Summary:
We occasionally get write stalls (>1s Write() calls) on HDD under read load. The following timers explain almost all of the stalls:
- perf_context.db_mutex_lock_nanos
- perf_context.db_condition_wait_nanos
- iostats_context.open_time
- iostats_context.allocate_time
- iostats_context.write_time
- iostats_context.range_sync_time
- iostats_context.logger_time
In my experiments each of these occasionally takes >1s on write path under some workload. There are rare cases when Write() takes long but none of these takes long.
Test Plan: Added code to our application to write the listed timings to log for slow writes. They usually add up to almost exactly the time Write() call took.
Reviewers: rven, yhchiang, sdong
Reviewed By: sdong
Subscribers: march, dhruba, tnovak
Differential Revision: https://reviews.facebook.net/D39177
Summary: In DB::CompactRange(), change parameter "reduce_level" to "change_level". Users can compact all data to the last level if needed. By doing it, users can migrate the DB to options.level_compaction_dynamic_level_bytes=true.
Test Plan: Add a unit test for it.
Reviewers: yhchiang, anthony, kradhakrishnan, igor, rven
Reviewed By: rven
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D39099
Summary:
DBImpl::notifying_events_ is a internal counter in DBImpl which is
used to prevent DB close when DB is notifying events. However, as
the current events all rely on either compaction or flush which
already have similar counters to prevent DB close, it is safe to
remove notifying_events_.
Test Plan:
listener_test
examples/compact_files_example
Reviewers: igor, anthony, kradhakrishnan, rven, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D39315
Summary: Broken by optimistic transaction diff. (I only built 'release' not 'static_lib' when testing).
Test Plan: build
Reviewers: yhchiang, sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D39219
Summary: Optimistic transactions supporting begin/commit/rollback semantics. Currently relies on checking the memtable to determine if there are any collisions at commit time. Not yet implemented would be a way of enuring the memtable has some minimum amount of history so that we won't fail to commit when the memtable is empty. You should probably start with transaction.h to get an overview of what is currently supported.
Test Plan: Added a new test, but still need to look into stress testing.
Reviewers: yhchiang, igor, rven, sdong
Reviewed By: sdong
Subscribers: adamretter, MarkCallaghan, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D33435
Summary:
For transactions, we are using the memtables to validate that there are no write conflicts. But after flushing, we don't have any memtables, and transactions could fail to commit. So we want to someone keep around some extra history to use for conflict checking. In addition, we want to provide a way to increase the size of this history if too many transactions fail to commit.
After chatting with people, it seems like everyone prefers just using Memtables to store this history (instead of a separate history structure). It seems like the best place for this is abstracted inside the memtable_list. I decide to create a separate list in MemtableListVersion as using the same list complicated the flush/installalflushresults logic too much.
This diff adds a new parameter to control how much memtable history to keep around after flushing. However, it sounds like people aren't too fond of adding new parameters. So I am making the default size of flushed+not-flushed memtables be set to max_write_buffers. This should not change the maximum amount of memory used, but make it more likely we're using closer the the limit. (We are now postponing deleting flushed memtables until the max_write_buffer limit is reached). So while we might use more memory on average, we are still obeying the limit set (and you could argue it's better to go ahead and use up memory now instead of waiting for a write stall to happen to test this limit).
However, if people are opposed to this default behavior, we can easily set it to 0 and require this parameter be set in order to use transactions.
Test Plan: Added a xfunc test to play around with setting different values of this parameter in all tests. Added testing in memtablelist_test and planning on adding more testing here.
Reviewers: sdong, rven, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D37443
Summary:
Rename EventLoggerHelpers EventHelpers, as it's going to include
all event-related helper functions instead of EventLogger only stuffs.
Test Plan: make
Reviewers: sdong, rven, anthony
Reviewed By: anthony
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D39093
Summary:
Compaction now boosts the size of deletion entries of a file only when
the number of deletion entries is greater than the number of non-deletion
entries in the file. The motivation here is that in a stable workload,
the number of deletion entries should be roughly equal to the number of
non-deletion entries. If we compensate the size of deletion entries in a
stable workload, the deletion compensation logic might introduce unwanted
effet which changes the shape of LSM tree.
Test Plan: db_test --gtest_filter="*Deletion*"
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38703
Summary:
Fixed a missing "}" at the end of the generated JSON Log
in EventLoggerHelpers::LogTableFileCreation.
Test Plan: db_bench
Reviewers: igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38919
Summary: Removed an unused private variable in db_impl.h
Test Plan: make db_test
Reviewers: sdong, anthony, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38925
Summary: We have a bug where we don't report the last level's files as being compacted. This fixes it.
Test Plan: See the fix in action here: https://phabricator.fb.com/P19845738
Reviewers: MarkCallaghan, sdong
Reviewed By: sdong
Subscribers: yhchiang, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38727
Summary:
This patch fixes the following two bugs on logging file deletion.
1. Previously, file deletion failure was only logged in INFO_LEVEL.
This patch changes it to ERROR_LEVEL and does some code clean.
2. EventLogger previously will always generate the same log on
table file deletion even when file deletion is not successful.
Now the resulting status of file deletion will also be logged.
Test Plan: make all check
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38817
Summary: Ensure ColumnFamilyOptions.num_levels >= 2 when level compaction is used.
Test Plan: Extend SanitizeOptions test in column_family_test
Reviewers: sdong, rven, anthony, krishnanm86, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38829
Summary: Avoid logging under mutex in DBImpl::WriteLevel0TableForRecovery().
Test Plan: make all check
Reviewers: igor, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38823
Summary:
Allow EventLogger to directly log from a JSONWriter. This allows
the JSONWriter to be shared by EventLogger and potentially EventListener,
which is an important step to integrate EventLogger and EventListener.
This patch also rewrites EventLoggerHelpers::LogTableFileCreation(),
which uses the new API to generate identical log.
Test Plan:
Run db_bench in debug mode and make sure the log is correct and no
assertions fail.
Reviewers: sdong, anthony, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38709
Summary:
This turns out to be pretty bad because if we prioritize L0->L1 then L1 can grow artificially large, which makes L0->L1 more and more expensive. For example:
256MB @ L0 + 256MB @ L1 --> 512MB @ L1
256MB @ L0 + 512MB @ L1 --> 768MB @ L1
256MB @ L0 + 768MB @ L1 --> 1GB @ L1
....
256MB @ L0 + 10GB @ L1 --> 10.2GB @ L1
At some point we need to start compacting L1->L2 to speed up L0->L1.
Test Plan:
The performance improvement is massive for heavy write workload. This is the benchmark I ran: https://phabricator.fb.com/P19842671. Before this change, the benchmark took 47 minutes to complete. After, the benchmark finished in 2minutes. You can see full results here: https://phabricator.fb.com/P19842674
Also, we ran this diff on MongoDB on RocksDB on one replicaset. Before the change, our initial sync was so slow that it couldn't keep up with primary writes. After the change, the import finished without any issues
Reviewers: dynamike, MarkCallaghan, rven, yhchiang, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38637
Summary: Dump db stats in WARN level
Test Plan: run db_bench and verify the LOG
Reviewers: igor, MarkCallaghan
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38691
Summary:
See https://gist.github.com/mdcallag/89ebb2b8cbd331854865 for the IO stats.
I added "Cumulative compaction:" and "Interval compaction:" lines. The IO rates
can be confusing. Rates fro per-level stats lines, Wr(MB/s) & Rd(MB/s), are computed
using the duration of the compaction job. If the job reads 10MB, writes 9MB and the job
(IO & merging) takes 1 second then the rates are 10MB/s for read and 9MB/s for writes.
The IO rates in the Cumulative compaction line uses the total uptime. The IO rates in the
Interval compaction line uses the interval uptime. So these Cumalative & Interval
compaction IO rates cannot be compared to the per-level IO rates. But both forms of
the rates are useful for debugging perf.
Task ID: #
Blame Rev:
Test Plan:
run db_bench
Revert Plan:
Database Impact:
Memcache Impact:
Other Notes:
EImportant:
- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -
Reviewers: igor
Reviewed By: igor
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D38667
Summary: Not sure why this fails on some compilers and doesn't on others.
Test Plan: none
Reviewers: meyering, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38673
Summary:
sync_file_range is not always asyncronous and thus can block writes if we do this for WAL in the foreground thread. See more here: http://yoshinorimatsunobu.blogspot.com/2014/03/how-syncfilerange-really-works.html
Some users don't want us to call sync_file_range on WALs. Some other do.
Thus, I'm adding a separate option wal_bytes_per_sync to control calling
sync_file_range on WAL files. bytes_per_sync will apply only to table
files now.
Test Plan: no more sync_file_range for WAL as evidenced by strace
Reviewers: yhchiang, rven, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38253
Summary: As title. I spent some time thinking about it and I don't think there should be any issue with running manual compaction and flushes in parallel
Test Plan: make check works
Reviewers: rven, yhchiang, sdong
Reviewed By: yhchiang, sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38355
Summary: DBTest.DynamicLevelMaxBytesCompactRange needs to make sure L0 is not empty to properly cover the code paths we want to cover. However, current codes have a bug that might leave the condition not held. Improve the test to ensure it.
Test Plan: Run the test in an environment that is used to fail. Also run it many times.
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D38631
Summary: CompactRange() now is much more expensive for dynamic level base size as it goes through all the levels. Skip those not used levels between level 0 an base level.
Test Plan: Run all unit tests
Reviewers: yhchiang, rven, anthony, kradhakrishnan, igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D37125
Summary:
Allow GetThreadList to report Flush properties, which includes:
* job id
* number of bytes that has been written since flush started.
* total size of input mem-tables
Test Plan:
./db_bench --threads=30 --num=1000000 --benchmarks=fillrandom --thread_status_per_interval=100 --value_size=1000
Sample output from db_bench which tracks same flush job
ThreadID ThreadType cfName Operation ElapsedTime Stage State OperationProperties
140213879898240 High Pri default Flush 5789 us FlushJob::WriteLevel0Table BytesMemtables 4112835 | BytesWritten 577104 | JobID 8 |
ThreadID ThreadType cfName Operation ElapsedTime Stage State OperationProperties
140213879898240 High Pri default Flush 30.634 ms FlushJob::WriteLevel0Table BytesMemtables 4112835 | BytesWritten 1734865 | JobID 8 |
Reviewers: rven, igor, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38505
Summary:
When trying to compact entire database with SuggestCompactRange(), we'll first try the left-most files. This is pretty bad, because:
1) the left part of LSM tree will be overly compacted, but right part will not be touched
2) First compaction will pick up the left-most file. Second compaction will try to pick up next left-most, but this will not be possible, because there's a big chance that second's file range on N+1 level is already being compacted.
I observe both of those problems when running Mongo+RocksDB and trying to compact the DB to clean up tombstones. I'm unable to clean them up :(
This diff adds a bit of randomness into choosing a file. First, it chooses a file at random and tries to compact that one. This should solve both problems specified here.
Test Plan: make check
Reviewers: yhchiang, rven, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38379
Summary: DBTest.DynamicLevelMaxBytesBase2 has a check that is not necessary and may fail. Remove it, and add two unrelated check.
Test Plan: Run the test
Reviewers: yhchiang, rven, kradhakrishnan, anthony, igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D38457
Summary: Universal compactions with multiple levels should use file preallocation size based on file size if output level is not level 0
Test Plan: Run all tests.
Reviewers: igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D38439
Summary: Add --rate_limiter_bytes_per_sec to db_bench to allow rater limit to disk
Test Plan:
Run
./db_bench --benchmarks=fillseq --num=30000000 --rate_limiter_bytes_per_sec=3000000 --num_multi_db=8 -disable_wal
And see io_stats to have the rate limited.
Reviewers: yhchiang, rven, anthony, kradhakrishnan, igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D38385
Summary:
Fixed the following compile error in db/column_family.cc
db/column_family.cc:633:33: error: ‘ASSERT_GT’ was not declared in this scope
16:14:45 ASSERT_GT(listeners.size(), 0U);
Test Plan: make db_test
Reviewers: igor, sdong, rven
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38367
Summary:
Fixed a bug in EventListener::OnCompactionCompleted() that returns
incorrect list of input / output file names.
Test Plan: Extend existing test in listener_test.cc
Reviewers: sdong, rven, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38349
Summary: This caused a crash of our MongoDB + RocksDB instance. PickCompactionBySize() sets its own parent_index. We never reset this parent_index when picking PickFilesMarkedForCompactionExperimental(). So we might end up doing SetupOtherInputs() with parent_index that was set by PickCompactionBySize, although we're using compaction calculated using PickFilesMarkedForCompactionExperimental.
Test Plan: Added a unit test that fails with assertion on master.
Reviewers: yhchiang, rven, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38337
Summary:
Added a couple functions to WriteBatchWithIndex to make it easier to query the value of a key including reading pending writes from a batch. (This is needed for transactions).
I created write_batch_with_index_internal.h to use to store an internal-only helper function since there wasn't a good place in the existing class hierarchy to store this function (and it didn't seem right to stick this function inside WriteBatchInternal::Rep).
Since I needed to access the WriteBatchEntryComparator, I moved some helper classes from write_batch_with_index.cc into write_batch_with_index_internal.h/.cc. WriteBatchIndexEntry, ReadableWriteBatch, and WriteBatchEntryComparator are all unchanged (just moved to a different file(s)).
Test Plan: Added new unit tests.
Reviewers: rven, yhchiang, sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38037
Summary: In new clang we need to add override to every overriden function
Test Plan: none
Reviewers: rven
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D38259
Summary: When reporting compaction that was started because of SuggestCompactRange() we should treat it as manual compaction.
Test Plan: none
Reviewers: yhchiang, rven
Reviewed By: rven
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38139
Summary:
Couple changes:
1. instead of SnapshotList, just take a vector of snapshots
2. don't take a separate parameter is_snapshots_supported. If there are snapshots in the list, that means they are supported. I actually think we should get rid of this notion of snapshots not being supported.
3. don't pass in mutable_cf_options as a parameter. Lifetime of mutable_cf_options is a bit tricky to maintain, so it's better to not pass it in for the whole compaction job. We only really need it when we install the compaction results.
Test Plan: make check
Reviewers: sdong, rven, yhchiang
Reviewed By: yhchiang
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D36627
Summary:
Fixes#6840824, running "make check" on centos6 hits
a deadlock in column_family_test
Test Plan:
seq 10000 | parallel --gnu --eta 't=/dev/shm/rdb-{}; rm -rf
$t; mkdir $t && export TEST_TMPDIR=$t; ./column_family_test > $t/log-{}'
Made the test deterministic by narrrowing the window for the flush.
Reviewers: igor, meyering
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D38079
Summary: Optimize GetRange Function by checking the level of the files
Test Plan: pass make all check
Reviewers: rven, yhchiang, igor
Reviewed By: igor
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D37977
Summary:
[noticed a new warning when building with the very latest gcc]
* db/memtablerep_bench.cc (FLAGS_env): Remove declaration
of unused varaible, to avoid this warning/error:
db/memtablerep_bench.cc:135:22: error: ‘FLAGS_env’ defined but not\
used [-Werror=unused-variable]
static rocksdb::Env* FLAGS_env = rocksdb::Env::Default();
^
Test Plan: compile
Reviewers: ljin, rven, igor.sugak, yhchiang, sdong, igor
Reviewed By: igor
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D37983
Summary:
The default, use one iter for the whole test, isn't good. This cost me
a few hours of debugging and a few days of tessting. For readonly
that isn't realistic and for read-write that keeps a lot of old sst files around.
I remove the option because nothing uses it and not calling gettimeofday per
loop iteration adds about 3% to QPS at 20 threads.
Task ID: #
Blame Rev:
Test Plan:
run db_bench
Revert Plan:
Database Impact:
Memcache Impact:
Other Notes:
EImportant:
- begin *PUBLIC* platform impact section -
Bugzilla: #
- end platform impact -
Reviewers: igor
Reviewed By: igor
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D37965
Mike Kolupaev