rocksdb/db/db_impl_debug.cc
Nathan Bronson b7198c3afe reduce db mutex contention for write batch groups
Summary:
This diff allows a Writer to join the next write batch group
without acquiring any locks. Waiting is performed via a per-Writer mutex,
so all of the non-leader writers never need to acquire the db mutex.
It is now possible to join a write batch group after the leader has been
chosen but before the batch has been constructed. This diff doesn't
increase parallelism, but reduces synchronization overheads.

For some CPU-bound workloads (no WAL, RAM-sized working set) this can
substantially reduce contention on the db mutex in a multi-threaded
environment.  With T=8 N=500000 in a CPU-bound scenario (see the test
plan) this is good for a 33% perf win.  Not all scenarios see such a
win, but none show a loss.  This code is slightly faster even for the
single-threaded case (about 2% for the CPU-bound scenario below).

Test Plan:
1. unit tests
2. COMPILE_WITH_TSAN=1 make check
3. stress high-contention scenarios with db_bench -benchmarks=fillrandom -threads=$T -batch_size=1 -memtablerep=skip_list -value_size=0 --num=$N -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

Reviewers: sdong, igor, rven, ljin, yhchiang

Subscribers: dhruba

Differential Revision: https://reviews.facebook.net/D43887
2015-08-14 10:55:43 -07:00

157 lines
4.6 KiB
C++

// Copyright (c) 2013, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#ifndef ROCKSDB_LITE
#include "db/db_impl.h"
#include "util/thread_status_updater.h"
namespace rocksdb {
uint64_t DBImpl::TEST_GetLevel0TotalSize() {
InstrumentedMutexLock l(&mutex_);
return default_cf_handle_->cfd()->current()->storage_info()->NumLevelBytes(0);
}
Iterator* DBImpl::TEST_NewInternalIterator(Arena* arena,
ColumnFamilyHandle* column_family) {
ColumnFamilyData* cfd;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
cfd = cfh->cfd();
}
mutex_.Lock();
SuperVersion* super_version = cfd->GetSuperVersion()->Ref();
mutex_.Unlock();
ReadOptions roptions;
return NewInternalIterator(roptions, cfd, super_version, arena);
}
int64_t DBImpl::TEST_MaxNextLevelOverlappingBytes(
ColumnFamilyHandle* column_family) {
ColumnFamilyData* cfd;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
cfd = cfh->cfd();
}
InstrumentedMutexLock l(&mutex_);
return cfd->current()->storage_info()->MaxNextLevelOverlappingBytes();
}
void DBImpl::TEST_GetFilesMetaData(
ColumnFamilyHandle* column_family,
std::vector<std::vector<FileMetaData>>* metadata) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
auto cfd = cfh->cfd();
InstrumentedMutexLock l(&mutex_);
metadata->resize(NumberLevels());
for (int level = 0; level < NumberLevels(); level++) {
const std::vector<FileMetaData*>& files =
cfd->current()->storage_info()->LevelFiles(level);
(*metadata)[level].clear();
for (const auto& f : files) {
(*metadata)[level].push_back(*f);
}
}
}
uint64_t DBImpl::TEST_Current_Manifest_FileNo() {
return versions_->manifest_file_number();
}
Status DBImpl::TEST_CompactRange(int level, const Slice* begin,
const Slice* end,
ColumnFamilyHandle* column_family,
bool disallow_trivial_move) {
ColumnFamilyData* cfd;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
cfd = cfh->cfd();
}
int output_level =
(cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO)
? level
: level + 1;
return RunManualCompaction(cfd, level, output_level, 0, begin, end,
disallow_trivial_move);
}
Status DBImpl::TEST_FlushMemTable(bool wait) {
FlushOptions fo;
fo.wait = wait;
return FlushMemTable(default_cf_handle_->cfd(), fo);
}
Status DBImpl::TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family) {
ColumnFamilyData* cfd;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
cfd = cfh->cfd();
}
return WaitForFlushMemTable(cfd);
}
Status DBImpl::TEST_WaitForCompact() {
// Wait until the compaction completes
// TODO: a bug here. This function actually does not necessarily
// wait for compact. It actually waits for scheduled compaction
// OR flush to finish.
InstrumentedMutexLock l(&mutex_);
while ((bg_compaction_scheduled_ || bg_flush_scheduled_) && bg_error_.ok()) {
bg_cv_.Wait();
}
return bg_error_;
}
void DBImpl::TEST_LockMutex() {
mutex_.Lock();
}
void DBImpl::TEST_UnlockMutex() {
mutex_.Unlock();
}
void* DBImpl::TEST_BeginWrite() {
auto w = new WriteThread::Writer();
write_thread_.EnterUnbatched(w, &mutex_);
return reinterpret_cast<void*>(w);
}
void DBImpl::TEST_EndWrite(void* w) {
auto writer = reinterpret_cast<WriteThread::Writer*>(w);
write_thread_.ExitUnbatched(writer);
delete writer;
}
size_t DBImpl::TEST_LogsToFreeSize() {
InstrumentedMutexLock l(&mutex_);
return logs_to_free_.size();
}
uint64_t DBImpl::TEST_LogfileNumber() {
InstrumentedMutexLock l(&mutex_);
return logfile_number_;
}
} // namespace rocksdb
#endif // ROCKSDB_LITE