the block-based table
should be consolidated under table/block_based/filter_policy*. I don't
foresee sharing these APIs with e.g. the Plain Table because they don't
expose hashes for reuse in indexing.
This change is step 1 of 2:
(a) mv table/full_filter_bits_builder.h to
table/block_based/filter_policy_internal.h which I expect to expand
soon to internally reveal more implementation details for testing.
(b) consolidate eventual contents of table/block_based/filter_policy.cc
in util/bloom.cc, which has the most elaborate revision history
(see step 2 ...)
Step 2 soon to follow:
mv util/bloom.cc table/block_based/filter_policy.cc
This gets its own PR so that git has the best chance of following the
rename for blame purposes. Note that low-level shared implementation
details of Bloom filters are in util/bloom_impl.h.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5963
Test Plan: make check
Differential Revision: D18121199
Pulled By: pdillinger
fbshipit-source-id: 8f21732c3d8909777e3240e4ac3123d73140326a
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:
A recent change introduced readahead inside VerifyChecksum(). However it is not compatible with mmap mode and generated wrong checksum verification failure. Fix it by not enabling readahead in mmap
mode.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5945
Test Plan: Add a unit test that used to fail.
Differential Revision: D18021443
fbshipit-source-id: 6f2eb600f81b26edb02222563a4006869d576bff
Summary:
Amongst other things, PR https://github.com/facebook/rocksdb/issues/5504 refactored the filter block readers so that
only the filter block contents are stored in the block cache (as opposed to the
earlier design where the cache stored the filter block reader itself, leading to
potentially dangling pointers and concurrency bugs). However, this change
introduced a performance hit since with the new code, the metadata fields are
re-parsed upon every access. This patch reunites the block contents with the
filter bits reader to eliminate this overhead; since this is still a self-contained
pure data object, it is safe to store it in the cache. (Note: this is similar to how
the zstd digest is handled.)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5936
Test Plan:
make asan_check
filter_bench results for the old code:
```
$ ./filter_bench -quick
WARNING: Assertions are enabled; benchmarks unnecessarily slow
Building...
Build avg ns/key: 26.7153
Number of filters: 16669
Total memory (MB): 200.009
Bits/key actual: 10.0647
----------------------------
Inside queries...
Dry run (46b) ns/op: 33.4258
Single filter ns/op: 42.5974
Random filter ns/op: 217.861
----------------------------
Outside queries...
Dry run (25d) ns/op: 32.4217
Single filter ns/op: 50.9855
Random filter ns/op: 219.167
Average FP rate %: 1.13993
----------------------------
Done. (For more info, run with -legend or -help.)
$ ./filter_bench -quick -use_full_block_reader
WARNING: Assertions are enabled; benchmarks unnecessarily slow
Building...
Build avg ns/key: 26.5172
Number of filters: 16669
Total memory (MB): 200.009
Bits/key actual: 10.0647
----------------------------
Inside queries...
Dry run (46b) ns/op: 32.3556
Single filter ns/op: 83.2239
Random filter ns/op: 370.676
----------------------------
Outside queries...
Dry run (25d) ns/op: 32.2265
Single filter ns/op: 93.5651
Random filter ns/op: 408.393
Average FP rate %: 1.13993
----------------------------
Done. (For more info, run with -legend or -help.)
```
With the new code:
```
$ ./filter_bench -quick
WARNING: Assertions are enabled; benchmarks unnecessarily slow
Building...
Build avg ns/key: 25.4285
Number of filters: 16669
Total memory (MB): 200.009
Bits/key actual: 10.0647
----------------------------
Inside queries...
Dry run (46b) ns/op: 31.0594
Single filter ns/op: 43.8974
Random filter ns/op: 226.075
----------------------------
Outside queries...
Dry run (25d) ns/op: 31.0295
Single filter ns/op: 50.3824
Random filter ns/op: 226.805
Average FP rate %: 1.13993
----------------------------
Done. (For more info, run with -legend or -help.)
$ ./filter_bench -quick -use_full_block_reader
WARNING: Assertions are enabled; benchmarks unnecessarily slow
Building...
Build avg ns/key: 26.5308
Number of filters: 16669
Total memory (MB): 200.009
Bits/key actual: 10.0647
----------------------------
Inside queries...
Dry run (46b) ns/op: 33.2968
Single filter ns/op: 58.6163
Random filter ns/op: 291.434
----------------------------
Outside queries...
Dry run (25d) ns/op: 32.1839
Single filter ns/op: 66.9039
Random filter ns/op: 292.828
Average FP rate %: 1.13993
----------------------------
Done. (For more info, run with -legend or -help.)
```
Differential Revision: D17991712
Pulled By: ltamasi
fbshipit-source-id: 7ea205550217bfaaa1d5158ebd658e5832e60f29
Summary:
Plain table SSTs could crash sst_dump because of a bug in
PlainTableReader that can leave table_properties_ as null. Even if it
was intended not to keep the table properties in some cases, they were
leaked on the offending code path.
Steps to reproduce:
$ db_bench --benchmarks=fillrandom --num=2000000 --use_plain_table --prefix-size=12
$ sst_dump --file=0000xx.sst --show_properties
from [] to []
Process /dev/shm/dbbench/000014.sst
Sst file format: plain table
Raw user collected properties
------------------------------
Segmentation fault (core dumped)
Also added missing unit testing of plain table full_scan_mode, and
an assertion in NewIterator to check for regression.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5940
Test Plan: new unit test, manual, make check
Differential Revision: D18018145
Pulled By: pdillinger
fbshipit-source-id: 4310c755e824c4cd6f3f86a3abc20dfa417c5e07
Summary:
Partition Filters make use of a top-level index to find the partition that might have the bloom hash of the key. The index is with internal key format (before format version 3). Each partition contains the i) blooms of the keys in that range ii) bloom of prefixes of keys in that range, iii) the bloom of the prefix of the last key in the previous partition.
When ::SeekForPrev(key), we first perform a prefix bloom test on the SST file. The partition however is identified using the full internal key, rather than the prefix key. The reason is to be compatible with the internal key format of the top-level index. This creates a corner case. Example:
- SST k, Partition N: P1K1, P1K2
- SST k, top-level index: P1K2
- SST k+1, Partition 1: P2K1, P3K1
- SST k+1 top-level index: P3K1
When SeekForPrev(P1K3), it should point us to P1K2. However SST k top-level index would reject P1K3 since it is out of range.
One possible fix would be to search with the prefix P1 (instead of full internal key P1K3) however the details of properly comparing prefix with full internal key might get complicated. The fix we apply in this PR is to look into the last partition anyway even if the key is out of range.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5907
Differential Revision: D17889918
Pulled By: maysamyabandeh
fbshipit-source-id: 169fd7b3c71dbc08808eae5a8340611ebe5bdc1e
Summary:
Since we do not evict a file's blocks from block cache before that file
is deleted, we require a file's cache ID prefix is both unique and
non-reusable. However, the Windows functionality we were relying on only
guaranteed uniqueness. That meant a newly created file could be assigned
the same cache ID prefix as a deleted file. If the newly created file
had block offsets matching the deleted file, full cache keys could be
exactly the same, resulting in obsolete data blocks returned from cache
when trying to read from the new file.
We noticed this when running on FAT32 where compaction was writing out
of order keys due to reading obsolete blocks from its input files. The
functionality is documented as behaving the same on NTFS, although I
wasn't able to repro it there.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5844
Test Plan:
we had a reliable repro of out-of-order keys on FAT32 that
was fixed by this change
Differential Revision: D17752442
fbshipit-source-id: 95d983f9196cf415f269e19293b97341edbf7e00
Summary:
RocksDB has a MultiGet() API that implements batched key lookup for higher performance (https://github.com/facebook/rocksdb/blob/master/include/rocksdb/db.h#L468). Currently, batching is implemented in BlockBasedTableReader::MultiGet() for SST file lookups. One of the ways it improves performance is by pipelining bloom filter lookups (by prefetching required cachelines for all the keys in the batch, and then doing the probe) and thus hiding the cache miss latency. The same concept can be extended to the memtable as well. This PR involves implementing a pipelined bloom filter lookup in DynamicBloom, and implementing MemTable::MultiGet() that can leverage it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5818
Test Plan:
Existing tests
Performance Test:
Ran the below command which fills up the memtable and makes sure there are no flushes and then call multiget. Ran it on master and on the new change and see atleast 1% performance improvement across all the test runs I did. Sometimes the improvement was upto 5%.
TEST_TMPDIR=/data/users/$USER/benchmarks/feature/ numactl -C 10 ./db_bench -benchmarks="fillseq,multireadrandom" -num=600000 -compression_type="none" -level_compaction_dynamic_level_bytes -write_buffer_size=200000000 -target_file_size_base=200000000 -max_bytes_for_level_base=16777216 -reads=90000 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 -statistics -memtable_whole_key_filtering=true -memtable_bloom_size_ratio=10
Differential Revision: D17578869
Pulled By: vjnadimpalli
fbshipit-source-id: 23dc651d9bf49db11d22375bf435708875a1f192
Summary:
Example: using the tool before and after PR https://github.com/facebook/rocksdb/issues/5784 shows that
the refactoring, presumed performance-neutral, actually sped up SST
filters by about 3% to 8% (repeatable result):
Before:
- Dry run ns/op: 22.4725
- Single filter ns/op: 51.1078
- Random filter ns/op: 120.133
After:
+ Dry run ns/op: 22.2301
+ Single filter run ns/op: 47.4313
+ Random filter ns/op: 115.9
Only tests filters for the block-based table (full filters and
partitioned filters - same implementation; not block-based filters),
which seems to be the recommended format/implementation.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5825
Differential Revision: D17804987
Pulled By: pdillinger
fbshipit-source-id: 0f18a9c254c57f7866030d03e7fa4ba503bac3c5
Summary:
When an iterator reseek happens with the user specifying a new iterate_upper_bound in ReadOptions, and the new seek position is at the end of the same data block, the Seek() ends up using a stale value of data_block_within_upper_bound_ and may return incorrect results.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5883
Test Plan: Added a new test case DBIteratorTest.IterReseekNewUpperBound. Verified that it failed due to the assertion failure without the fix, and passes with the fix.
Differential Revision: D17752740
Pulled By: anand1976
fbshipit-source-id: f9b635ff5d6aeb0e1bef102cf8b2f900efd378e3
Summary:
Conflict resolving in 846e05005d ("Revert "Merging iterator to avoid child iterator reseek for some cases") caused some timer misplaced. Fix it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5874
Test Plan: See it build.
Differential Revision: D17705073
fbshipit-source-id: 9bd3a8dc4901ac33c2c6fc5b1091ffbc56a8529f
Summary:
Without this fix, compiler complains.
```
$ROCKSDB_NO_FBCODE=1 USE_CLANG=1 make ldb
table/block_based/full_filter_block.cc: In constructor ‘rocksdb::FullFilterBlockBuilder::FullFilterBlockBuilder(const rocksdb::SliceTransform*, bool, rocksdb::FilterBitsBuilder*)’:
table/block_based/full_filter_block.cc:20:43: error: declaration of ‘prefix_extractor’ shadows a member of 'this' [-Werror=shadow]
FilterBitsBuilder* filter_bits_builder)
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5872
Test Plan:
```
$ROCKSDB_NO_FBCODE=1 make all
```
Differential Revision: D17690058
Pulled By: riversand963
fbshipit-source-id: 19e3d9bd86e1123847095240e73d30da5d66240e
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:
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:
Further apply formatter to more recent commits.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5830
Test Plan: Run all existing tests.
Differential Revision: D17488031
fbshipit-source-id: 137458fd94d56dd271b8b40c522b03036943a2ab
Summary:
We are seeing a bug of wrong results with merging iterator's reseek avoidence feature and prefix extractor. Disable this optimization for now.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5815
Test Plan: Validated the same MyRocks case was fixed; run all existing tests.
Differential Revision: D17430776
fbshipit-source-id: aef664277ba0ab8a2e68331ff0db6ae682535371
Summary:
Refactoring to consolidate implementation details of legacy
Bloom filters. This helps to organize and document some related,
obscure code.
Also added make/cpp var TEST_CACHE_LINE_SIZE so that it's easy to
compile and run unit tests for non-native cache line size. (Fixed a
related test failure in db_properties_test.)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5784
Test Plan:
make check, including Recently added Bloom schema unit tests
(in ./plain_table_db_test && ./bloom_test), and including with
TEST_CACHE_LINE_SIZE=128U and TEST_CACHE_LINE_SIZE=256U. Tested the
schema tests with temporary fault injection into new implementations.
Some performance testing with modified unit tests suggest a small to moderate
improvement in speed.
Differential Revision: D17381384
Pulled By: pdillinger
fbshipit-source-id: ee42586da996798910fc45ac0b6289147f16d8df
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:
This will allow us to fix history by having the code changes for PR#5784 properly attributed to it.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5810
Differential Revision: D17400231
Pulled By: pdillinger
fbshipit-source-id: 2da8b1cdf2533cfedb35b5526eadefb38c291f09
Summary:
file_reader_writer.h and .cc contain several files and helper function, and it's hard to navigate. Separate it to multiple files and put them under file/
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5803
Test Plan: Build whole project using make and cmake.
Differential Revision: D17374550
fbshipit-source-id: 10efca907721e7a78ed25bbf74dc5410dea05987
Summary:
1. Put the similar logic of adding valid iterator to heap and check invalid iterator's status code to the same helper functions.
2. Because of 1, in the changing direction case, move around the places where we check status a little bit so that we can call the helper function there too. The logic would only divert in the case where the iterator is valid but status is not OK, which is not expected to happen. Add an assertion for that.
3. Put the logic of changing direction from forward to backward to a separate function so the unlikely code path is not in Prev().
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5793
Test Plan: run all existing tests.
Differential Revision: D17374397
fbshipit-source-id: d595ffcf156095c4bd0f5532bacba854482a2332
Summary:
Use delete to disable automatic generated methods instead of private, and put the constructor together for more clear.This modification cause the unused field warning, so add unused attribute to disable this warning.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5009
Differential Revision: D17288733
fbshipit-source-id: 8a767ce096f185f1db01bd28fc88fef1cdd921f3
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:
DynamicBloom was being used both for memory-only and for on-disk filters, as part of the PlainTable format. To set up enhancements to the memtable Bloom filter, this splits the code into two copies and removes unused features from each copy. Adds test PlainTableDBTest.BloomSchema to ensure no accidental change to that format.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5767
Differential Revision: D17206963
Pulled By: pdillinger
fbshipit-source-id: 6cce8d55305ed0df051b4c58bdc98c8ad81d0553
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:
PR https://github.com/facebook/rocksdb/issues/5584 decoupled the uncompression dictionary object from the underlying block data; however, this defeats the purpose of the digested ZSTD dictionary, since the whole point
of the digest is to create it once and reuse it over and over again. This patch goes back to
storing the uncompression dictionary itself in the cache (which should be now safe to do,
since it no longer includes a Statistics pointer), while preserving the rest of the refactoring.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5645
Test Plan: make asan_check
Differential Revision: D16551864
Pulled By: ltamasi
fbshipit-source-id: 2a7e2d34bb16e70e3c816506d5afe1d842057800
Summary:
To improve code readability, since RetrieveBlock already calls MaybeReadBlockAndLoadToCache, we avoid name similarity of the functions that call RetrieveBlock with MaybeReadBlockAndLoadToCache. The patch thus renames MaybeLoadBlocksToCache to RetrieveMultipleBlock and deletes GetDataBlockFromCache, which contains only two lines.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5726
Differential Revision: D16962535
Pulled By: maysamyabandeh
fbshipit-source-id: 99e8946808ce4eb7857592b9003812e3004f92d6
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:
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:
Previous PR https://github.com/facebook/rocksdb/pull/3601 added support for making prefix_extractor dynamically mutable. However, there was a missing check for hash index when creating new BlockBasedTableIterator. While the check may be redundant because no other types of IndexReader makes uses of the flag, it is less error-prone to add the missing check so that future index reader implementation will not worry about violating the contract.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5712
Differential Revision: D16842052
Pulled By: miasantreble
fbshipit-source-id: aef11c0ff7a690ed248f5b8fe23481cac486b381
