Summary:
This PR implements a fault injection mechanism for injecting errors in reads in db_stress. The FaultInjectionTestFS is used for this purpose. A thread local structure is used to track the errors, so that each db_stress thread can independently enable/disable error injection and verify observed errors against expected errors. This is initially enabled only for Get and MultiGet, but can be extended to iterator as well once its proven stable.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6538
Test Plan:
crash_test
make check
Reviewed By: riversand963
Differential Revision: D20714347
Pulled By: anand1976
fbshipit-source-id: d7598321d4a2d72bda0ced57411a337a91d87dc7
Summary:
Towards making compaction logic compatible with user timestamp.
When computing boundaries and overlapping ranges for inputs of compaction, We need to compare SSTs by user key without timestamp.
Test plan (devserver):
```
make check
```
Several individual tests:
```
./version_set_test --gtest_filter=VersionStorageInfoTimestampTest.GetOverlappingInputs
./db_with_timestamp_compaction_test
./db_with_timestamp_basic_test
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6645
Reviewed By: ltamasi
Differential Revision: D20960012
Pulled By: riversand963
fbshipit-source-id: ad377fa9eb481bf7a8a3e1824aaade48cdc653a4
Summary:
New memory technologies are being developed by various hardware vendors (Intel DCPMM is one such technology currently available). These new memory types require different libraries for allocation and management (such as PMDK and memkind). The high capacities available make it possible to provision large caches (up to several TBs in size), beyond what is achievable with DRAM.
The new allocator provided in this PR uses the memkind library to allocate memory on different media.
**Performance**
We tested the new allocator using db_bench.
- For each test, we vary the size of the block cache (relative to the size of the uncompressed data in the database).
- The database is filled sequentially. Throughput is then measured with a readrandom benchmark.
- We use a uniform distribution as a worst-case scenario.
The plot shows throughput (ops/s) relative to a configuration with no block cache and default allocator.
For all tests, p99 latency is below 500 us.
**Changes**
- Add MemkindKmemAllocator
- Add --use_cache_memkind_kmem_allocator db_bench option (to create an LRU block cache with the new allocator)
- Add detection of memkind library with KMEM DAX support
- Add test for MemkindKmemAllocator
**Minimum Requirements**
- kernel 5.3.12
- ndctl v67 - https://github.com/pmem/ndctl
- memkind v1.10.0 - https://github.com/memkind/memkind
**Memory Configuration**
The allocator uses the MEMKIND_DAX_KMEM memory kind. Follow the instructions on[ memkind’s GitHub page](https://github.com/memkind/memkind) to set up NVDIMM memory accordingly.
Note on memory allocation with NVDIMM memory exposed as system memory.
- The MemkindKmemAllocator will only allocate from NVDIMM memory (using memkind_malloc with MEMKIND_DAX_KMEM kind).
- The default allocator is not restricted to RAM by default. Based on NUMA node latency, the kernel should allocate from local RAM preferentially, but it’s a kernel decision. numactl --preferred/--membind can be used to allocate preferentially/exclusively from the local RAM node.
**Usage**
When creating an LRU cache, pass a MemkindKmemAllocator object as argument.
For example (replace capacity with the desired value in bytes):
```
#include "rocksdb/cache.h"
#include "memory/memkind_kmem_allocator.h"
NewLRUCache(
capacity /*size_t*/,
6 /*cache_numshardbits*/,
false /*strict_capacity_limit*/,
false /*cache_high_pri_pool_ratio*/,
std::make_shared<MemkindKmemAllocator>());
```
Refer to [RocksDB’s block cache documentation](https://github.com/facebook/rocksdb/wiki/Block-Cache) to assign the LRU cache as block cache for a database.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6214
Reviewed By: cheng-chang
Differential Revision: D19292435
fbshipit-source-id: 7202f47b769e7722b539c86c2ffd669f64d7b4e1
Summary:
Although there are tests related to locking in transaction_test, this new test directly tests against TransactionLockMgr.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6599
Test Plan: make transaction_lock_mgr_test && ./transaction_lock_mgr_test
Reviewed By: lth
Differential Revision: D20673749
Pulled By: cheng-chang
fbshipit-source-id: 1fa4a13218e68d785f5a99924556751a8c5c0f31
Summary:
Adding a simple timer support to schedule work at a fixed time.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6543
Test Plan: TODO: clean up the unit tests, and make them better.
Reviewed By: siying
Differential Revision: D20465390
Pulled By: sagar0
fbshipit-source-id: cba143f70b6339863e1d0f8b8bf92e51c2b3d678
Summary:
When Travis times out, it's hard to determine whether
the last executing thing took an excessively long time or the
sum of all the work just exceeded the time limit. This
change inserts some timestamps in the output that should
make this easier to determine.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6643
Test Plan: CI (Travis mostly)
Reviewed By: anand1976
Differential Revision: D20843901
Pulled By: pdillinger
fbshipit-source-id: e7aae5434b0c609931feddf238ce4355964488b7
Summary:
This PR adds support for pipelined & parallel compression optimization for `BlockBasedTableBuilder`. This optimization makes block building, block compression and block appending a pipeline, and uses multiple threads to accelerate block compression. Users can set `CompressionOptions::parallel_threads` greater than 1 to enable compression parallelism.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6262
Reviewed By: ajkr
Differential Revision: D20651306
fbshipit-source-id: 62125590a9c15b6d9071def9dc72589c1696a4cb
Summary:
By supporting direct IO in RandomAccessFileReader::MultiRead, the benefits of parallel IO (IO uring) and direct IO can be combined.
In direct IO mode, read requests are aligned and merged together before being issued to RandomAccessFile::MultiRead, so blocks in the original requests might share the same underlying buffer, the shared buffers are returned in `aligned_bufs`, which is a new parameter of the `MultiRead` API.
For example, suppose alignment requirement for direct IO is 4KB, one request is (offset: 1KB, len: 1KB), another request is (offset: 3KB, len: 1KB), then since they all belong to page (offset: 0, len: 4KB), `MultiRead` only reads the page with direct IO into a buffer on heap, and returns 2 Slices referencing regions in that same buffer. See `random_access_file_reader_test.cc` for more examples.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6446
Test Plan: Added a new test `random_access_file_reader_test.cc`.
Reviewed By: anand1976
Differential Revision: D20097518
Pulled By: cheng-chang
fbshipit-source-id: ca48a8faf9c3af146465c102ef6b266a363e78d1
Summary:
In some of the test, db_basic_test may cause time out due to its long running time. Separate the timestamp related test from db_basic_test to avoid the potential issue.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6516
Test Plan: pass make asan_check
Differential Revision: D20423922
Pulled By: zhichao-cao
fbshipit-source-id: d6306f89a8de55b07bf57233e4554c09ef1fe23a
Summary:
In the `.travis.yml` file the `jdk: openjdk7` element is ignored when `language: cpp`. So whatever version of the JDK that was installed in the Travis container was used - typically JDK 11.
To ensure our RocksJava builds are working, we now instead install and use OpenJDK 8. Ideally we would use OpenJDK 7, as RocksJava supports Java 7, but many of the newer Travis containers don't support Java 7, so Java 8 is the next best thing.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6512
Differential Revision: D20388296
Pulled By: pdillinger
fbshipit-source-id: 8bbe6b59b70cfab7fe81ff63867d907fefdd2df1
Summary:
In Linux, when reopening DB with many SST files, profiling shows that 100% system cpu time spent for a couple of seconds for `GetLogicalBufferSize`. This slows down MyRocks' recovery time when site is down.
This PR introduces two new APIs:
1. `Env::RegisterDbPaths` and `Env::UnregisterDbPaths` lets `DB` tell the env when it starts or stops using its database directories . The `PosixFileSystem` takes this opportunity to set up a cache from database directories to the corresponding logical block sizes.
2. `LogicalBlockSizeCache` is defined only for OS_LINUX to cache the logical block sizes.
Other modifications:
1. rename `logical buffer size` to `logical block size` to be consistent with Linux terms.
2. declare `GetLogicalBlockSize` in `PosixHelper` to expose it to `PosixFileSystem`.
3. change the functions `IOError` and `IOStatus` in `env/io_posix.h` to have external linkage since they are used in other translation units too.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6457
Test Plan:
1. A new unit test is added for `LogicalBlockSizeCache` in `env/io_posix_test.cc`.
2. A new integration test is added for `DB` operations related to the cache in `db/db_logical_block_size_cache_test.cc`.
`make check`
Differential Revision: D20131243
Pulled By: cheng-chang
fbshipit-source-id: 3077c50f8065c0bffb544d8f49fb10bba9408d04
Summary:
* **macOS version:** 10.15.2 (Catalina)
* **XCode/Clang version:** Apple clang version 11.0.0 (clang-1100.0.33.16)
Before this bugfix the error generated is:
```
In file included from ./util/compression.h:23:
./snappy-1.1.7/snappy.h:76:59: error: unknown type name 'string'; did you mean 'std::string'?
size_t Compress(const char* input, size_t input_length, string* output);
^~~~~~
std::string
/Library/Developer/CommandLineTools/usr/bin/../include/c++/v1/iosfwd:211:65: note: 'std::string' declared here
typedef basic_string<char, char_traits<char>, allocator<char> > string;
^
In file included from db/builder.cc:10:
In file included from ./db/builder.h:12:
In file included from ./db/range_tombstone_fragmenter.h:15:
In file included from ./db/pinned_iterators_manager.h:12:
In file included from ./table/internal_iterator.h:13:
In file included from ./table/format.h:25:
In file included from ./options/cf_options.h:14:
In file included from ./util/compression.h:23:
./snappy-1.1.7/snappy.h:85:19: error: unknown type name 'string'; did you mean 'std::string'?
string* uncompressed);
^~~~~~
std::string
/Library/Developer/CommandLineTools/usr/bin/../include/c++/v1/iosfwd:211:65: note: 'std::string' declared here
typedef basic_string<char, char_traits<char>, allocator<char> > string;
^
2 errors generated.
make: *** [jls/db/builder.o] Error 1
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6496
Differential Revision: D20389254
Pulled By: pdillinger
fbshipit-source-id: 2864245c8d0dba7b2ab81294241a62f2adf02e20
Summary:
It's never too soon to refactor something. The patch splits the recently
introduced (`VersionEdit` related) `BlobFileState` into two classes
`BlobFileAddition` and `BlobFileGarbage`. The idea is that once blob files
are closed, they are immutable, and the only thing that changes is the
amount of garbage in them. In the new design, `BlobFileAddition` contains
the immutable attributes (currently, the count and total size of all blobs, checksum
method, and checksum value), while `BlobFileGarbage` contains the mutable
GC-related information elements (count and total size of garbage blobs). This is a
better fit for the GC logic and is more consistent with how SST files are handled.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6502
Test Plan: `make check`
Differential Revision: D20348352
Pulled By: ltamasi
fbshipit-source-id: ff93f0121e80ab15e0e0a6525ba0d6af16a0e008
Summary:
In the current code base, we can use FaultInjectionTestEnv to simulate the env issue such as file write/read errors, which are used in most of the test. The PR https://github.com/facebook/rocksdb/issues/5761 introduce the File System as a new Env API. This PR implement the FaultInjectionTestFS, which can be used to simulate when File System has issues such as IO error. user can specify any IOStatus error as input, such that FS corresponding actions will return certain error to the caller.
A set of ErrorHandlerFSTests are introduced for testing
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6414
Test Plan: pass make asan_check, pass error_handler_fs_test.
Differential Revision: D20252421
Pulled By: zhichao-cao
fbshipit-source-id: e922038f8ce7e6d1da329fd0bba7283c4b779a21
Summary:
BlobDB currently does not keep track of blob files: no records are written to
the manifest when a blob file is added or removed, and upon opening a database,
the list of blob files is populated simply based on the contents of the blob directory.
This means that lost blob files cannot be detected at the moment. We plan to solve
this issue by making blob files a part of `Version`; as a first step, this patch makes
it possible to store information about blob files in `VersionEdit`. Currently, this information
includes blob file number, total number and size of all blobs, and total number and size
of garbage blobs. However, the format is extensible: new fields can be added in
both a forward compatible and a forward incompatible manner if needed (similarly
to `kNewFile4`).
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6416
Test Plan: `make check`
Differential Revision: D19894234
Pulled By: ltamasi
fbshipit-source-id: f9753e1f2aedf6dadb70c09b345207cb9c58c329
Summary:
Add a utility class `Defer` to defer the execution of a function until the Defer object goes out of scope.
Used in VersionSet:: ProcessManifestWrites as an example.
The inline comments for class `Defer` have more details.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6382
Test Plan: `make defer_test version_set_test && ./defer_test && ./version_set_test`
Differential Revision: D19797538
Pulled By: cheng-chang
fbshipit-source-id: b1a9b7306e4fd4f48ec2ab55783caa561a315f0f
Summary:
It's logically correct for PinnableSlice to support move semantics to transfer ownership of the pinned memory region. This PR adds both move constructor and move assignment to PinnableSlice.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6374
Test Plan:
A set of unit tests for the move semantics are added in slice_test.
So `make slice_test && ./slice_test`.
Differential Revision: D19739254
Pulled By: cheng-chang
fbshipit-source-id: f898bd811bb05b2d87384ec58b645e9915e8e0b1
Summary:
I set up a mirror of our Java deps on github so we can download
them through github URLs rather than maven.org, which is proving
terribly unreliable from Travis builds.
Also sanitized calls to curl, so they are easier to read and
appropriately fail on download failure.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6348
Test Plan: CI
Differential Revision: D19633621
Pulled By: pdillinger
fbshipit-source-id: 7eb3f730953db2ead758dc94039c040f406790f3
Summary:
Both changes are related to RocksJava:
1. Allow dependencies that are already present on the host system due to Maven to be reused in Docker builds.
2. Extend the `make clean-not-downloaded` target to RocksJava, so that libraries needed as dependencies for the test suite are not deleted and re-downloaded unnecessarily.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6318
Differential Revision: D19608742
Pulled By: pdillinger
fbshipit-source-id: 25e25649e3e3212b537ac4512b40e2e53dc02ae7
Summary:
Clang analyzer was falsely reporting on use of txn=nullptr.
Added a new const variable so that it can properly prune impossible
control flows.
Also, 'make analyze' previously required setting USE_CLANG=1 as an
environment variable, not a make variable, or else compilation errors
like
g++: error: unrecognized command line option ‘-Wshorten-64-to-32’
Now USE_CLANG is not required for 'make analyze' (it's implied) and you
can do an incremental analysis (recompile what has changed) with
'USE_CLANG=1 make analyze_incremental'
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6244
Test Plan: 'make -j24 analyze', 'make crash_test'
Differential Revision: D19225950
Pulled By: pdillinger
fbshipit-source-id: 14f4039aa552228826a2de62b2671450e0fed3cb
Summary:
Currently, db_stress performs verification by calling `VerifyDb()` at the end of test and optionally before tests start. In case of corruption or incorrect result, it will be too late. This PR adds more verification in two ways.
1. For cf consistency test, each test thread takes a snapshot and verifies every N ops. N is configurable via `-verify_db_one_in`. This option is not supported in other stress tests.
2. For cf consistency test, we use another background thread in which a secondary instance periodically tails the primary (interval is configurable). We verify the secondary. Once an error is detected, we terminate the test and report. This does not affect other stress tests.
Test plan (devserver)
```
$./db_stress -test_cf_consistency -verify_db_one_in=0 -ops_per_thread=100000 -continuous_verification_interval=100
$./db_stress -test_cf_consistency -verify_db_one_in=1000 -ops_per_thread=10000 -continuous_verification_interval=0
$make crash_test
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6173
Differential Revision: D19047367
Pulled By: riversand963
fbshipit-source-id: aeed584ad71f9310c111445f34975e5ab47a0615
Summary:
Adds a python script to syntax check all python files in the
repository and report any errors.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6209
Test Plan:
'make check' with and without seeded syntax errors. Also look
for "No syntax errors in 34 .py files" on success, and in java_test CI output
Differential Revision: D19166756
Pulled By: pdillinger
fbshipit-source-id: 537df464b767260d66810b4cf4c9808a026c58a4
Summary:
And clean up related code, especially in stress test.
(More clean up of db_stress_test_base.cc coming after this.)
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6154
Test Plan: make check, make blackbox_crash_test for a bit
Differential Revision: D18938180
Pulled By: pdillinger
fbshipit-source-id: 524d27621b8dbb25f6dff40f1081e7c00630357e
Summary:
Error message when running `make` on Mac OS with master branch (v6.6.0):
```
$ make
$DEBUG_LEVEL is 1
Makefile:168: Warning: Compiling in debug mode. Don't use the resulting binary in production
third-party/folly/folly/synchronization/WaitOptions.cpp:6:10: fatal error: 'folly/synchronization/WaitOptions.h' file not found
#include <folly/synchronization/WaitOptions.h>
^
1 error generated.
third-party/folly/folly/synchronization/ParkingLot.cpp:6:10: fatal error: 'folly/synchronization/ParkingLot.h' file not found
#include <folly/synchronization/ParkingLot.h>
^
1 error generated.
third-party/folly/folly/synchronization/DistributedMutex.cpp:6:10: fatal error: 'folly/synchronization/DistributedMutex.h' file not found
#include <folly/synchronization/DistributedMutex.h>
^
1 error generated.
third-party/folly/folly/synchronization/AtomicNotification.cpp:6:10: fatal error: 'folly/synchronization/AtomicNotification.h' file not found
#include <folly/synchronization/AtomicNotification.h>
^
1 error generated.
third-party/folly/folly/detail/Futex.cpp:6:10: fatal error: 'folly/detail/Futex.h' file not found
#include <folly/detail/Futex.h>
^
1 error generated.
GEN util/build_version.cc
$DEBUG_LEVEL is 1
Makefile:168: Warning: Compiling in debug mode. Don't use the resulting binary in production
third-party/folly/folly/synchronization/WaitOptions.cpp:6:10: fatal error: 'folly/synchronization/WaitOptions.h' file not found
#include <folly/synchronization/WaitOptions.h>
^
1 error generated.
third-party/folly/folly/synchronization/ParkingLot.cpp:6:10: fatal error: 'folly/synchronization/ParkingLot.h' file not found
#include <folly/synchronization/ParkingLot.h>
^
1 error generated.
third-party/folly/folly/synchronization/DistributedMutex.cpp:6:10: fatal error: 'folly/synchronization/DistributedMutex.h' file not found
#include <folly/synchronization/DistributedMutex.h>
^
1 error generated.
third-party/folly/folly/synchronization/AtomicNotification.cpp:6:10: fatal error: 'folly/synchronization/AtomicNotification.h' file not found
#include <folly/synchronization/AtomicNotification.h>
^
1 error generated.
third-party/folly/folly/detail/Futex.cpp:6:10: fatal error: 'folly/detail/Futex.h' file not found
#include <folly/detail/Futex.h>
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6145
Differential Revision: D18910812
fbshipit-source-id: 5a4475466c2d0601657831a0b48d34316b2f0816
Summary:
db_stress_tool.cc now is a giant file. In order to main it easier to improve and maintain, break it down to multiple source files.
Most classes are turned into their own files. Separate .h and .cc files are created for gflag definiations. Another .h and .cc files are created for some common functions. Some test execution logic that is only loosely related to class StressTest is moved to db_stress_driver.h and db_stress_driver.cc. All the files are located under db_stress_tool/. The directory name is created as such because if we end it with either stress or test, .gitignore will ignore any file under it and makes it prone to issues in developements.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6134
Test Plan: Build under GCC7 with and without LITE on using GNU Make. Build with GCC 4.8. Build with cmake with -DWITH_TOOL=1
Differential Revision: D18876064
fbshipit-source-id: b25d0a7451840f31ac0f5ebb0068785f783fdf7d
Summary:
Before this fix, `make all` will emit full compilation command when building
object files in the third-party/folly directory even if default verbosity is
0 (AM_DEFAULT_VERBOSITY).
Test Plan (devserver):
```
$make all | tee build.log
$make check
```
Check build.log to verify.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6120
Differential Revision: D18795621
Pulled By: riversand963
fbshipit-source-id: 04641a8359cd4fd55034e6e797ed85de29ee2fe2
Summary:
Add the jni library for musl-libc, specifically for incorporating into Alpine based docker images. The classifier is `musl64`.
I have signed the CLA electronically.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/3143
Differential Revision: D18719372
fbshipit-source-id: 6189d149310b6436d6def7d808566b0234b23313
Summary:
**NOTE**: This also needs to be back-ported to be 6.4.6
Fix a regression introduced in f2bf0b2 by https://github.com/facebook/rocksdb/pull/5674 whereby the compiled library would get the wrong name on PPC64LE platforms.
On PPC64LE, the regression caused the library to be named `librocksdbjni-linux64.so` instead of `librocksdbjni-linux-ppc64le.so`.
This PR corrects the name back to `librocksdbjni-linux-ppc64le.so` and also corrects the ordering of conditional arguments in the Makefile to match the expected order as defined in the documentation for Make.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6080
Differential Revision: D18710351
fbshipit-source-id: d4db87ef378263b57de7f9edce1b7d15644cf9de
Summary:
* We can reuse downloaded 3rd-party libraries
* We can isolate the build to a Docker volume. This is useful for investigating failed builds, as we can examine the volume by assigning it a name during the build.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6079
Differential Revision: D18710263
fbshipit-source-id: 93f456ba44b49e48941c43b0c4d53995ecc1f404
Summary:
Had complications with LITE build and valgrind test.
Reverts/fixes small parts of PR https://github.com/facebook/rocksdb/issues/6007
Pull Request resolved: https://github.com/facebook/rocksdb/pull/6036
Test Plan:
make LITE=1 all check
and
ROCKSDB_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 make -j24 db_bloom_filter_test && ROCKSDB_VALGRIND_RUN=1 DISABLE_JEMALLOC=1 ./db_bloom_filter_test
Differential Revision: D18512238
Pulled By: pdillinger
fbshipit-source-id: 37213cf0d309edf11c483fb4b2fb6c02c2cf2b28
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:
From bzip2's official [download page](http://www.bzip.org/downloads.html), we could download it from sourceforge. This source would be more credible than previous web archive.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5995
Differential Revision: D18377662
fbshipit-source-id: e8353f83d5d6ea6067f78208b7bfb7f0d5b49c05
Summary:
include db_stress_tool in rocksdb tools lib
Test Plan (on devserver):
```
$make db_stress
$./db_stress
$make all && make check
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5950
Differential Revision: D18044399
Pulled By: riversand963
fbshipit-source-id: 895585abbbdfd8b954965921dba4b1400b7af1b1
Summary:
expose db stress test by providing db_stress_tool.h in public header.
This PR does the following:
- adds a new header, db_stress_tool.h, in include/rocksdb/
- renames db_stress.cc to db_stress_tool.cc
- adds a db_stress.cc which simply invokes a test function.
- update Makefile accordingly.
Test Plan (dev server):
```
make db_stress
./db_stress
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5937
Differential Revision: D17997647
Pulled By: riversand963
fbshipit-source-id: 1a8d9994f89ce198935566756947c518f0052410
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:
Make class ObsoleteFilesTest inherit from DBTestBase.
Test plan (on devserver):
```
$COMPILE_WITH_ASAN=1 make obsolete_files_test
$./obsolete_files_test
```
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5820
Differential Revision: D17452348
Pulled By: riversand963
fbshipit-source-id: b09f4581a18022ca2bfd79f2836c0bf7083f5f25
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:
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:
AtomicFlushStressTest is a powerful test, but right now we only run it for atomic_flush=true + disable_wal=true. We further extend it to the case where atomic_flush=false + disable_wal = false. All the workload generation and validation can stay the same.
Atomic flush crash test is also changed to switch between the two test scenarios. It makes the name "atomic flush crash test" out of sync from what it really does. We leave it as it is to avoid troubles with continous test set-up.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5729
Test Plan: Run "CRASH_TEST_KILL_ODD=188 TEST_TMPDIR=/dev/shm/ USE_CLANG=1 make whitebox_crash_test_with_atomic_flush", observe the settings used and see it passed.
Differential Revision: D16969791
fbshipit-source-id: 56e37487000ae631e31b0100acd7bdc441c04163
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
anand76
