Summary:
- "rocksdb.compaction.key.drop.range_del" - number of keys dropped during compaction due to a range tombstone covering them
- "rocksdb.compaction.range_del.drop.obsolete" - number of range tombstones dropped due to compaction to bottom level and no snapshot saving them
- s/CompactionIteratorStats/CompactionIterationStats/g since this class is no longer specific to CompactionIterator -- it's also updated for range tombstone iteration during compaction
- Move the above class into a separate .h file to avoid circular dependency.
Closes https://github.com/facebook/rocksdb/pull/1520
Differential Revision: D4187179
Pulled By: ajkr
fbshipit-source-id: 10c2103
Summary:
Remove the ticker count because:
* Having to reset the ticker count in WriteImpl is ineffiecent;
* It doesn't make sense to have it as a ticker count if multiple db
instance share a statistics object.
Closes https://github.com/facebook/rocksdb/pull/1531
Differential Revision: D4194442
Pulled By: yiwu-arbug
fbshipit-source-id: e2110a9
Summary: A convience method to atomically get and reset ticker count. I'm wanting to use it to have a thin wrapper to the statistics object to export ticker counts to ODS for LogDevice (since they don't even use fb303).
Test Plan:
test in LogDevice shadow cluster.
https://fburl.com/461868822
Reviewers: andrewkr, yhchiang, IslamAbdelRahman
Reviewed By: IslamAbdelRahman
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D64869
* Fix StatsLevel so that kExceptTimeForMutex leaves compression stats enabled and kExceptDetailedTimers disables mutex lock stats. Also change default stats level to kExceptDetailedTimers (disabling both compression and mutex timing).
* Changed order of StatsLevel enum to simplify logic for determining what stats to record.
Summary:
Add ReadOptions::read_amp_bytes_per_bit option which allow us to create a bitmap for every data block we read
the bitmap will contain (block_size / read_amp_bytes_per_bit) bits.
We will use this bitmap to mark which bytes have been used of the block so we can calculate the read amplification
Test Plan: added new tests
Reviewers: andrewkr, yhchiang, sdong
Reviewed By: sdong
Subscribers: yiwu, leveldb, march, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D58707
Summary:
add SIM_BLOCK_CACHE_HIT and SIM_BLOCK_CACHE_MISS tickers.
maybe can be combined with Histograms like DB_GET to evaluate the current setting of the size of block cache.
Test Plan: make all check
Reviewers: sdong, andrewkr, IslamAbdelRahman, yiwu
Reviewed By: yiwu
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D61803
Summary:
TickersNameMap is not consistent with Tickers enum.
this cause us to report wrong statistics and sometimes to access TickersNameMap outside it's boundary causing crashes (in Fb303 statistics)
Test Plan: added new unit test
Reviewers: sdong, kradhakrishnan
Reviewed By: kradhakrishnan
Subscribers: andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D61083
* Added new statistics and refactored to allow ioptions to be passed around as required to access environment and statistics pointers (and, as a convenient side effect, info_log pointer).
* Prevent incrementing compression counter when compression is turned off in options.
* Prevent incrementing compression counter when compression is turned off in options.
* Added two more supported compression types to test code in db_test.cc
* Prevent incrementing compression counter when compression is turned off in options.
* Added new StatsLevel that excludes compression timing.
* Fixed casting error in coding.h
* Fixed CompressionStatsTest for new StatsLevel.
* Removed unused variable that was breaking the Linux build
Summary: With `table_options.cache_index_and_filter_blocks = true`, index and filter blocks are stored in block cache. Then people are curious how much of the block cache total size is used by indexes and bloom filters. It will be nice we have a way to report that. It can help people tune performance and plan for optimized hardware setting. We add several enum values for db Statistics. BLOCK_CACHE_INDEX/FILTER_BYTES_INSERT - BLOCK_CACHE_INDEX/FILTER_BYTES_ERASE = current INDEX/FILTER total block size in bytes.
Test Plan:
write a test case called `DBBlockCacheTest.IndexAndFilterBlocksStats`. The result is:
```
[gzh@dev9927.prn1 ~/local/rocksdb] make db_block_cache_test -j64 && ./db_block_cache_test --gtest_filter=DBBlockCacheTest.IndexAndFilterBlocksStats
Makefile:101: Warning: Compiling in debug mode. Don't use the resulting binary in production
GEN util/build_version.cc
make: `db_block_cache_test' is up to date.
Note: Google Test filter = DBBlockCacheTest.IndexAndFilterBlocksStats
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from DBBlockCacheTest
[ RUN ] DBBlockCacheTest.IndexAndFilterBlocksStats
[ OK ] DBBlockCacheTest.IndexAndFilterBlocksStats (689 ms)
[----------] 1 test from DBBlockCacheTest (689 ms total)
[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (689 ms total)
[ PASSED ] 1 test.
```
Reviewers: IslamAbdelRahman, andrewkr, sdong
Reviewed By: sdong
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D58677
Summary:
Added a new abstraction to cache page to RocksDB designed for the read
cache use.
RocksDB current block cache is more of an object cache. For the persistent read cache
project, what we need is a page cache equivalent. This changes adds a cache
abstraction to RocksDB to cache pages called PersistentCache. PersistentCache can cache
uncompressed pages or raw pages (content as in filesystem). The user can
choose to operate PersistentCache either in COMPRESSED or UNCOMPRESSED mode.
Blame Rev:
Test Plan: Run unit tests
Reviewers: sdong
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D55707
Summary:
Cache to have an option to fail Cache::Insert() when full. Update call sites to check status and handle error.
I totally have no idea what's correct behavior of all the call sites when they encounter error. Please let me know if you see something wrong or more unit test is needed.
Test Plan: make check -j32, see tests pass.
Reviewers: anthony, yhchiang, andrewkr, IslamAbdelRahman, kradhakrishnan, sdong
Reviewed By: sdong
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D54705
Summary: Measuring mutex duration will measure time inside DB mutex, which breaks our best practice. Add a stat level in Statistics class. By default, disable to measure the mutex operations.
Test Plan: Add a unit test to make sure it is off by default.
Reviewers: rven, anthony, IslamAbdelRahman, kradhakrishnan, andrewkr, yhchiang
Reviewed By: yhchiang
Subscribers: MarkCallaghan, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D53367
Summary:
This diff adds support for concurrent adds to the skiplist memtable
implementations. Memory allocation is made thread-safe by the addition of
a spinlock, with small per-core buffers to avoid contention. Concurrent
memtable writes are made via an additional method and don't impose a
performance overhead on the non-concurrent case, so parallelism can be
selected on a per-batch basis.
Write thread synchronization is an increasing bottleneck for higher levels
of concurrency, so this diff adds --enable_write_thread_adaptive_yield
(default off). This feature causes threads joining a write batch
group to spin for a short time (default 100 usec) using sched_yield,
rather than going to sleep on a mutex. If the timing of the yield calls
indicates that another thread has actually run during the yield then
spinning is avoided. This option improves performance for concurrent
situations even without parallel adds, although it has the potential to
increase CPU usage (and the heuristic adaptation is not yet mature).
Parallel writes are not currently compatible with
inplace updates, update callbacks, or delete filtering.
Enable it with --allow_concurrent_memtable_write (and
--enable_write_thread_adaptive_yield). Parallel memtable writes
are performance neutral when there is no actual parallelism, and in
my experiments (SSD server-class Linux and varying contention and key
sizes for fillrandom) they are always a performance win when there is
more than one thread.
Statistics are updated earlier in the write path, dropping the number
of DB mutex acquisitions from 2 to 1 for almost all cases.
This diff was motivated and inspired by Yahoo's cLSM work. It is more
conservative than cLSM: RocksDB's write batch group leader role is
preserved (along with all of the existing flush and write throttling
logic) and concurrent writers are blocked until all memtable insertions
have completed and the sequence number has been advanced, to preserve
linearizability.
My test config is "db_bench -benchmarks=fillrandom -threads=$T
-batch_size=1 -memtablerep=skip_list -value_size=100 --num=1000000/$T
-level0_slowdown_writes_trigger=9999 -level0_stop_writes_trigger=9999
-disable_auto_compactions --max_write_buffer_number=8
-max_background_flushes=8 --disable_wal --write_buffer_size=160000000
--block_size=16384 --allow_concurrent_memtable_write" on a two-socket
Xeon E5-2660 @ 2.2Ghz with lots of memory and an SSD hard drive. With 1
thread I get ~440Kops/sec. Peak performance for 1 socket (numactl
-N1) is slightly more than 1Mops/sec, at 16 threads. Peak performance
across both sockets happens at 30 threads, and is ~900Kops/sec, although
with fewer threads there is less performance loss when the system has
background work.
Test Plan:
1. concurrent stress tests for InlineSkipList and DynamicBloom
2. make clean; make check
3. make clean; DISABLE_JEMALLOC=1 make valgrind_check; valgrind db_bench
4. make clean; COMPILE_WITH_TSAN=1 make all check; db_bench
5. make clean; COMPILE_WITH_ASAN=1 make all check; db_bench
6. make clean; OPT=-DROCKSDB_LITE make check
7. verify no perf regressions when disabled
Reviewers: igor, sdong
Reviewed By: sdong
Subscribers: MarkCallaghan, IslamAbdelRahman, anthony, yhchiang, rven, sdong, guyg8, kradhakrishnan, dhruba
Differential Revision: https://reviews.facebook.net/D50589
Summary: Add 2 new counters BLOCK_CACHE_BYTES_WRITE, BLOCK_CACHE_BYTES_READ to keep track of how many bytes were written to the cache and how many bytes that we read from cache
Test Plan: make check
Reviewers: sdong, yhchiang, igor
Reviewed By: igor
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D48195
Summary:
Although there are currently counters to keep track of the
stall caused by having too many L0 files, there is no distinction as
to whether when that stall occurs either (A) L0-L1 compaction is taking
place to try and mitigate it, or (B) no L0-L1 compaction has been scheduled
at the moment. This diff adds a counter for (A) so that the nature of L0
stalls can be better understood.
Test Plan: make all && make check
Reviewers: sdong, igor, anthony, noetzli, yhchiang
Reviewed By: yhchiang
Subscribers: MarkCallaghan, dhruba
Differential Revision: https://reviews.facebook.net/D46749
Summary:
There are currently no statistics on seeks, only on gets. This adds the following counters:
rocksdb.number.db.seek
rocksdb.number.db.next
rocksdb.number.db.prev
(number of calls)
rocksdb.db.iterate.bytes.read
(number of bytes read from key + value using seek/next/prev)
rocksdb.number.keys.seek.found
rocksdb.number.keys.next.found
rocksdb.number.keys.prev.found
(number of calls where seek/next/prev found a value)
Test Plan:
./db_bench -statistics -benchmarks fillrandom,seekrandom -seek_nexts 5
./db_bench -statistics -benchmarks fillrandom,seekrandom -seek_nexts 5 -reverse_iterator
Reviewers: yhchiang, rven, kradhakrishnan, IslamAbdelRahman, MarkCallaghan, sdong, igor
Reviewed By: sdong
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D46605
Summary:
Up to this point, the subcompactions that make up a compaction
job have been divided based on the key range of the L1 files, and each
subcompaction has handled the key range of only one file. However
DBOption.max_subcompactions allows the user to designate how many
subcompactions at most to perform. This patch updates the
CompactionJob::GetSubcompactionBoundaries() to determine these
divisions accordingly based on that option and other input/system factors.
The current approach orders the starting and/or ending keys of certain
compaction input files and then generates a histogram to approximate the
size covered by the key range between each consecutive pair of keys. Then
it groups these ranges into groups so that the sizes are approximately equal
to one another. The approach has also been adapted to work for universal
compaction as well instead of just for level-based compaction as it was before.
These subcompactions are then executed in parallel by locally spawning
threads, one for each. The results are then aggregated and the compaction
completed.
Test Plan: make all && make check
Reviewers: yhchiang, anthony, igor, noetzli, sdong
Reviewed By: sdong
Subscribers: MarkCallaghan, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43269
Summary: Measure read latency histogram and put in statistics. Compaction inputs are excluded from it when possible (unfortunately usually no possible as we usually take table reader from table cache.
Test Plan:
Run db_bench and it shows the stats, like:
rocksdb.sst.read.micros statistics Percentiles :=> 50 : 1.238522 95 : 2.529740 99 : 3.912180
Reviewers: kradhakrishnan, rven, anthony, IslamAbdelRahman, MarkCallaghan, yhchiang
Reviewed By: yhchiang
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D43275
Summary:
In one of our recent meetings, we discussed deprecating features that are not being actively used. One of those features, at least within Facebook, is timeout_hint. The feature is really nicely implemented, but if nobody needs it, we should remove it from our code-base (until we get a valid use-case). Some arguments:
* Less code == better icache hit rate, smaller builds, simpler code
* The motivation for adding timeout_hint_us was to work-around RocksDB's stall issue. However, we're currently addressing the stall issue itself (see @sdong's recent work on stall write_rate), so we should never see sharp lock-ups in the future.
* Nobody is using the feature within Facebook's code-base. Googling for `timeout_hint_us` also doesn't yield any users.
Test Plan: make check
Reviewers: anthony, kradhakrishnan, sdong, yhchiang
Reviewed By: yhchiang
Subscribers: sdong, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D41937
Summary:
BYTES_READ only count the number of logical bytes read from
the DB::Get() function. It neither includes all logical bytes read
nor indicates IO read bytes.
This patch improves the comment for BYTES_READ.
Test Plan: Only change comment.
Reviewers: sdong, rven, anthony, kradhakrishnan, IslamAbdelRahman, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D40599
Summary:
Implementation of a table-level row cache.
It only caches point queries done through the `DB::Get` interface, queries done through the `Iterator` interface will completely skip the cache.
Supports snapshots and merge operations.
Test Plan: Ran `make valgrind_check commit-prereq`
Reviewers: igor, philipp, sdong
Reviewed By: sdong
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D39849
Summary:
We have addded new stats and perf_context for measuring the merge and filter operation time consumption.
We have bounded all the merge operations within the GUARD statment and collected the total time for these operations in the DB.
Test Plan: WIP
Reviewers: rven, yhchiang, kradhakrishnan, igor, sdong
Reviewed By: sdong
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D34377
Summary:
- In statistics.h , added tickers.
- In version_set.cc,
-- Added a getter method for hit_file_level_ in the class FilePicker
-- Added a line in the Get() method in case of a found, increment the corresponding counters based on the level of the file respectively.
Corresponding task: https://our.intern.facebook.com/intern/tasks/?s=506100481&t=5952818
Personal fork: 0c3f2e3600
Test Plan:
In terminal,
```
make -j32 db_test
ROCKSDB_TESTS=L0L1L2AndUpHitCounter ./db_test
```
Or to use debugger,
```
make -j32 db_test
export ROCKSDB_TESTS=L0L1L2AndUpHitCounter
gdb db_test
```
Reviewers: rven, sdong
Reviewed By: sdong
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D32205
Summary:
Add a counter for collecting the wait time on db mutex.
Also add MutexWrapper and CondVarWrapper for measuring wait time.
Test Plan:
./db_test
export ROCKSDB_TESTS=MutexWaitStats
./db_test
verify stats output using db_bench
make clean
make release
./db_bench --statistics=1 --benchmarks=fillseq,readwhilewriting --num=10000 --threads=10
Sample output:
rocksdb.db.mutex.wait.micros COUNT : 7546866
Reviewers: MarkCallaghan, rven, sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D32787
Summary:
Introducing WriteController, which is a source of truth about per-DB write delays. Let's define an DB epoch as a period where there are no flushes and compactions (i.e. new epoch is started when flush or compaction finishes). Each epoch can either:
* proceed with all writes without delay
* delay all writes by fixed time
* stop all writes
The three modes are recomputed at each epoch change (flush, compaction), rather than on every write (which is currently the case).
When we have a lot of column families, our current pull behavior adds a big overhead, since we need to loop over every column family for every write. With new push model, overhead on Write code-path is minimal.
This is just the start. Next step is to also take care of stalls introduced by slow memtable flushes. The final goal is to eliminate function MakeRoomForWrite(), which currently needs to be called for every column family by every write.
Test Plan: make check for now. I'll add some unit tests later. Also, perf test.
Reviewers: dhruba, yhchiang, MarkCallaghan, sdong, ljin
Reviewed By: ljin
Subscribers: leveldb
Differential Revision: https://reviews.facebook.net/D22791
Summary:
1. Make filter_block.h a base class. Derive block_based_filter_block and full_filter_block. The previous one is the traditional filter block. The full_filter_block is newly added. It would generate a filter block that contain all the keys in SST file.
2. When querying a key, table would first check if full_filter is available. If not, it would go to the exact data block and check using block_based filter.
3. User could choose to use full_filter or tradional(block_based_filter). They would be stored in SST file with different meta index name. "filter.filter_policy" or "full_filter.filter_policy". Then, Table reader is able to know the fllter block type.
4. Some optimizations have been done for full_filter_block, thus it requires a different interface compared to the original one in filter_policy.h.
5. Actual implementation of filter bits coding/decoding is placed in util/bloom_impl.cc
Benchmark: base commit 1d23b5c470
Command:
db_bench --db=/dev/shm/rocksdb --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --write_buffer_size=134217728 --max_write_buffer_number=2 --target_file_size_base=33554432 --max_bytes_for_level_base=1073741824 --verify_checksum=false --max_background_compactions=4 --use_plain_table=0 --memtablerep=prefix_hash --open_files=-1 --mmap_read=1 --mmap_write=0 --bloom_bits=10 --bloom_locality=1 --memtable_bloom_bits=500000 --compression_type=lz4 --num=393216000 --use_hash_search=1 --block_size=1024 --block_restart_interval=16 --use_existing_db=1 --threads=1 --benchmarks=readrandom —disable_auto_compactions=1
Read QPS increase for about 30% from 2230002 to 2991411.
Test Plan:
make all check
valgrind db_test
db_stress --use_block_based_filter = 0
./auto_sanity_test.sh
Reviewers: igor, yhchiang, ljin, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D20979
Summary: as title
Test Plan: make release
Reviewers: sdong, yhchiang
Reviewed By: yhchiang
Subscribers: leveldb
Differential Revision: https://reviews.facebook.net/D21717
Summary: as title
Test Plan: make release
Reviewers: yhchiang, sdong
Reviewed By: sdong
Subscribers: leveldb
Differential Revision: https://reviews.facebook.net/D21549
Summary:
Make StatisticsImpl being able to forward stats to provided statistics
implementation. The main purpose is to allow us to collect internal
stats in the future even when user supplies custom statistics
implementation. It avoids intrumenting 2 sets of stats collection code.
One immediate use case is tuning advisor, which needs to collect some
internal stats, users may not be interested.
Test Plan:
ran db_bench and see stats show up at the end of run
Will run make all check since some tests rely on statistics
Reviewers: yhchiang, sdong, igor
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D20145
Summary:
This diff allows the I/O stats about Flush and Compaction to be reported
in a more accurate way. Instead of measuring the size of a file, it
measure I/O cost in per read / write basis.
Test Plan: make all check
Reviewers: sdong, igor, ljin
Reviewed By: ljin
Subscribers: leveldb
Differential Revision: https://reviews.facebook.net/D19383
Summary:
This diff adds timeout_hint_us to WriteOptions. If it's non-zero, then
1) writes associated with this options MAY be aborted when it has been
waiting for longer than the specified time. If an abortion happens,
associated writes will return Status::TimeOut.
2) the stall time of the associated write caused by flush or compaction
will be limited by timeout_hint_us.
The default value of timeout_hint_us is 0 (i.e., OFF.)
The statistics of timeout writes will be recorded in WRITE_TIMEDOUT.
Test Plan:
export ROCKSDB_TESTS=WriteTimeoutAndDelayTest
make db_test
./db_test
Reviewers: igor, ljin, haobo, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D18837
Summary: This is a temporary solution to a issue that we have with compression libraries. See task #4453446.
Test Plan: make check doesn't complain :)
Reviewers: haobo, ljin, yhchiang, dhruba, sdong
Reviewed By: sdong
Subscribers: leveldb
Differential Revision: https://reviews.facebook.net/D18975
Summary: KSVObsolete is no longer nullptr and needs to be checked explicitly. Also did some minor code cleanup and added a stat counter to track superversion cleanups incurred in the foreground.
Test Plan: make check
Reviewers: ljin
Reviewed By: ljin
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16701
Summary:
Add a check at the end of GetImpl to release SuperVersion if it becomes
obsolete. Also do Scrape() inside InstallSuperVersion so it happens more
frequent.
Test Plan:
make all check
running asan_check now
Reviewers: igor, haobo, sdong, dhruba
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16641
Summary: as title
Test Plan:
asan_check
will post results later
Reviewers: haobo, igor, dhruba, sdong
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16257
Summary: Provide a public API for users to access the table properties for each SSTable.
Test Plan: Added a unit tests to test the function correctness under differnet conditions.
Reviewers: haobo, dhruba, sdong
Reviewed By: haobo
CC: leveldb
Differential Revision: https://reviews.facebook.net/D16083
Summary: I'm separating code-cleanup part of https://reviews.facebook.net/D14517. This will make D14517 easier to understand and this diff easier to review.
Test Plan: make check
Reviewers: haobo, kailiu, sdong, dhruba, tnovak
Reviewed By: tnovak
CC: leveldb
Differential Revision: https://reviews.facebook.net/D15099
Summary: liveness of the statistics object is already ensured by the shared pointer in DB options. There's no reason to pass again shared pointer among internal functions. Raw pointer is sufficient and efficient.
Test Plan: make check
Reviewers: dhruba, MarkCallaghan, igor
Reviewed By: dhruba
CC: leveldb, reconnect.grayhat
Differential Revision: https://reviews.facebook.net/D14289
Summary: This diff leverage the existing block cache and extend it to cache index/filter block.
Test Plan:
Added new tests in db_test and table_test
The correctness is checked by:
1. make check
2. make valgrind_check
Performance is test by:
1. 10 times of build_tools/regression_build_test.sh on two versions of rocksdb before/after the code change. Test results suggests no significant difference between them. For the two key operatons `overwrite` and `readrandom`, the average iops are both 20k and ~260k, with very small variance).
2. db_stress.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, haobo, xjin
Differential Revision: https://reviews.facebook.net/D13167
Andrew Kryczka