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:
Fixing two types of clang-analyzer false positives:
* db is deleted and then reopen, and clang-analyzer thinks we are reusing the pointer after it has been deleted. Adding asserts to hint clang-analyzer the pointer is recreated.
* ParsedInternalKey is (intentionally) uninitialized. Initialize the struct only when clang-analyzer is running.
Closes https://github.com/facebook/rocksdb/pull/2334
Differential Revision: D5093801
Pulled By: yiwu-arbug
fbshipit-source-id: f51355382098eb3da5ab9f64e094c6d03e6bdf7d
Summary:
Upgrading a shared lock was silently succeeding because the actual locking code was skipped. This is because if the keys are tracked, it is assumed that they are already locked and do not require locking. Fix this by recording in tracked keys whether the key was locked exclusively or not.
Note that lock downgrades are impossible, which is the behaviour we expect.
This fixesfacebook/mysql-5.6#587.
Closes https://github.com/facebook/rocksdb/pull/2122
Differential Revision: D4861489
Pulled By: IslamAbdelRahman
fbshipit-source-id: 58c7ebe7af098bf01b9774b666d3e9867747d8fd
Summary:
Extend TransactionOptions to include max_write_batch_size which determines the maximum size of the writebatch representation. If memory limit is exceeded, the operation will abort with subcode kMemoryLimit.
Closes https://github.com/facebook/rocksdb/pull/2124
Differential Revision: D4861842
Pulled By: lth
fbshipit-source-id: 46fd172ea67cc90bbba829bf0d70cfab2261c161
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:
introduce new methods into a public threadpool interface,
- allow submission of std::functions as they allow greater flexibility.
- add Joining methods to the implementation to join scheduled and submitted jobs with
an option to cancel jobs that did not start executing.
- Remove ugly `#ifdefs` between pthread and std implementation, make it uniform.
- introduce pimpl for a drop in replacement of the implementation
- Introduce rocksdb::port::Thread typedef which is a replacement for std::thread. On Posix Thread defaults as before std::thread.
- Implement WindowsThread that allocates memory in a more controllable manner than windows std::thread with a replaceable implementation.
- should be no functionality changes.
Closes https://github.com/facebook/rocksdb/pull/1823
Differential Revision: D4492902
Pulled By: siying
fbshipit-source-id: c74cb11
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:
Enable directIO on WritableFileImpl::Append
with offset being current length of the file.
Enable UniqueID tests on Windows, disable others but
leeting them to compile. Unique tests are valuable to
detect failures on different filesystems and upcoming
ReFS.
Clear output in WinEnv Getchildren.This is different from
previous strategy, do not touch output on failure.
Make sure DBTest.OpenWhenOpen works with windows error message
Closes https://github.com/facebook/rocksdb/pull/1746
Differential Revision: D4385681
Pulled By: IslamAbdelRahman
fbshipit-source-id: c07b702
Summary:
In the test the last change to AAAZZZ in handles[1] is deleting it. The
result of the get must be NotFound then. Previosuly the test did not
check for the return value of Get and assumed that the status is ok. It
then move ahead asserting the returned value. The passed-by-reference
string value however was not changed (since the key was not found) and
the asserted value is what it contained before doing the Get.
Closes https://github.com/facebook/rocksdb/pull/1753
Differential Revision: D4390982
Pulled By: maysamyabandeh
fbshipit-source-id: dd55a34
Summary:
Fixes compile error:
In file included from ./util/statistics.h:17:0,
from ./util/stop_watch.h:8,
from ./util/perf_step_timer.h:9,
from ./util/iostats_context_imp.h:8,
from ./util/posix_logger.h:27,
from ./port/util_logger.h:18,
from ./db/auto_roll_logger.h:15,
from db/auto_roll_logger.cc:6:
./util/thread_local.h:65:16: error: 'function' in namespace 'std' does not name a template type
typedef std::function<void(void*, void*)> FoldFunc;
Closes https://github.com/facebook/rocksdb/pull/1656
Differential Revision: D4318702
Pulled By: yiwu-arbug
fbshipit-source-id: 8c5d17a
Summary:
This is an implementation of non-exclusive locks for pessimistic transactions. It is relatively simple and does not prevent starvation (ie. it's possible that request for exclusive access will never be granted if there are always threads holding shared access). It is done by changing `KeyLockInfo` to hold an set a transaction ids, instead of just one, and adding a flag specifying whether this lock is currently held with exclusive access or not.
Some implementation notes:
- Some lock diagnostic functions had to be updated to return a set of transaction ids for a given lock, eg. `GetWaitingTxn` and `GetLockStatusData`.
- Deadlock detection is a bit more complicated since a transaction can now wait on multiple other transactions. A BFS is done in this case, and deadlock detection depth is now just a limit on the number of transactions we visit.
- Expirable transactions do not work efficiently with shared locks at the moment, but that's okay for now.
Closes https://github.com/facebook/rocksdb/pull/1573
Differential Revision: D4239097
Pulled By: lth
fbshipit-source-id: da7c074
Summary:
Originally sequence ids were calculated, in recovery, based off of the first seqid found if the first log recovered. The working seqid was then incremented from that value based on every insertion that took place. This was faulty because of the potential for missing log files or inserts that skipped the WAL. The current recovery scheme grabs sequence from current recovering batch and increments using memtableinserter to track how many actual inserts take place. This works for 2PC batches as well scenarios where some logs are missing or inserts that skip the WAL.
Closes https://github.com/facebook/rocksdb/pull/1486
Differential Revision: D4156064
Pulled By: reidHoruff
fbshipit-source-id: a6da8d9
Summary: Implement deadlock detection. This is done by maintaining a TxnID -> TxnID map which represents the edges in the wait for graph (this is named `wait_txn_map_`).
Test Plan: transaction_test
Reviewers: IslamAbdelRahman, sdong
Reviewed By: sdong
Subscribers: andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D64491
Summary: Modifies the lock info export test to test multiple column families after I was experiencing a bug while developing the MyRocks front-end for this.
Test Plan: is test.
Reviewers: mung
Reviewed By: mung
Subscribers: andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D64725
Summary:
This diff does 3 things:
Expose TransactionID so that we can identify transactions when we retrieve locking and lock wait information. This is exposed as `Transaction::GetID`.
Expose lock state information by locking all stripes in all column families and copying their contents to a data structure. This is exposed as `TransactionDB::GetLockStatusData`.
Adds support for tracking the transaction and the key being waited on, and exposes this as `Transaction::GetWaitingTxn`.
Test Plan: unit tests
Reviewers: horuff, sdong
Reviewed By: sdong
Subscribers: vasilep, hermanlee4, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D64413
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: make transactionDB working with StackableDB
Test Plan: make all check -j64
Reviewers: andrewkr, yiwu, sdong
Reviewed By: sdong
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D60705
Summary: This tests that a prepared transaction is not lost after several crashes, restarts, and memtable flushes.
Test Plan: TwoPhaseLongPrepareTest
Reviewers: sdong
Subscribers: hermanlee4, andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D58185
Summary:
TransactionTest.TwoPhaseMultiThreadTest runs forever under TSAN and our CI builds time out
looks like the reason is that some threads keep running and other threads dont get a chance to increment the counter
Test Plan: run the test under TSAN
Reviewers: sdong, horuff
Reviewed By: horuff
Subscribers: andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D58359
Summary:
- Make sure we clean up recovered_transactions_ on DBImpl destructor
- delete leaked txns and env in TransactionTest
Test Plan: Run transaction_test under valgrind
Reviewers: sdong, andrewkr, yhchiang, horuff
Reviewed By: horuff
Subscribers: andrewkr, dhruba
Differential Revision: https://reviews.facebook.net/D58263
Summary:
1. prepare()
2. crash
3. recover
4. commit()
5. crash
6. data is lost
This is due to the transaction data still only residing in the WAL but because the logs were flushed on the first recovery the data is ignored on the second recovery. We must scan all logs found on recovery and only ignore redundant data at the time of replay. It is not possible to know which logs still contain relevant data at time of recovery. We cannot simply ignore a log because all of the non-2pc data it contains has already been written to L0.
The changes made to MemTableInserter are to ensure that prepared sections are still recovered even if all of the non-2pc data in that log has already been flushed to L0.
Test Plan: Provided test.
Reviewers: sdong
Subscribers: andrewkr, hermanlee4, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D57729
Summary:
Consider the following WAL with 4 batch entries prefixed with their sequence at time of memtable insert.
[1: BEGIN_PREPARE, PUT, PUT, PUT, PUT, END_PREPARE(a)]
[1: BEGIN_PREPARE, PUT, PUT, PUT, PUT, END_PREPARE(b)]
[4: COMMIT(a)]
[7: COMMIT(b)]
The first two batches do not consume any sequence numbers so are both prefixed with seq=1.
For 2pc commit, memtable insertion takes place before COMMIT batch is written to WAL.
We can see that sequence number consumption takes place between WAL entries giving us the seemingly sparse sequence prefix for WAL entries.
This is a valid WAL.
Because with 2PC markers one WriteBatch points to another batch containing its inserts a writebatch can consume more or less sequence numbers than the number of sequence consuming entries that it contains.
We can see that, given the entries in the WAL, 6 sequence ids were consumed. Yet on recovery the maximum sequence consumed would be 7 + 3 (the number of sequence numbers consumed by COMMIT(b))
So, now upon recovery we must track the actual consumption of sequence numbers.
In the provided scenario there will be no sequence gaps, but it is possible to produce a sequence gap. This should not be a problem though. correct?
Test Plan: provided test.
Reviewers: sdong
Subscribers: andrewkr, leveldb, dhruba, hermanlee4
Differential Revision: https://reviews.facebook.net/D57645
Summary: Refactored db_bench transaction stress tests so that they can be called from unit tests as well.
Test Plan: run new unit test as well as db_bench
Reviewers: yhchiang, IslamAbdelRahman, sdong
Reviewed By: IslamAbdelRahman
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D55203
Summary: Previously, reusing a transaction (by passing it as an argument to BeginTransaction) would not clear the transaction's snapshot. This is not a clear, well-definited behavior.
Test Plan: improved test
Reviewers: sdong, IslamAbdelRahman, horuff, jkedgar
Reviewed By: jkedgar
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D55053
Summary: Add function to reinitialize a transaction object so that it can be reused. This is an optimization so users can potentially avoid reallocating transaction objects.
Test Plan: added tests
Reviewers: yhchiang, kradhakrishnan, IslamAbdelRahman, sdong
Reviewed By: sdong
Subscribers: jkedgar, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D53835
Summary: There is an issue in DBImpl::WriteImpl where if an empty writebatch comes in and sync=true then the logs will be marked as being synced yet the sync never actually happens because there is no data in the writebatch. This causes the next incoming batch to hang while waiting for the logs to complete syncing. This fix syncs logs even if the writebatch is empty.
Test Plan: DoubleEmptyBatch unit test in transaction_test.
Reviewers: yoshinorim, hermanlee4, sdong, ngbronson, anthony
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D54057
Summary:
One test in transaction_test.cc forgets to call SyncPoint::DisableProcessing().
As a result, a program might to access the SyncPoint singleton after it
already goes out of scope.
This patch fix this error by calling SyncPoint::DisableProcessing().
Test Plan: transaction_test
Reviewers: sdong, IslamAbdelRahman, kradhakrishnan, anthony
Reviewed By: anthony
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D54033
Summary: MyRocks wants to be able to un-lock a key that was just locked by GetForUpdate(). To do this safely, I am now keeping track of the number of reads(for update) and writes for each key in a transaction. UndoGetForUpdate() will only unlock a key if it hasn't been written and the read count reaches 0.
Test Plan: more unit tests
Reviewers: igor, rven, yhchiang, spetrunia, sdong
Reviewed By: spetrunia, sdong
Subscribers: spetrunia, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47043
Summary:
Doing inline checking of transaction expiration instead of
using a callback.
Test Plan: To be added
Reviewers: anthony
Reviewed By: anthony
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D53673
Summary: Add support to change write options after creating a transaction. This is needed for MongoRocks.
Test Plan: added test
Reviewers: sdong, rven, kradhakrishnan, IslamAbdelRahman, yhchiang
Reviewed By: yhchiang
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D51867
Summary:
Currently, transactions can fail even if there is no actual write conflict. This is due to relying on only the memtables to check for write-conflicts. Users have to tune memtable settings to try to avoid this, but it's hard to figure out exactly how to tune these settings.
With this diff, TransactionDB will use both memtables and SST files to determine if there are any write conflicts. This relies on the fact that BlockBasedTable stores sequence numbers for all writes that happen after any open snapshot. Also, D50295 is needed to prevent SingleDelete from disappearing writes (the TODOs in this test code will be fixed once the other diff is approved and merged).
Note that Optimistic transactions will still rely on tuning memtable settings as we do not want to read from SST while on the write thread. Also, memtable settings can still be used to reduce how often TransactionDB needs to read SST files.
Test Plan: unit tests, db bench
Reviewers: rven, yhchiang, kradhakrishnan, IslamAbdelRahman, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb, yoshinorim
Differential Revision: https://reviews.facebook.net/D50475
Summary: When SetSnapshot() is used the caller immediately knows a snapshot has been created, but when SetSnapshotOnNextOperation() is used the caller needs a way to get notified when that snapshot has been generated. This creates an interface that the client can implement that will be called at the time the snapshot is created.
Test Plan: Added a new SetSnapshotOnNextOperationWithNotification test into the transaction_test.
Reviewers: sdong, anthony
Reviewed By: anthony
Subscribers: yoshinorim, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D51177
Summary:
Fixes T8781168.
Added a new function EnableAutoCompactions in db.h to be publicly
avialable. This allows compaction to be re-enabled after disabling it via
SetOptions
Refactored code to set the dbptr earlier on in TransactionDB::Open and DB::Open
Temporarily disable auto_compaction in TransactionDB::Open until dbptr is set to
prevent race condition.
Test Plan:
Ran make all check
verified fix on myrocks side:
was able to reproduce the seg fault with
../tools/mysqltest.sh --mem --force rocksdb.drop_table
method was to manually sleep the thread after DB::Open but before TransactionDB ptr was
assigned in transaction_db_impl.cc:
DB::Open(db_options, dbname, column_families_copy, handles, &db);
clock_t goal = (60000 * 10) + clock();
while (goal > clock());
...dbptr(aka rdb) gets assigned below
verified my changes fixed the issue.
Also added unit test 'ToggleAutoCompaction' in transaction_test.cc
Reviewers: hermanlee4, anthony
Reviewed By: anthony
Subscribers: alex, dhruba
Differential Revision: https://reviews.facebook.net/D51147
Summary:
MyRocks needs the ability to clear a snapshot for Read Committed support
Test Plan: transaction_test
Reviewers: anthony
Reviewed By: anthony
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D48861
Summary: Support for Transaction::CreateSnapshotOnNextOperation(). This is to fix a write-conflict race-condition that Yoshinori was running into when testing MyRocks with LinkBench.
Test Plan: New tests
Reviewers: yhchiang, spetrunia, rven, igor, yoshinorim, sdong
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D48099
Summary:
MyRocks reported some perfomance issues when inserting many keys into a transaction due to the cost of inserting new keys into WriteBatchWithIndex. Frequently, they don't even need the keys to be indexed as they don't need to read them back. DisableIndexing() can be used to avoid the cost of indexing.
I also plan on eventually investigating if we can improve WriteBatchWithIndex performance. But even if we improved the perf here, it is still beneficial to be able to disable the indexing all together for large transactions.
Test Plan: unit test
Reviewers: igor, rven, yoshinorim, spetrunia, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D48471
Summary:
WriteBatchWithIndex::GetFromBatchAndDB only works correctly for overwrite_key=false. Transactions use overwrite_key=true (since WriteBatchWithIndex::GetIteratorWithBase only works when overwrite_key=true). So currently, Transactions could return incorrectly merged results when calling Get/GetForUpdate().
Until a permanent fix can be put in place, Transaction::Get[ForUpdate] and WriteBatchWithIndex::GetFromBatch[AndDB] will now return MergeInProgress if the most recent write to a key in the batch is a Merge.
Test Plan: more tests
Reviewers: sdong, yhchiang, rven, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47817
Summary: Transactional SingleDelete is needed for MyRocks. Note: This diff requires D47529.
Test Plan: Added some new tests in this diff as well as more tests added in D47529
Reviewers: rven, sdong, igor, yhchiang
Reviewed By: yhchiang
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47535
Summary: Transaction::RollbackToSavePoint() will now release any locks that were taken since the previous SavePoint. To do this cleanly, I moved tracked_keys_ management into TransactionBase.
Test Plan: New Transaction test.
Reviewers: igor, rven, sdong
Reviewed By: sdong
Subscribers: dhruba, spetrunia, leveldb
Differential Revision: https://reviews.facebook.net/D46761
Yedidya Feldblum