Summary:
Throughput: 46k tps in our sysbench settings (filling the details later)
The idea is to have the simplest change that gives us a reasonable boost
in 2PC throughput.
Major design changes:
1. The WAL file internal buffer is not flushed after each write. Instead
it is flushed before critical operations (WAL copy via fs) or when
FlushWAL is called by MySQL. Flushing the WAL buffer is also protected
via mutex_.
2. Use two sequence numbers: last seq, and last seq for write. Last seq
is the last visible sequence number for reads. Last seq for write is the
next sequence number that should be used to write to WAL/memtable. This
allows to have a memtable write be in parallel to WAL writes.
3. BatchGroup is not used for writes. This means that we can have
parallel writers which changes a major assumption in the code base. To
accommodate for that i) allow only 1 WriteImpl that intends to write to
memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes
come via group commit phase which is serial anyway, ii) make all the
parts in the code base that assumed to be the only writer (via
EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are
protected via a stat_mutex_.
Note: the first commit has the approach figured out but is not clean.
Submitting the PR anyway to get the early feedback on the approach. If
we are ok with the approach I will go ahead with this updates:
0) Rebase with Yi's pipelining changes
1) Currently batching is disabled by default to make sure that it will be
consistent with all unit tests. Will make this optional via a config.
2) A couple of unit tests are disabled. They need to be updated with the
serial commit of 2PC taken into account.
3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires
releasing mutex_ beforehand (the same way EnterUnbatched does). This
needs to be cleaned up.
Closes https://github.com/facebook/rocksdb/pull/2345
Differential Revision: D5210732
Pulled By: maysamyabandeh
fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
Summary:
Some users want to prevent rocksdb from entering read-only mode in certain error cases. This diff gives them a callback, `OnBackgroundError`, that they can use to achieve it.
- call `OnBackgroundError` every time we consider setting `bg_error_`. Use its result to assign `bg_error_` but not to change the function's return status.
- classified calls using `BackgroundErrorReason` to give the callback some info about where the error happened
- renamed `ParanoidCheck` to something more specific so we can provide a clear `BackgroundErrorReason`
- unit tests for the most common cases: flush or compaction errors
Closes https://github.com/facebook/rocksdb/pull/2477
Differential Revision: D5300190
Pulled By: ajkr
fbshipit-source-id: a0ea4564249719b83428e3f4c6ca2c49e366e9b3
Summary:
If ReadOptions.low_pri=true and compaction is behind, the write will either return immediate or be slowed down based on ReadOptions.no_slowdown.
Closes https://github.com/facebook/rocksdb/pull/2369
Differential Revision: D5127619
Pulled By: siying
fbshipit-source-id: d30e1cff515890af0eff32dfb869d2e4c9545eb0
Summary:
- `max_background_flushes` and `max_background_compactions` are still supported for backwards compatibility
- `base_background_compactions` is completely deprecated. Now we just throttle to one background compaction when there's no pressure.
- `max_background_jobs` is added to automatically partition the concurrent background jobs into flushes vs compactions. Currently it's very simple as we just allocate one-fourth of the jobs to flushes, and the remaining can be used for compactions.
- The test cases that set `base_background_compactions > 1` needed to be updated. I just grab the pressure token such that the desired number of compactions can be scheduled.
Closes https://github.com/facebook/rocksdb/pull/2205
Differential Revision: D4937461
Pulled By: ajkr
fbshipit-source-id: df52cbbd497e13bbc9a60560a5ac2a2526b3f1f9
Summary:
PipelineWriteImpl is an alternative approach to WriteImpl. In WriteImpl, only one thread is allow to write at the same time. This thread will do both WAL and memtable writes for all write threads in the write group. Pending writers wait in queue until the current writer finishes. In the pipeline write approach, two queue is maintained: one WAL writer queue and one memtable writer queue. All writers (regardless of whether they need to write WAL) will still need to first join the WAL writer queue, and after the house keeping work and WAL writing, they will need to join memtable writer queue if needed. The benefit of this approach is that
1. Writers without memtable writes (e.g. the prepare phase of two phase commit) can exit write thread once WAL write is finish. They don't need to wait for memtable writes in case of group commit.
2. Pending writers only need to wait for previous WAL writer finish to be able to join the write thread, instead of wait also for previous memtable writes.
Merging #2056 and #2058 into this PR.
Closes https://github.com/facebook/rocksdb/pull/2286
Differential Revision: D5054606
Pulled By: yiwu-arbug
fbshipit-source-id: ee5b11efd19d3e39d6b7210937b11cefdd4d1c8d
Summary:
Adding DB::CreateColumnFamilie() and DB::DropColumnFamilies() to bulk create/drop column families. This is to address the problem creating/dropping 1k column families takes minutes. The bottleneck is we persist options files for every single column family create/drop, and it parses the persisted options file for verification, which take a lot CPU time.
The new APIs simply create/drop column families individually, and persist options file once at the end. This improves create 1k column families to within ~0.1s. Further improvement can be merge manifest write to one IO.
Closes https://github.com/facebook/rocksdb/pull/2248
Differential Revision: D5001578
Pulled By: yiwu-arbug
fbshipit-source-id: d4e00bda671451e0b314c13e12ad194b1704aa03
Summary:
Add a function to allow users to reset internal stats without restarting the DB.
Closes https://github.com/facebook/rocksdb/pull/2167
Differential Revision: D4907939
Pulled By: siying
fbshipit-source-id: ab2dd85b88aabe9380da7485320a1d460d3e1f68
Summary:
The concept about early exit in write thread implementation is a confusing one. It means that if early exit is allowed, batch group leader will not responsible to exit the batch group, but the last finished writer do. In case we need to mark log synced, or encounter memtable insert error, early exit is disallowed.
This patch remove such a concept by:
* In all cases, the last finished writer (not necessary leader) is responsible to exit batch group.
* In case of parallel memtable write, leader will also mark log synced after memtable insert and before signal finish (call `CompleteParallelWorker()`). The purpose is to allow mark log synced (which require locking mutex) can run in parallel to memtable insert in other writers.
* The last finish writer should handle memtable insert error (update bg_error_) before exiting batch group.
Closes https://github.com/facebook/rocksdb/pull/2134
Differential Revision: D4869667
Pulled By: yiwu-arbug
fbshipit-source-id: aec170847c85b90f4179d6a4608a4fe1361544e3
Summary:
Move some files under util/ to new directories env/, monitoring/ options/ and cache/
Closes https://github.com/facebook/rocksdb/pull/2090
Differential Revision: D4833681
Pulled By: siying
fbshipit-source-id: 2fd8bef
Summary:
db_impl.cc is too large to manage. Divide db_impl.cc into db/db_impl.cc, db/db_impl_compaction_flush.cc, db/db_impl_files.cc, db/db_impl_open.cc and db/db_impl_write.cc.
Closes https://github.com/facebook/rocksdb/pull/2095
Differential Revision: D4838188
Pulled By: siying
fbshipit-source-id: c5f3059
Summary:
Refactor WriteImpl() so when I plug-in the pipeline write code (which is
an alternative approach for WriteThread), some of the logic can be
reuse. I split out the following methods from WriteImpl():
* PreprocessWrite()
* HandleWALFull() (previous MaybeFlushColumnFamilies())
* HandleWriteBufferFull()
* WriteToWAL()
Also adding a constructor to WriteThread::Writer, and move WriteContext into db_impl.h.
No real logic change in this patch.
Closes https://github.com/facebook/rocksdb/pull/2042
Differential Revision: D4781014
Pulled By: yiwu-arbug
fbshipit-source-id: d45ca18
Summary:
Add two DB properties: rocksdb.actual_delayed_write_rate and rocksdb.is_write_stooped, for people to know whether current writes are being throttled.
Closes https://github.com/facebook/rocksdb/pull/2043
Differential Revision: D4782975
Pulled By: siying
fbshipit-source-id: 6b2f5cf
Summary:
PinnableSlice
Summary:
Currently the point lookup values are copied to a string provided by the
user. This incures an extra memcpy cost. This patch allows doing point lookup
via a PinnableSlice which pins the source memory location (instead of
copying their content) and releases them after the content is consumed
by the user. The old API of Get(string) is translated to the new API
underneath.
Here is the summary for improvements:
value 100 byte: 1.8% regular, 1.2% merge values
value 1k byte: 11.5% regular, 7.5% merge values
value 10k byte: 26% regular, 29.9% merge values
The improvement for merge could be more if we extend this approach to
pin the merge output and delay the full merge operation until the user
actually needs it. We have put that for future work.
PS:
Sometimes we observe a small decrease in performance when switching from
t5452014 to this patch but with the old Get(string) API. The d
Closes https://github.com/facebook/rocksdb/pull/1756
Differential Revision: D4391738
Pulled By: maysamyabandeh
fbshipit-source-id: 6f3edd3
Summary:
Seems to me `has_unpersisted_data_` is read from read thread and write
from write thread concurrently without synchronization. Making it an
atomic.
I update the logic not because seeing any problem with it, but it just
feel confusing.
Closes https://github.com/facebook/rocksdb/pull/1869
Differential Revision: D4555837
Pulled By: yiwu-arbug
fbshipit-source-id: eff2ab8
Summary:
Added method that returns approx num of entries as well as size for memtables.
Closes https://github.com/facebook/rocksdb/pull/1841
Differential Revision: D4511990
Pulled By: VitaliyLi
fbshipit-source-id: 9a4576e
Summary:
GetAndRefSuperVersionUnlocked
ReturnAndCleanupSuperVersionUnlocked
GetColumnFamilyHandleUnlocked
Are dead code that are not used any where
Closes https://github.com/facebook/rocksdb/pull/1802
Differential Revision: D4459948
Pulled By: IslamAbdelRahman
fbshipit-source-id: 30fa89d
Summary:
Added an option to GetApproximateSizes to exclude file stats, as MyRocks has those counted exactly and we need only stats from memtables.
Closes https://github.com/facebook/rocksdb/pull/1787
Differential Revision: D4441111
Pulled By: IslamAbdelRahman
fbshipit-source-id: c11f4c3
Summary:
Consider the following single column family scenario:
prepare in log A
commit in log B
*WAL is too large, flush all CFs to releast log A*
*CFA is on log B so we do not see CFA is depending on log A so no flush is requested*
To fix this we must also consider the log containing the prepare section when determining what log a CF is dependent on.
Closes https://github.com/facebook/rocksdb/pull/1768
Differential Revision: D4403265
Pulled By: reidHoruff
fbshipit-source-id: ce800ff
Summary:
Currently the point lookup values are copied to a string provided by the user.
This incures an extra memcpy cost. This patch allows doing point lookup
via a PinnableSlice which pins the source memory location (instead of
copying their content) and releases them after the content is consumed
by the user. The old API of Get(string) is translated to the new API
underneath.
Here is the summary for improvements:
1. value 100 byte: 1.8% regular, 1.2% merge values
2. value 1k byte: 11.5% regular, 7.5% merge values
3. value 10k byte: 26% regular, 29.9% merge values
The improvement for merge could be more if we extend this approach to
pin the merge output and delay the full merge operation until the user
actually needs it. We have put that for future work.
PS:
Sometimes we observe a small decrease in performance when switching from
t5452014 to this patch but with the old Get(string) API. The difference
is a little and could be noise. More importantly it is safely
cancelled
Closes https://github.com/facebook/rocksdb/pull/1732
Differential Revision: D4374613
Pulled By: maysamyabandeh
fbshipit-source-id: a077f1a
Summary:
If 2PC is enabled, checkpoint may not copy previous log files that contain uncommitted prepare records. In this diff we keep those files.
Closes https://github.com/facebook/rocksdb/pull/1724
Differential Revision: D4368319
Pulled By: siying
fbshipit-source-id: cc2c746
Summary:
Made delete_obsolete_files_period_micros option dynamic. It can be updating using DB::SetDBOptions().
Closes https://github.com/facebook/rocksdb/pull/1595
Differential Revision: D4246569
Pulled By: tonek
fbshipit-source-id: d23f560
Summary:
If the WriteOptions.no_slowdown flag is set AND we need to wait or sleep for
the write request, then fail immediately with Status::Incomplete().
Closes https://github.com/facebook/rocksdb/pull/1527
Differential Revision: D4191405
Pulled By: maysamyabandeh
fbshipit-source-id: 7f3ce3f
Summary:
Currently the compaction stats are printed to stdout. We want to export the compaction stats in a map format so that the upper layer apps (e.g., MySQL) could present
the stats in any format required by the them.
Closes https://github.com/facebook/rocksdb/pull/1477
Differential Revision: D4149836
Pulled By: maysamyabandeh
fbshipit-source-id: b3df19f
Summary:
Note: reviewed in https://reviews.facebook.net/D65115
- DBIter maintains a range tombstone accumulator. We don't cleanup obsolete tombstones yet, so if the user seeks back and forth, the same tombstones would be added to the accumulator multiple times.
- DBImpl::NewInternalIterator() (used to make DBIter's underlying iterator) adds memtable/L0 range tombstones, L1+ range tombstones are added on-demand during NewSecondaryIterator() (see D62205)
- DBIter uses ShouldDelete() when advancing to check whether keys are covered by range tombstones
Closes https://github.com/facebook/rocksdb/pull/1464
Differential Revision: D4131753
Pulled By: ajkr
fbshipit-source-id: be86559
Summary:
change ioptions.comparator to user_comparator instread of internal_comparator.
Also change Comparator* to InternalKeyComparator* to make its type explicitly.
Test Plan: make all check -j64
Reviewers: andrewkr, sdong, yiwu
Reviewed By: yiwu
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D65121
Summary:
Changes in the diff
API changes:
- Introduce IngestExternalFile to replace AddFile (I think this make the API more clear)
- Introduce IngestExternalFileOptions (This struct will encapsulate the options for ingesting the external file)
- Deprecate AddFile() API
Logic changes:
- If our file overlap with the memtable we will flush the memtable
- We will find the first level in the LSM tree that our file key range overlap with the keys in it
- We will find the lowest level in the LSM tree above the the level we found in step 2 that our file can fit in and ingest our file in it
- We will assign a global sequence number to our new file
- Remove AddFile restrictions by using global sequence numbers
Other changes:
- Refactor all AddFile logic to be encapsulated in ExternalSstFileIngestionJob
Test Plan:
unit tests (still need to add more)
addfile_stress (https://reviews.facebook.net/D65037)
Reviewers: yiwu, andrewkr, lightmark, yhchiang, sdong
Reviewed By: sdong
Subscribers: jkedgar, hcz, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D65061
Summary:
MyRocks hit a regression, @mung generated perf reports showing that the reason is the cost of calling `GetDBOptions()` inside `GetFromBatchAndDB()`
This diff avoid calling `GetDBOptions` and use the `ImmutableDBOptions` instead
Test Plan: make check -j64
Reviewers: sdong, yiwu
Reviewed By: yiwu
Subscribers: andrewkr, dhruba, mung
Differential Revision: https://reviews.facebook.net/D65151
Summary:
Issue scenario:
(1) We have 3 files in L1 and we issue a compaction that will compact them into 1 file in L2
(2) While compaction (1) is running, we flush a file into L0 and trigger another compaction that decide to move this file to L1 and then move it again to L2 (this file don't overlap with any other files)
(3) compaction (1) finishes and install the file it generated in L2, but this file overlap with the file we generated in (2) so we break the LSM consistency
Looks like this issue can be triggered by using non-exclusive manual compaction or AddFile()
Test Plan: unit tests
Reviewers: sdong
Reviewed By: sdong
Subscribers: hermanlee4, jkedgar, andrewkr, dhruba, yoshinorim
Differential Revision: https://reviews.facebook.net/D64947
Summary:
Fix the conflict bug between AddFile() and CompactRange() by
- Make sure that no AddFile calls are running when asking CompactionPicker to pick compaction for manual compaction
- If AddFile() run after we pick the compaction for the manual compaction it will be aware of it since we will add the manual compaction to running_compactions_ after picking it
This will solve these 2 scenarios
- If AddFile() is running, we will wait for it to finish before we pick a compaction for the manual compaction
- If we already picked a manual compaction and then AddFile() started ... we ensure that it never ingest a file in a level that will overlap with the manual compaction
Test Plan: unit tests
Reviewers: sdong
Reviewed By: sdong
Subscribers: andrewkr, yoshinorim, jkedgar, dhruba
Differential Revision: https://reviews.facebook.net/D64449
Summary:
Since AddFile unlock/lock the mutex inside LogAndApply() we need to ensure that during this period other compactions cannot run since such compactions are not aware of the file we are ingesting and could create a compaction that overlap wit this file
this diff add
- WaitForAddFile() call that will ensure that no AddFile() calls are being processed right now
- Call `WaitForAddFile()` in 3 locations
-- When doing manual Compaction
-- When starting automatic Compaction
-- When doing CompactFiles()
Test Plan: unit test
Reviewers: lightmark, yiwu, andrewkr, sdong
Reviewed By: sdong
Subscribers: andrewkr, yoshinorim, jkedgar, dhruba
Differential Revision: https://reviews.facebook.net/D64383
Summary: Use ImmutableDBOptions/MutableDBOptions internally and DBOptions only for user-facing APIs. MutableDBOptions is barely a placeholder for now. I'll start to move options to MutableDBOptions in following diffs.
Test Plan:
make all check
Reviewers: yhchiang, IslamAbdelRahman, sdong
Reviewed By: sdong
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D64065
Summary:
Revert the behavior where we don't read sequence id from WAL, but increase it as we replay the log. We still keep the behave for 2PC for now but will fix later.
This change fixes github issue 1339, where some writes come with WAL disabled and we may recover records with wrong sequence id.
Test Plan: Added unit test.
Subscribers: andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D64275
Summary: WritableFile::SetPreallocationBlockSize() requires parameter as size_t, and options used in DBImpl::GetWalPreallocateBlockSize() are all size_t. WritableFile::SetPreallocationBlockSize() should return size_t to avoid build break if size_t is not uint64_t.
Test Plan: Run existing tests.
Reviewers: andrewkr, IslamAbdelRahman, yiwu
Reviewed By: yiwu
Subscribers: leveldb, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D64137
Summary: Currently the WAL file preallocation size is 1.1 * write_buffer_size. This, however, will be over-estimated if options.db_write_buffer_size or options.max_total_wal_size is set and is much smaller.
Test Plan: Add a unit test.
Reviewers: andrewkr, yiwu
Reviewed By: yiwu
Subscribers: leveldb, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D63957
Summary: As title, make sure Prev() works as expected with Next() when the current iter->key() in the range of the same prefix in prefix seek mode
Test Plan: make all check -j64 (add prefix_test with PrefixSeekModePrev test case)
Reviewers: andrewkr, sdong, IslamAbdelRahman
Reviewed By: IslamAbdelRahman
Subscribers: yoshinorim, andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D61419
Summary: 1. Range Deletion Tombstone structure 2. Modify Add() in table_builder to make it usable for adding range del tombstones 3. Expose NewTombstoneIterator() API in table_reader
Test Plan: table_test.cc (now BlockBasedTableBuilder::Add() only accepts InternalKey. I make table_test only pass InternalKey to BlockBasedTableBuidler. Also test writing/reading range deletion tombstones in table_test )
Reviewers: sdong, IslamAbdelRahman, lightmark, andrewkr
Reviewed By: andrewkr
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D61473
Summary: With read_options.background_purge_on_iterator_cleanup=true, File deletion and closing can still happen in forward iterator, or WAL file closing. Cover those cases too.
Test Plan: I am adding unit tests.
Reviewers: andrewkr, IslamAbdelRahman, yiwu
Reviewed By: yiwu
Subscribers: leveldb, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D61503
Maysam Yabandeh