You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
rocksdb/db/column_family.cc

1018 lines
38 KiB

// Copyright (c) 2011-present, 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.
#include "db/column_family.h"
#ifndef __STDC_FORMAT_MACROS
#define __STDC_FORMAT_MACROS
#endif
#include <inttypes.h>
#include <vector>
#include <string>
#include <algorithm>
#include <limits>
#include "db/compaction_picker.h"
#include "db/db_impl.h"
#include "db/internal_stats.h"
#include "db/job_context.h"
#include "db/table_properties_collector.h"
#include "db/version_set.h"
#include "db/write_controller.h"
#include "db/writebuffer.h"
#include "memtable/hash_skiplist_rep.h"
#include "util/autovector.h"
#include "util/compression.h"
#include "util/options_helper.h"
#include "util/thread_status_util.h"
Support saving history in memtable_list 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
10 years ago
#include "util/xfunc.h"
namespace rocksdb {
ColumnFamilyHandleImpl::ColumnFamilyHandleImpl(
ColumnFamilyData* column_family_data, DBImpl* db, InstrumentedMutex* mutex)
: cfd_(column_family_data), db_(db), mutex_(mutex) {
if (cfd_ != nullptr) {
cfd_->Ref();
}
}
ColumnFamilyHandleImpl::~ColumnFamilyHandleImpl() {
if (cfd_ != nullptr) {
// Job id == 0 means that this is not our background process, but rather
// user thread
JobContext job_context(0);
mutex_->Lock();
if (cfd_->Unref()) {
delete cfd_;
}
db_->FindObsoleteFiles(&job_context, false, true);
mutex_->Unlock();
if (job_context.HaveSomethingToDelete()) {
db_->PurgeObsoleteFiles(job_context);
}
job_context.Clean();
}
}
uint32_t ColumnFamilyHandleImpl::GetID() const { return cfd()->GetID(); }
const std::string& ColumnFamilyHandleImpl::GetName() const {
return cfd()->GetName();
}
Status ColumnFamilyHandleImpl::GetDescriptor(ColumnFamilyDescriptor* desc) {
#ifndef ROCKSDB_LITE
// accessing mutable cf-options requires db mutex.
InstrumentedMutexLock l(mutex_);
*desc = ColumnFamilyDescriptor(
cfd()->GetName(),
BuildColumnFamilyOptions(*cfd()->options(),
*cfd()->GetLatestMutableCFOptions()));
return Status::OK();
#else
return Status::NotSupported();
#endif // !ROCKSDB_LITE
}
const Comparator* ColumnFamilyHandleImpl::user_comparator() const {
return cfd()->user_comparator();
}
void GetIntTblPropCollectorFactory(
const ColumnFamilyOptions& cf_options,
std::vector<std::unique_ptr<IntTblPropCollectorFactory>>*
int_tbl_prop_collector_factories) {
auto& collector_factories = cf_options.table_properties_collector_factories;
for (size_t i = 0; i < cf_options.table_properties_collector_factories.size();
++i) {
assert(collector_factories[i]);
int_tbl_prop_collector_factories->emplace_back(
new UserKeyTablePropertiesCollectorFactory(collector_factories[i]));
}
// Add collector to collect internal key statistics
int_tbl_prop_collector_factories->emplace_back(
new InternalKeyPropertiesCollectorFactory);
}
Status CheckCompressionSupported(const ColumnFamilyOptions& cf_options) {
if (!cf_options.compression_per_level.empty()) {
for (size_t level = 0; level < cf_options.compression_per_level.size();
++level) {
if (!CompressionTypeSupported(cf_options.compression_per_level[level])) {
return Status::InvalidArgument(
"Compression type " +
CompressionTypeToString(cf_options.compression_per_level[level]) +
" is not linked with the binary.");
}
}
} else {
if (!CompressionTypeSupported(cf_options.compression)) {
return Status::InvalidArgument(
"Compression type " +
CompressionTypeToString(cf_options.compression) +
" is not linked with the binary.");
}
}
return Status::OK();
}
support for concurrent adds to memtable Summary: This diff adds support for concurrent adds to the skiplist memtable implementations. Memory allocation is made thread-safe by the addition of a spinlock, with small per-core buffers to avoid contention. Concurrent memtable writes are made via an additional method and don't impose a performance overhead on the non-concurrent case, so parallelism can be selected on a per-batch basis. Write thread synchronization is an increasing bottleneck for higher levels of concurrency, so this diff adds --enable_write_thread_adaptive_yield (default off). This feature causes threads joining a write batch group to spin for a short time (default 100 usec) using sched_yield, rather than going to sleep on a mutex. If the timing of the yield calls indicates that another thread has actually run during the yield then spinning is avoided. This option improves performance for concurrent situations even without parallel adds, although it has the potential to increase CPU usage (and the heuristic adaptation is not yet mature). Parallel writes are not currently compatible with inplace updates, update callbacks, or delete filtering. Enable it with --allow_concurrent_memtable_write (and --enable_write_thread_adaptive_yield). Parallel memtable writes are performance neutral when there is no actual parallelism, and in my experiments (SSD server-class Linux and varying contention and key sizes for fillrandom) they are always a performance win when there is more than one thread. Statistics are updated earlier in the write path, dropping the number of DB mutex acquisitions from 2 to 1 for almost all cases. This diff was motivated and inspired by Yahoo's cLSM work. It is more conservative than cLSM: RocksDB's write batch group leader role is preserved (along with all of the existing flush and write throttling logic) and concurrent writers are blocked until all memtable insertions have completed and the sequence number has been advanced, to preserve linearizability. My test config is "db_bench -benchmarks=fillrandom -threads=$T -batch_size=1 -memtablerep=skip_list -value_size=100 --num=1000000/$T -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 --block_size=16384 --allow_concurrent_memtable_write" on a two-socket Xeon E5-2660 @ 2.2Ghz with lots of memory and an SSD hard drive. With 1 thread I get ~440Kops/sec. Peak performance for 1 socket (numactl -N1) is slightly more than 1Mops/sec, at 16 threads. Peak performance across both sockets happens at 30 threads, and is ~900Kops/sec, although with fewer threads there is less performance loss when the system has background work. Test Plan: 1. concurrent stress tests for InlineSkipList and DynamicBloom 2. make clean; make check 3. make clean; DISABLE_JEMALLOC=1 make valgrind_check; valgrind db_bench 4. make clean; COMPILE_WITH_TSAN=1 make all check; db_bench 5. make clean; COMPILE_WITH_ASAN=1 make all check; db_bench 6. make clean; OPT=-DROCKSDB_LITE make check 7. verify no perf regressions when disabled Reviewers: igor, sdong Reviewed By: sdong Subscribers: MarkCallaghan, IslamAbdelRahman, anthony, yhchiang, rven, sdong, guyg8, kradhakrishnan, dhruba Differential Revision: https://reviews.facebook.net/D50589
10 years ago
Status CheckConcurrentWritesSupported(const ColumnFamilyOptions& cf_options) {
if (cf_options.inplace_update_support) {
return Status::InvalidArgument(
"In-place memtable updates (inplace_update_support) is not compatible "
"with concurrent writes (allow_concurrent_memtable_write)");
}
if (cf_options.filter_deletes) {
return Status::InvalidArgument(
"Delete filtering (filter_deletes) is not compatible with concurrent "
"memtable writes (allow_concurrent_memtable_writes)");
}
if (!cf_options.memtable_factory->IsInsertConcurrentlySupported()) {
return Status::InvalidArgument(
"Memtable doesn't concurrent writes (allow_concurrent_memtable_write)");
}
support for concurrent adds to memtable Summary: This diff adds support for concurrent adds to the skiplist memtable implementations. Memory allocation is made thread-safe by the addition of a spinlock, with small per-core buffers to avoid contention. Concurrent memtable writes are made via an additional method and don't impose a performance overhead on the non-concurrent case, so parallelism can be selected on a per-batch basis. Write thread synchronization is an increasing bottleneck for higher levels of concurrency, so this diff adds --enable_write_thread_adaptive_yield (default off). This feature causes threads joining a write batch group to spin for a short time (default 100 usec) using sched_yield, rather than going to sleep on a mutex. If the timing of the yield calls indicates that another thread has actually run during the yield then spinning is avoided. This option improves performance for concurrent situations even without parallel adds, although it has the potential to increase CPU usage (and the heuristic adaptation is not yet mature). Parallel writes are not currently compatible with inplace updates, update callbacks, or delete filtering. Enable it with --allow_concurrent_memtable_write (and --enable_write_thread_adaptive_yield). Parallel memtable writes are performance neutral when there is no actual parallelism, and in my experiments (SSD server-class Linux and varying contention and key sizes for fillrandom) they are always a performance win when there is more than one thread. Statistics are updated earlier in the write path, dropping the number of DB mutex acquisitions from 2 to 1 for almost all cases. This diff was motivated and inspired by Yahoo's cLSM work. It is more conservative than cLSM: RocksDB's write batch group leader role is preserved (along with all of the existing flush and write throttling logic) and concurrent writers are blocked until all memtable insertions have completed and the sequence number has been advanced, to preserve linearizability. My test config is "db_bench -benchmarks=fillrandom -threads=$T -batch_size=1 -memtablerep=skip_list -value_size=100 --num=1000000/$T -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 --block_size=16384 --allow_concurrent_memtable_write" on a two-socket Xeon E5-2660 @ 2.2Ghz with lots of memory and an SSD hard drive. With 1 thread I get ~440Kops/sec. Peak performance for 1 socket (numactl -N1) is slightly more than 1Mops/sec, at 16 threads. Peak performance across both sockets happens at 30 threads, and is ~900Kops/sec, although with fewer threads there is less performance loss when the system has background work. Test Plan: 1. concurrent stress tests for InlineSkipList and DynamicBloom 2. make clean; make check 3. make clean; DISABLE_JEMALLOC=1 make valgrind_check; valgrind db_bench 4. make clean; COMPILE_WITH_TSAN=1 make all check; db_bench 5. make clean; COMPILE_WITH_ASAN=1 make all check; db_bench 6. make clean; OPT=-DROCKSDB_LITE make check 7. verify no perf regressions when disabled Reviewers: igor, sdong Reviewed By: sdong Subscribers: MarkCallaghan, IslamAbdelRahman, anthony, yhchiang, rven, sdong, guyg8, kradhakrishnan, dhruba Differential Revision: https://reviews.facebook.net/D50589
10 years ago
return Status::OK();
}
ColumnFamilyOptions SanitizeOptions(const DBOptions& db_options,
const InternalKeyComparator* icmp,
const ColumnFamilyOptions& src) {
ColumnFamilyOptions result = src;
result.comparator = icmp;
size_t clamp_max = std::conditional<
sizeof(size_t) == 4, std::integral_constant<size_t, 0xffffffff>,
std::integral_constant<uint64_t, 64ull << 30>>::type::value;
ClipToRange(&result.write_buffer_size, ((size_t)64) << 10, clamp_max);
// if user sets arena_block_size, we trust user to use this value. Otherwise,
// calculate a proper value from writer_buffer_size;
if (result.arena_block_size <= 0) {
result.arena_block_size = result.write_buffer_size / 8;
// Align up to 4k
const size_t align = 4 * 1024;
result.arena_block_size =
((result.arena_block_size + align - 1) / align) * align;
}
result.min_write_buffer_number_to_merge =
std::min(result.min_write_buffer_number_to_merge,
result.max_write_buffer_number - 1);
if (result.num_levels < 1) {
result.num_levels = 1;
}
if (result.compaction_style == kCompactionStyleLevel &&
result.num_levels < 2) {
result.num_levels = 2;
}
if (result.max_write_buffer_number < 2) {
result.max_write_buffer_number = 2;
}
Support saving history in memtable_list 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
10 years ago
if (result.max_write_buffer_number_to_maintain < 0) {
result.max_write_buffer_number_to_maintain = result.max_write_buffer_number;
}
XFUNC_TEST("memtablelist_history", "transaction_xftest_SanitizeOptions",
xf_transaction_set_memtable_history1,
xf_transaction_set_memtable_history,
&result.max_write_buffer_number_to_maintain);
XFUNC_TEST("memtablelist_history_clear", "transaction_xftest_SanitizeOptions",
xf_transaction_clear_memtable_history1,
xf_transaction_clear_memtable_history,
&result.max_write_buffer_number_to_maintain);
if (!result.prefix_extractor) {
assert(result.memtable_factory);
Slice name = result.memtable_factory->Name();
if (name.compare("HashSkipListRepFactory") == 0 ||
name.compare("HashLinkListRepFactory") == 0) {
result.memtable_factory = std::make_shared<SkipListFactory>();
}
}
if (result.compaction_style == kCompactionStyleFIFO) {
result.num_levels = 1;
// since we delete level0 files in FIFO compaction when there are too many
// of them, these options don't really mean anything
result.level0_file_num_compaction_trigger = std::numeric_limits<int>::max();
result.level0_slowdown_writes_trigger = std::numeric_limits<int>::max();
result.level0_stop_writes_trigger = std::numeric_limits<int>::max();
}
if (result.level0_file_num_compaction_trigger == 0) {
Warn(db_options.info_log.get(),
"level0_file_num_compaction_trigger cannot be 0");
result.level0_file_num_compaction_trigger = 1;
}
if (result.level0_stop_writes_trigger <
result.level0_slowdown_writes_trigger ||
result.level0_slowdown_writes_trigger <
result.level0_file_num_compaction_trigger) {
Warn(db_options.info_log.get(),
"This condition must be satisfied: "
"level0_stop_writes_trigger(%d) >= "
"level0_slowdown_writes_trigger(%d) >= "
"level0_file_num_compaction_trigger(%d)",
result.level0_stop_writes_trigger,
result.level0_slowdown_writes_trigger,
result.level0_file_num_compaction_trigger);
if (result.level0_slowdown_writes_trigger <
result.level0_file_num_compaction_trigger) {
result.level0_slowdown_writes_trigger =
result.level0_file_num_compaction_trigger;
}
if (result.level0_stop_writes_trigger <
result.level0_slowdown_writes_trigger) {
result.level0_stop_writes_trigger = result.level0_slowdown_writes_trigger;
}
Warn(db_options.info_log.get(),
"Adjust the value to "
"level0_stop_writes_trigger(%d)"
"level0_slowdown_writes_trigger(%d)"
"level0_file_num_compaction_trigger(%d)",
result.level0_stop_writes_trigger,
result.level0_slowdown_writes_trigger,
result.level0_file_num_compaction_trigger);
}
if (result.soft_pending_compaction_bytes_limit == 0) {
result.soft_pending_compaction_bytes_limit =
result.hard_pending_compaction_bytes_limit;
} else if (result.hard_pending_compaction_bytes_limit > 0 &&
result.soft_pending_compaction_bytes_limit >
result.hard_pending_compaction_bytes_limit) {
result.soft_pending_compaction_bytes_limit =
result.hard_pending_compaction_bytes_limit;
}
if (result.level_compaction_dynamic_level_bytes) {
if (result.compaction_style != kCompactionStyleLevel ||
db_options.db_paths.size() > 1U) {
// 1. level_compaction_dynamic_level_bytes only makes sense for
// level-based compaction.
// 2. we don't yet know how to make both of this feature and multiple
// DB path work.
result.level_compaction_dynamic_level_bytes = false;
}
}
return result;
}
int SuperVersion::dummy = 0;
void* const SuperVersion::kSVInUse = &SuperVersion::dummy;
void* const SuperVersion::kSVObsolete = nullptr;
SuperVersion::~SuperVersion() {
for (auto td : to_delete) {
delete td;
}
}
SuperVersion* SuperVersion::Ref() {
refs.fetch_add(1, std::memory_order_relaxed);
return this;
}
bool SuperVersion::Unref() {
// fetch_sub returns the previous value of ref
uint32_t previous_refs = refs.fetch_sub(1);
assert(previous_refs > 0);
return previous_refs == 1;
}
void SuperVersion::Cleanup() {
assert(refs.load(std::memory_order_relaxed) == 0);
imm->Unref(&to_delete);
MemTable* m = mem->Unref();
if (m != nullptr) {
auto* memory_usage = current->cfd()->imm()->current_memory_usage();
assert(*memory_usage >= m->ApproximateMemoryUsage());
*memory_usage -= m->ApproximateMemoryUsage();
to_delete.push_back(m);
}
current->Unref();
}
void SuperVersion::Init(MemTable* new_mem, MemTableListVersion* new_imm,
Version* new_current) {
mem = new_mem;
imm = new_imm;
current = new_current;
mem->Ref();
imm->Ref();
current->Ref();
refs.store(1, std::memory_order_relaxed);
}
namespace {
void SuperVersionUnrefHandle(void* ptr) {
// UnrefHandle is called when a thread exists or a ThreadLocalPtr gets
// destroyed. When former happens, the thread shouldn't see kSVInUse.
// When latter happens, we are in ~ColumnFamilyData(), no get should happen as
// well.
SuperVersion* sv = static_cast<SuperVersion*>(ptr);
if (sv->Unref()) {
sv->db_mutex->Lock();
sv->Cleanup();
sv->db_mutex->Unlock();
delete sv;
}
}
} // anonymous namespace
Rewritten system for scheduling background work 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
10 years ago
ColumnFamilyData::ColumnFamilyData(
uint32_t id, const std::string& name, Version* _dummy_versions,
Cache* _table_cache, WriteBuffer* write_buffer,
const ColumnFamilyOptions& cf_options, const DBOptions* db_options,
const EnvOptions& env_options, ColumnFamilySet* column_family_set)
: id_(id),
name_(name),
dummy_versions_(_dummy_versions),
current_(nullptr),
refs_(0),
dropped_(false),
internal_comparator_(cf_options.comparator),
options_(*db_options,
SanitizeOptions(*db_options, &internal_comparator_, cf_options)),
ioptions_(options_),
mutable_cf_options_(options_, ioptions_),
write_buffer_(write_buffer),
mem_(nullptr),
Support saving history in memtable_list 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
10 years ago
imm_(options_.min_write_buffer_number_to_merge,
options_.max_write_buffer_number_to_maintain),
super_version_(nullptr),
super_version_number_(0),
local_sv_(new ThreadLocalPtr(&SuperVersionUnrefHandle)),
next_(nullptr),
prev_(nullptr),
log_number_(0),
Rewritten system for scheduling background work 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
10 years ago
column_family_set_(column_family_set),
pending_flush_(false),
pending_compaction_(false),
prev_compaction_needed_bytes_(0) {
Ref();
// Convert user defined table properties collector factories to internal ones.
GetIntTblPropCollectorFactory(options_, &int_tbl_prop_collector_factories_);
// if _dummy_versions is nullptr, then this is a dummy column family.
if (_dummy_versions != nullptr) {
make internal stats independent of statistics Summary: also make it aware of column family output from db_bench ``` ** Compaction Stats [default] ** Level Files Size(MB) Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) RW-Amp W-Amp Rd(MB/s) Wr(MB/s) Rn(cnt) Rnp1(cnt) Wnp1(cnt) Wnew(cnt) Comp(sec) Comp(cnt) Avg(sec) Stall(sec) Stall(cnt) Avg(ms) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 14 956 0.9 0.0 0.0 0.0 2.7 2.7 0.0 0.0 0.0 111.6 0 0 0 0 24 40 0.612 75.20 492387 0.15 L1 21 2001 2.0 5.7 2.0 3.7 5.3 1.6 5.4 2.6 71.2 65.7 31 43 55 12 82 2 41.242 43.72 41183 1.06 L2 217 18974 1.9 16.5 2.0 14.4 15.1 0.7 15.6 7.4 70.1 64.3 17 182 185 3 241 16 15.052 0.00 0 0.00 L3 1641 188245 1.8 9.1 1.1 8.0 8.5 0.5 15.4 7.4 61.3 57.2 9 75 76 1 152 9 16.887 0.00 0 0.00 L4 4447 449025 0.4 13.4 4.8 8.6 9.1 0.5 4.7 1.9 77.8 52.7 38 79 100 21 176 38 4.639 0.00 0 0.00 Sum 6340 659201 0.0 44.7 10.0 34.7 40.6 6.0 32.0 15.2 67.7 61.6 95 379 416 37 676 105 6.439 118.91 533570 0.22 Int 0 0 0.0 1.2 0.4 0.8 1.3 0.5 5.2 2.7 59.1 65.6 3 7 9 2 20 10 2.003 0.00 0 0.00 Stalls(secs): 75.197 level0_slowdown, 0.000 level0_numfiles, 0.000 memtable_compaction, 43.717 leveln_slowdown Stalls(count): 492387 level0_slowdown, 0 level0_numfiles, 0 memtable_compaction, 41183 leveln_slowdown ** DB Stats ** Uptime(secs): 202.1 total, 13.5 interval Cumulative writes: 6291456 writes, 6291456 batches, 1.0 writes per batch, 4.90 ingest GB Cumulative WAL: 6291456 writes, 6291456 syncs, 1.00 writes per sync, 4.90 GB written Interval writes: 1048576 writes, 1048576 batches, 1.0 writes per batch, 836.0 ingest MB Interval WAL: 1048576 writes, 1048576 syncs, 1.00 writes per sync, 0.82 MB written Test Plan: ran it Reviewers: sdong, yhchiang, igor Reviewed By: igor Subscribers: leveldb Differential Revision: https://reviews.facebook.net/D19917
11 years ago
internal_stats_.reset(
new InternalStats(ioptions_.num_levels, db_options->env, this));
table_cache_.reset(new TableCache(ioptions_, env_options, _table_cache));
if (ioptions_.compaction_style == kCompactionStyleLevel) {
compaction_picker_.reset(
new LevelCompactionPicker(ioptions_, &internal_comparator_));
#ifndef ROCKSDB_LITE
} else if (ioptions_.compaction_style == kCompactionStyleUniversal) {
compaction_picker_.reset(
new UniversalCompactionPicker(ioptions_, &internal_comparator_));
} else if (ioptions_.compaction_style == kCompactionStyleFIFO) {
compaction_picker_.reset(
new FIFOCompactionPicker(ioptions_, &internal_comparator_));
} else if (ioptions_.compaction_style == kCompactionStyleNone) {
compaction_picker_.reset(new NullCompactionPicker(
ioptions_, &internal_comparator_));
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"Column family %s does not use any background compaction. "
"Compactions can only be done via CompactFiles\n",
GetName().c_str());
#endif // !ROCKSDB_LITE
} else {
Log(InfoLogLevel::ERROR_LEVEL, ioptions_.info_log,
"Unable to recognize the specified compaction style %d. "
"Column family %s will use kCompactionStyleLevel.\n",
ioptions_.compaction_style, GetName().c_str());
compaction_picker_.reset(
new LevelCompactionPicker(ioptions_, &internal_comparator_));
}
Rewritten system for scheduling background work 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
10 years ago
if (column_family_set_->NumberOfColumnFamilies() < 10) {
Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
"--------------- Options for column family [%s]:\n", name.c_str());
options_.DumpCFOptions(ioptions_.info_log);
Rewritten system for scheduling background work 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
10 years ago
} else {
Log(InfoLogLevel::INFO_LEVEL, ioptions_.info_log,
"\t(skipping printing options)\n");
}
}
RecalculateWriteStallConditions(mutable_cf_options_);
}
// DB mutex held
ColumnFamilyData::~ColumnFamilyData() {
assert(refs_.load(std::memory_order_relaxed) == 0);
// remove from linked list
auto prev = prev_;
auto next = next_;
prev->next_ = next;
next->prev_ = prev;
if (!dropped_ && column_family_set_ != nullptr) {
// If it's dropped, it's already removed from column family set
// If column_family_set_ == nullptr, this is dummy CFD and not in
// ColumnFamilySet
column_family_set_->RemoveColumnFamily(this);
}
if (current_ != nullptr) {
current_->Unref();
}
Rewritten system for scheduling background work 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
10 years ago
// It would be wrong if this ColumnFamilyData is in flush_queue_ or
// compaction_queue_ and we destroyed it
assert(!pending_flush_);
assert(!pending_compaction_);
if (super_version_ != nullptr) {
// Release SuperVersion reference kept in ThreadLocalPtr.
// This must be done outside of mutex_ since unref handler can lock mutex.
super_version_->db_mutex->Unlock();
local_sv_.reset();
super_version_->db_mutex->Lock();
bool is_last_reference __attribute__((unused));
is_last_reference = super_version_->Unref();
assert(is_last_reference);
super_version_->Cleanup();
delete super_version_;
super_version_ = nullptr;
}
if (dummy_versions_ != nullptr) {
// List must be empty
assert(dummy_versions_->TEST_Next() == dummy_versions_);
bool deleted __attribute__((unused)) = dummy_versions_->Unref();
assert(deleted);
}
if (mem_ != nullptr) {
delete mem_->Unref();
}
autovector<MemTable*> to_delete;
imm_.current()->Unref(&to_delete);
for (MemTable* m : to_delete) {
delete m;
}
}
void ColumnFamilyData::SetDropped() {
// can't drop default CF
assert(id_ != 0);
dropped_ = true;
write_controller_token_.reset();
// remove from column_family_set
column_family_set_->RemoveColumnFamily(this);
}
const double kSlowdownRatio = 1.2;
namespace {
std::unique_ptr<WriteControllerToken> SetupDelay(
uint64_t max_write_rate, WriteController* write_controller,
uint64_t compaction_needed_bytes, uint64_t prev_compaction_neeed_bytes,
bool auto_comapctions_disabled) {
const uint64_t kMinWriteRate = 1024u; // Minimum write rate 1KB/s.
uint64_t write_rate = write_controller->delayed_write_rate();
if (auto_comapctions_disabled) {
// When auto compaction is disabled, always use the value user gave.
write_rate = max_write_rate;
} else if (write_controller->NeedsDelay() && max_write_rate > kMinWriteRate) {
// If user gives rate less than kMinWriteRate, don't adjust it.
//
// If already delayed, need to adjust based on previous compaction debt.
// When there are two or more column families require delay, we always
// increase or reduce write rate based on information for one single
// column family. It is likely to be OK but we can improve if there is a
// problem.
// Ignore compaction_needed_bytes = 0 case because compaction_needed_bytes
// is only available in level-based compaction
//
// If the compaction debt stays the same as previously, we also further slow
// down. It usually means a mem table is full. It's mainly for the case
// where both of flush and compaction are much slower than the speed we
// insert to mem tables, so we need to actively slow down before we get
// feedback signal from compaction and flushes to avoid the full stop
// because of hitting the max write buffer number.
if (prev_compaction_neeed_bytes > 0 &&
prev_compaction_neeed_bytes <= compaction_needed_bytes) {
write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) /
kSlowdownRatio);
if (write_rate < kMinWriteRate) {
write_rate = kMinWriteRate;
}
} else if (prev_compaction_neeed_bytes > compaction_needed_bytes) {
// We are speeding up by ratio of kSlowdownRatio when we have paid
// compaction debt. But we'll never speed up to faster than the write rate
// given by users.
write_rate = static_cast<uint64_t>(static_cast<double>(write_rate) *
kSlowdownRatio);
if (write_rate > max_write_rate) {
write_rate = max_write_rate;
}
}
}
return write_controller->GetDelayToken(write_rate);
}
int GetL0ThresholdSpeedupCompaction(int level0_file_num_compaction_trigger,
int level0_slowdown_writes_trigger) {
// SanitizeOptions() ensures it.
assert(level0_file_num_compaction_trigger <= level0_slowdown_writes_trigger);
// 1/4 of the way between L0 compaction trigger threshold and slowdown
// condition.
// Or twice as compaction trigger, if it is smaller.
return std::min(level0_file_num_compaction_trigger * 2,
level0_file_num_compaction_trigger +
(level0_slowdown_writes_trigger -
level0_file_num_compaction_trigger) /
4);
}
} // namespace
void ColumnFamilyData::RecalculateWriteStallConditions(
const MutableCFOptions& mutable_cf_options) {
if (current_ != nullptr) {
auto* vstorage = current_->storage_info();
auto write_controller = column_family_set_->write_controller_;
uint64_t compaction_needed_bytes =
vstorage->estimated_compaction_needed_bytes();
Support saving history in memtable_list 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
10 years ago
if (imm()->NumNotFlushed() >= mutable_cf_options.max_write_buffer_number) {
write_controller_token_ = write_controller->GetStopToken();
internal_stats_->AddCFStats(InternalStats::MEMTABLE_COMPACTION, 1);
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Stopping writes because we have %d immutable memtables "
"(waiting for flush), max_write_buffer_number is set to %d",
Support saving history in memtable_list 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
10 years ago
name_.c_str(), imm()->NumNotFlushed(),
mutable_cf_options.max_write_buffer_number);
} else if (vstorage->l0_delay_trigger_count() >=
mutable_cf_options.level0_stop_writes_trigger) {
write_controller_token_ = write_controller->GetStopToken();
internal_stats_->AddCFStats(InternalStats::LEVEL0_NUM_FILES_TOTAL, 1);
if (compaction_picker_->IsLevel0CompactionInProgress()) {
internal_stats_->AddCFStats(
InternalStats::LEVEL0_NUM_FILES_WITH_COMPACTION, 1);
}
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Stopping writes because we have %d level-0 files",
name_.c_str(), vstorage->l0_delay_trigger_count());
} else if (mutable_cf_options.hard_pending_compaction_bytes_limit > 0 &&
compaction_needed_bytes >=
mutable_cf_options.hard_pending_compaction_bytes_limit) {
write_controller_token_ = write_controller->GetStopToken();
internal_stats_->AddCFStats(
InternalStats::HARD_PENDING_COMPACTION_BYTES_LIMIT, 1);
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Stopping writes because of estimated pending compaction "
"bytes %" PRIu64,
name_.c_str(), compaction_needed_bytes);
} else if (mutable_cf_options.max_write_buffer_number > 3 &&
imm()->NumNotFlushed() >=
mutable_cf_options.max_write_buffer_number - 1) {
write_controller_token_ =
SetupDelay(ioptions_.delayed_write_rate, write_controller,
compaction_needed_bytes, prev_compaction_needed_bytes_,
mutable_cf_options.disable_auto_compactions);
internal_stats_->AddCFStats(InternalStats::MEMTABLE_SLOWDOWN, 1);
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Stalling writes because we have %d immutable memtables "
"(waiting for flush), max_write_buffer_number is set to %d "
"rate %" PRIu64,
name_.c_str(), imm()->NumNotFlushed(),
mutable_cf_options.max_write_buffer_number,
write_controller->delayed_write_rate());
} else if (mutable_cf_options.level0_slowdown_writes_trigger >= 0 &&
vstorage->l0_delay_trigger_count() >=
mutable_cf_options.level0_slowdown_writes_trigger) {
write_controller_token_ =
SetupDelay(ioptions_.delayed_write_rate, write_controller,
compaction_needed_bytes, prev_compaction_needed_bytes_,
mutable_cf_options.disable_auto_compactions);
internal_stats_->AddCFStats(InternalStats::LEVEL0_SLOWDOWN_TOTAL, 1);
if (compaction_picker_->IsLevel0CompactionInProgress()) {
internal_stats_->AddCFStats(
InternalStats::LEVEL0_SLOWDOWN_WITH_COMPACTION, 1);
}
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Stalling writes because we have %d level-0 files "
"rate %" PRIu64,
name_.c_str(), vstorage->l0_delay_trigger_count(),
write_controller->delayed_write_rate());
} else if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0 &&
vstorage->estimated_compaction_needed_bytes() >=
mutable_cf_options.soft_pending_compaction_bytes_limit) {
write_controller_token_ =
SetupDelay(ioptions_.delayed_write_rate, write_controller,
compaction_needed_bytes, prev_compaction_needed_bytes_,
mutable_cf_options.disable_auto_compactions);
internal_stats_->AddCFStats(
InternalStats::SOFT_PENDING_COMPACTION_BYTES_LIMIT, 1);
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Stalling writes because of estimated pending compaction "
"bytes %" PRIu64 " rate %" PRIu64,
name_.c_str(), vstorage->estimated_compaction_needed_bytes(),
write_controller->delayed_write_rate());
} else if (vstorage->l0_delay_trigger_count() >=
GetL0ThresholdSpeedupCompaction(
mutable_cf_options.level0_file_num_compaction_trigger,
mutable_cf_options.level0_slowdown_writes_trigger)) {
write_controller_token_ = write_controller->GetCompactionPressureToken();
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Increasing compaction threads because we have %d level-0 "
"files ",
name_.c_str(), vstorage->l0_delay_trigger_count());
} else if (vstorage->estimated_compaction_needed_bytes() >=
mutable_cf_options.soft_pending_compaction_bytes_limit / 4) {
// Increase compaction threads if bytes needed for compaction exceeds
// 1/4 of threshold for slowing down.
// If soft pending compaction byte limit is not set, always speed up
// compaction.
write_controller_token_ = write_controller->GetCompactionPressureToken();
if (mutable_cf_options.soft_pending_compaction_bytes_limit > 0) {
Log(InfoLogLevel::WARN_LEVEL, ioptions_.info_log,
"[%s] Increasing compaction threads because of estimated pending "
"compaction "
"bytes %" PRIu64,
name_.c_str(), vstorage->estimated_compaction_needed_bytes());
}
} else {
write_controller_token_.reset();
}
prev_compaction_needed_bytes_ = compaction_needed_bytes;
}
}
const EnvOptions* ColumnFamilyData::soptions() const {
return &(column_family_set_->env_options_);
}
void ColumnFamilyData::SetCurrent(Version* current_version) {
current_ = current_version;
}
uint64_t ColumnFamilyData::GetNumLiveVersions() const {
return VersionSet::GetNumLiveVersions(dummy_versions_);
}
uint64_t ColumnFamilyData::GetTotalSstFilesSize() const {
return VersionSet::GetTotalSstFilesSize(dummy_versions_);
}
MemTable* ColumnFamilyData::ConstructNewMemtable(
const MutableCFOptions& mutable_cf_options, SequenceNumber earliest_seq) {
assert(current_ != nullptr);
return new MemTable(internal_comparator_, ioptions_, mutable_cf_options,
write_buffer_, earliest_seq);
}
void ColumnFamilyData::CreateNewMemtable(
const MutableCFOptions& mutable_cf_options, SequenceNumber earliest_seq) {
if (mem_ != nullptr) {
delete mem_->Unref();
}
SetMemtable(ConstructNewMemtable(mutable_cf_options, earliest_seq));
mem_->Ref();
}
Rewritten system for scheduling background work 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
10 years ago
bool ColumnFamilyData::NeedsCompaction() const {
return compaction_picker_->NeedsCompaction(current_->storage_info());
}
Compaction* ColumnFamilyData::PickCompaction(
const MutableCFOptions& mutable_options, LogBuffer* log_buffer) {
auto* result = compaction_picker_->PickCompaction(
GetName(), mutable_options, current_->storage_info(), log_buffer);
if (result != nullptr) {
result->SetInputVersion(current_);
}
return result;
}
const int ColumnFamilyData::kCompactAllLevels = -1;
const int ColumnFamilyData::kCompactToBaseLevel = -2;
Compaction* ColumnFamilyData::CompactRange(
const MutableCFOptions& mutable_cf_options, int input_level,
int output_level, uint32_t output_path_id, const InternalKey* begin,
const InternalKey* end, InternalKey** compaction_end, bool* conflict) {
auto* result = compaction_picker_->CompactRange(
GetName(), mutable_cf_options, current_->storage_info(), input_level,
output_level, output_path_id, begin, end, compaction_end, conflict);
if (result != nullptr) {
result->SetInputVersion(current_);
}
return result;
}
SuperVersion* ColumnFamilyData::GetReferencedSuperVersion(
InstrumentedMutex* db_mutex) {
SuperVersion* sv = nullptr;
sv = GetThreadLocalSuperVersion(db_mutex);
sv->Ref();
if (!ReturnThreadLocalSuperVersion(sv)) {
sv->Unref();
}
return sv;
}
SuperVersion* ColumnFamilyData::GetThreadLocalSuperVersion(
InstrumentedMutex* db_mutex) {
SuperVersion* sv = nullptr;
// The SuperVersion is cached in thread local storage to avoid acquiring
// mutex when SuperVersion does not change since the last use. When a new
// SuperVersion is installed, the compaction or flush thread cleans up
// cached SuperVersion in all existing thread local storage. To avoid
// acquiring mutex for this operation, we use atomic Swap() on the thread
// local pointer to guarantee exclusive access. If the thread local pointer
// is being used while a new SuperVersion is installed, the cached
// SuperVersion can become stale. In that case, the background thread would
// have swapped in kSVObsolete. We re-check the value at when returning
// SuperVersion back to thread local, with an atomic compare and swap.
// The superversion will need to be released if detected to be stale.
void* ptr = local_sv_->Swap(SuperVersion::kSVInUse);
// Invariant:
// (1) Scrape (always) installs kSVObsolete in ThreadLocal storage
// (2) the Swap above (always) installs kSVInUse, ThreadLocal storage
// should only keep kSVInUse before ReturnThreadLocalSuperVersion call
// (if no Scrape happens).
assert(ptr != SuperVersion::kSVInUse);
sv = static_cast<SuperVersion*>(ptr);
if (sv == SuperVersion::kSVObsolete ||
sv->version_number != super_version_number_.load()) {
RecordTick(ioptions_.statistics, NUMBER_SUPERVERSION_ACQUIRES);
SuperVersion* sv_to_delete = nullptr;
if (sv && sv->Unref()) {
RecordTick(ioptions_.statistics, NUMBER_SUPERVERSION_CLEANUPS);
db_mutex->Lock();
// NOTE: underlying resources held by superversion (sst files) might
// not be released until the next background job.
sv->Cleanup();
sv_to_delete = sv;
} else {
db_mutex->Lock();
}
sv = super_version_->Ref();
db_mutex->Unlock();
delete sv_to_delete;
}
assert(sv != nullptr);
return sv;
}
bool ColumnFamilyData::ReturnThreadLocalSuperVersion(SuperVersion* sv) {
assert(sv != nullptr);
// Put the SuperVersion back
void* expected = SuperVersion::kSVInUse;
if (local_sv_->CompareAndSwap(static_cast<void*>(sv), expected)) {
// When we see kSVInUse in the ThreadLocal, we are sure ThreadLocal
10 years ago
// storage has not been altered and no Scrape has happened. The
// SuperVersion is still current.
return true;
} else {
// ThreadLocal scrape happened in the process of this GetImpl call (after
// thread local Swap() at the beginning and before CompareAndSwap()).
// This means the SuperVersion it holds is obsolete.
assert(expected == SuperVersion::kSVObsolete);
}
return false;
}
SuperVersion* ColumnFamilyData::InstallSuperVersion(
SuperVersion* new_superversion, InstrumentedMutex* db_mutex) {
db_mutex->AssertHeld();
return InstallSuperVersion(new_superversion, db_mutex, mutable_cf_options_);
}
SuperVersion* ColumnFamilyData::InstallSuperVersion(
SuperVersion* new_superversion, InstrumentedMutex* db_mutex,
const MutableCFOptions& mutable_cf_options) {
new_superversion->db_mutex = db_mutex;
new_superversion->mutable_cf_options = mutable_cf_options;
new_superversion->Init(mem_, imm_.current(), current_);
SuperVersion* old_superversion = super_version_;
super_version_ = new_superversion;
++super_version_number_;
super_version_->version_number = super_version_number_;
// Reset SuperVersions cached in thread local storage
ResetThreadLocalSuperVersions();
RecalculateWriteStallConditions(mutable_cf_options);
if (old_superversion != nullptr && old_superversion->Unref()) {
old_superversion->Cleanup();
return old_superversion; // will let caller delete outside of mutex
}
return nullptr;
}
void ColumnFamilyData::ResetThreadLocalSuperVersions() {
autovector<void*> sv_ptrs;
local_sv_->Scrape(&sv_ptrs, SuperVersion::kSVObsolete);
for (auto ptr : sv_ptrs) {
assert(ptr);
if (ptr == SuperVersion::kSVInUse) {
continue;
}
auto sv = static_cast<SuperVersion*>(ptr);
if (sv->Unref()) {
sv->Cleanup();
delete sv;
}
}
}
#ifndef ROCKSDB_LITE
Status ColumnFamilyData::SetOptions(
const std::unordered_map<std::string, std::string>& options_map) {
MutableCFOptions new_mutable_cf_options;
Status s = GetMutableOptionsFromStrings(mutable_cf_options_, options_map,
&new_mutable_cf_options);
if (s.ok()) {
mutable_cf_options_ = new_mutable_cf_options;
mutable_cf_options_.RefreshDerivedOptions(ioptions_);
}
return s;
}
#endif // ROCKSDB_LITE
ColumnFamilySet::ColumnFamilySet(const std::string& dbname,
const DBOptions* db_options,
const EnvOptions& env_options,
Cache* table_cache,
WriteBuffer* write_buffer,
WriteController* write_controller)
: max_column_family_(0),
dummy_cfd_(new ColumnFamilyData(0, "", nullptr, nullptr, nullptr,
ColumnFamilyOptions(), db_options,
env_options, nullptr)),
default_cfd_cache_(nullptr),
db_name_(dbname),
db_options_(db_options),
env_options_(env_options),
table_cache_(table_cache),
write_buffer_(write_buffer),
write_controller_(write_controller) {
// initialize linked list
dummy_cfd_->prev_ = dummy_cfd_;
dummy_cfd_->next_ = dummy_cfd_;
}
ColumnFamilySet::~ColumnFamilySet() {
while (column_family_data_.size() > 0) {
// cfd destructor will delete itself from column_family_data_
auto cfd = column_family_data_.begin()->second;
cfd->Unref();
delete cfd;
}
dummy_cfd_->Unref();
delete dummy_cfd_;
}
ColumnFamilyData* ColumnFamilySet::GetDefault() const {
assert(default_cfd_cache_ != nullptr);
return default_cfd_cache_;
}
ColumnFamilyData* ColumnFamilySet::GetColumnFamily(uint32_t id) const {
auto cfd_iter = column_family_data_.find(id);
if (cfd_iter != column_family_data_.end()) {
return cfd_iter->second;
} else {
return nullptr;
}
}
ColumnFamilyData* ColumnFamilySet::GetColumnFamily(const std::string& name)
const {
auto cfd_iter = column_families_.find(name);
if (cfd_iter != column_families_.end()) {
auto cfd = GetColumnFamily(cfd_iter->second);
assert(cfd != nullptr);
return cfd;
} else {
return nullptr;
}
}
uint32_t ColumnFamilySet::GetNextColumnFamilyID() {
return ++max_column_family_;
}
uint32_t ColumnFamilySet::GetMaxColumnFamily() { return max_column_family_; }
void ColumnFamilySet::UpdateMaxColumnFamily(uint32_t new_max_column_family) {
max_column_family_ = std::max(new_max_column_family, max_column_family_);
}
size_t ColumnFamilySet::NumberOfColumnFamilies() const {
return column_families_.size();
}
// under a DB mutex AND write thread
ColumnFamilyData* ColumnFamilySet::CreateColumnFamily(
const std::string& name, uint32_t id, Version* dummy_versions,
const ColumnFamilyOptions& options) {
assert(column_families_.find(name) == column_families_.end());
ColumnFamilyData* new_cfd =
new ColumnFamilyData(id, name, dummy_versions, table_cache_,
write_buffer_, options, db_options_,
env_options_, this);
column_families_.insert({name, id});
column_family_data_.insert({id, new_cfd});
max_column_family_ = std::max(max_column_family_, id);
// add to linked list
new_cfd->next_ = dummy_cfd_;
auto prev = dummy_cfd_->prev_;
new_cfd->prev_ = prev;
prev->next_ = new_cfd;
dummy_cfd_->prev_ = new_cfd;
if (id == 0) {
default_cfd_cache_ = new_cfd;
}
return new_cfd;
}
// REQUIRES: DB mutex held
void ColumnFamilySet::FreeDeadColumnFamilies() {
autovector<ColumnFamilyData*> to_delete;
for (auto cfd = dummy_cfd_->next_; cfd != dummy_cfd_; cfd = cfd->next_) {
if (cfd->refs_.load(std::memory_order_relaxed) == 0) {
to_delete.push_back(cfd);
}
}
for (auto cfd : to_delete) {
// this is very rare, so it's not a problem that we do it under a mutex
delete cfd;
}
}
// under a DB mutex AND from a write thread
void ColumnFamilySet::RemoveColumnFamily(ColumnFamilyData* cfd) {
auto cfd_iter = column_family_data_.find(cfd->GetID());
assert(cfd_iter != column_family_data_.end());
column_family_data_.erase(cfd_iter);
column_families_.erase(cfd->GetName());
}
// under a DB mutex OR from a write thread
bool ColumnFamilyMemTablesImpl::Seek(uint32_t column_family_id) {
if (column_family_id == 0) {
// optimization for common case
current_ = column_family_set_->GetDefault();
} else {
current_ = column_family_set_->GetColumnFamily(column_family_id);
}
handle_.SetCFD(current_);
return current_ != nullptr;
}
uint64_t ColumnFamilyMemTablesImpl::GetLogNumber() const {
assert(current_ != nullptr);
return current_->GetLogNumber();
}
MemTable* ColumnFamilyMemTablesImpl::GetMemTable() const {
assert(current_ != nullptr);
return current_->mem();
}
ColumnFamilyHandle* ColumnFamilyMemTablesImpl::GetColumnFamilyHandle() {
assert(current_ != nullptr);
return &handle_;
}
uint32_t GetColumnFamilyID(ColumnFamilyHandle* column_family) {
uint32_t column_family_id = 0;
if (column_family != nullptr) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
column_family_id = cfh->GetID();
}
return column_family_id;
}
const Comparator* GetColumnFamilyUserComparator(
ColumnFamilyHandle* column_family) {
if (column_family != nullptr) {
auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
return cfh->user_comparator();
}
return nullptr;
}
} // namespace rocksdb