Summary:
When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue.
The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction.
Here are the performance results:
Command:
./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333
Before the patch:
fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s
After the patch:
fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s
Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got:
fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s
Test Plan:
make check
two stress tests:
Big number of compactions and flushes:
./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000
max_background_flushes=0, to verify that this case also works correctly
./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000
Reviewers: ljin, rven, yhchiang, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D30123
Summary:
- AssertionError when initialized with Non-Direct Buffer
- Tests + coverage for DirectSlice
- Slice sigsegv fixes when initializing from String and byte arrays
- Slice Tests
Test Plan: Run tests without source modifications.
Reviewers: yhchiang, adamretter, ankgup87
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D30081
Summary: Avoid unnecessary unlock and lock mutex when notifying events.
Test Plan: ./listener_test
Reviewers: igor
Reviewed By: igor
Subscribers: sdong, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D30267
Summary:
We now release the mutex before copying the files in the case
of the trivial move. This path does not use the compaction job.
Test Plan: DBTest.LevelCompactionThirdPath
Reviewers: yhchiang, igor, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D30381
Summary: Manual range compaction support in RocksJava.
Test Plan:
make rocksdbjava
make jtest
mvn -f rocksjni.pom package
Reviewers: adamretter, yhchiang, ankgup87
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D29283
Summary:
Allow Level-style compaction to place files in different paths
This diff provides the code for task 4854591. We now support level-compaction
to place files in different paths by specifying them in db_paths along with
the minimum level for files to store in that path.
Test Plan: ManualLevelCompactionOutputPathId in db_test.cc
Reviewers: yhchiang, MarkCallaghan, dhruba, yoshinorim, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29799
Summary:
This diff changes compile to optimize for native platform by default. This will automatically turn on crc32 optimizations for modern processors, which greatly improves rocksdb's performance.
I also did some more changes to compilation documentation.
Test Plan:
compile with `make`, observe -march=native
compile with `PORTABLE=1 make`, observe no -march=native
Reviewers: sdong, rven, yhchiang, MarkCallaghan
Reviewed By: MarkCallaghan
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D30225
Summary:
Priliminary diff to solicit comments.
Given DB path, dump all SST files (key/value and properties), WAL file and manifest
files. What command options do we need to support for this command? Maybe
output_hex for keys?
Test Plan: Create additional ldb unit tests.
Reviewers: sdong, rven
Reviewed By: rven
Subscribers: dhruba
Differential Revision: https://reviews.facebook.net/D29547
Summary:
ASAN build fails once for this error:
14:04:52 ==== Test DBTest.CompactFilesOnLevelCompaction
14:04:52 db_test: db/version_set.cc:1062: void rocksdb::VersionStorageInfo::AddFile(int, rocksdb::FileMetaData*): Assertion `level <= 0 || level_files->empty() || internal_comparator_->Compare( (*level_files)[level_files->size() - 1]->largest, f->smallest) < 0' failed.
Not abling figure out reason. We use std:vector for sorting for save and add one more assert to help figure out whether it is the sorting's problem.
Test Plan: make all check
Reviewers: yhchiang, rven, igor
Reviewed By: igor
Subscribers: leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D30117
Summary: Now DB::GetSnapshot() doesn't scale to more column families, as it needs to go through all the column families to find whether snapshot is supported. This patch optimizes it.
Test Plan:
Add unit tests to cover negative cases.
make all check
Reviewers: yhchiang, rven, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D30093
Summary:
Currently, blocks which have more than one reference (ie referenced by something other than cache itself) are evicted from cache. This doesn't make much sense:
- blocks are still in RAM, so the RAM usage reported by the cache is incorrect
- if the same block is needed by another iterator, it will be loaded and decompressed again
This diff changes the reference counting scheme a bit. Previously, if the cache contained the block, this was accounted for in its refcount. After this change, the refcount is only used to track external references. There is a boolean flag which indicates whether or not the block is contained in the cache.
This diff also changes how LRU list is used. Previously, both hashtable and the LRU list contained all blocks. After this change, the LRU list contains blocks with the refcount==0, ie those which can be evicted from the cache.
Note that this change still allows for cache to grow beyond its capacity. This happens when all blocks are pinned (ie refcount>0). This is consistent with the current behavior. The cache's insert function never fails. I spent lots of time trying to make table_reader and other places work with the insert which might failed. It turned out to be pretty hard. It might really destabilize some customers, so finally, I decided against doing this.
table_cache_remove_scan_count_limit option will be unneeded after this change, but I will remove it in the following diff, if this one gets approved
Test Plan: Ran tests, made sure they pass
Reviewers: sdong, ljin
Differential Revision: https://reviews.facebook.net/D25503
Summary: Set operations in VerisonBuilder is shown as a performance bottleneck of restarting DB when there are lots of files. Make both of added_files and deleted_files use unordered set or map. Only when adding the files, sort the added files.
Test Plan: make all check
Reviewers: yhchiang, rven, igor
Reviewed By: igor
Subscribers: hermanlee4, leveldb, dhruba, ljin
Differential Revision: https://reviews.facebook.net/D30051
Summary: as title
Test Plan: ran it
Reviewers: yhchiang, igor, sdong, MarkCallaghan
Reviewed By: MarkCallaghan
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D25563
Summary: Why do we assert here? This doesn't seem like user friendly thing to do :)
Test Plan: none
Reviewers: sdong, yhchiang, rven
Reviewed By: rven
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D30027
Summary: This is a regression bug introduced by https://reviews.facebook.net/D24729 . max_total_wal_size would be off the target it should be more and more in the case that the a user holds the current super version after flush or compaction. This patch fixes it
Test Plan: make all check
Reviewers: yhchiang, rven, igor
Reviewed By: igor
Subscribers: ljin, yoshinorim, MarkCallaghan, hermanlee4, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29961
Summary: When wal_dir is used, DestroyDB is not passed the wal_dir option and so we get a Corruption exception.
Test Plan:
Verified manually that the following command line works now:
./db_bench --db=/mnt/db/rocksdb ... --disable_wal=0 --wal_dir=/data/users/rocksdb/WAL... --benchmarks=filluniquerandom --use_existing_db=0...
Reviewers: sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29859
Summary: Add -fno-exceptions flag to ROCKSDB_LITE.
Test Plan: make OPT=-DROCKSDB_LITE shared_lib -j32
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29901
Summary:
Add a counter in SnapshotList to show number of snapshots. Also a unix timestamp in every snapshot.
Add two DB Properties to return number of snapshots and timestamp of the oldest one.
Test Plan: Add unit test checking
Reviewers: yhchiang, rven, igor
Reviewed By: igor
Subscribers: leveldb, dhruba, MarkCallaghan
Differential Revision: https://reviews.facebook.net/D29919
Summary:
While running rocksdb tests, we sometimes encounter errors and
the test run stops. We now provide a new make target call check_some
which restarts the test run from a specific test and continues from
there depending on the value of the environment variable ROCKSDBTESTS_START
Test Plan:
Run make check_some with different values of
ROCKSDBTESTS_START.
Reviewers: igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29913
Summary: Replace exception by abort() in dummy HdfsEnv implementation.
Test Plan: make dbg -j32
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29895
Summary:
Remove the compability check on log2 OS_ANDROID as it's already blocked by ROCKSDB_LITE
Test Plan:
make OPT="-DROCKSDB_LITE -DOS_ANDROID" shared_lib -j32
make shared_lib -j32
Summary: Replace runtime_error exception by abort() in thread_local
Test Plan: make dbg -j32
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29853
Summary:
Remove the use of exception in WriteBatch::Handler. Now the default
implementations of Put, Merge, and Delete in WriteBatch::Handler are no-op.
Test Plan:
Add three test cases in write_batch_test
./write_batch_test
Reviewers: sdong, igor
Reviewed By: sdong, igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29835
Summary: Replace exception by assertion in autovector
Test Plan: autovector_test
Reviewers: sdong, igor
Reviewed By: igor
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29847
Summary:
Fix the following compile warning in db_stress.cc on Mac
tools/db_stress.cc:1688:52: error: format specifies type 'unsigned long' but the argument has type '::google::uint64' (aka 'unsigned long long') [-Werror,-Wformat]
fprintf(stdout, "DB-write-buffer-size: %lu\n", FLAGS_db_write_buffer_size);
~~~ ^~~~~~~~~~~~~~~~~~~~~~~~~~
%llu
Test Plan:
make
Summary: Replace exception by setting valid_ = false in DBIter::MergeValuesNewToOld().
Test Plan:
Not sure if I am right at this, but it seems we currently don't have a good
way to test that code path as it requires dynamically set merge_operator = nullptr
at the time while Merge() is calling.
Reviewers: igor, sdong
Reviewed By: sdong
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D29811
