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/db_impl/db_impl_compaction_flush.cc

3816 lines
147 KiB

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root 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 <cinttypes>
#include <deque>
#include "db/builder.h"
#include "db/db_impl/db_impl.h"
#include "db/error_handler.h"
#include "db/event_helpers.h"
#include "file/sst_file_manager_impl.h"
#include "logging/logging.h"
#include "monitoring/iostats_context_imp.h"
#include "monitoring/perf_context_imp.h"
#include "monitoring/thread_status_updater.h"
#include "monitoring/thread_status_util.h"
#include "test_util/sync_point.h"
#include "util/cast_util.h"
#include "util/concurrent_task_limiter_impl.h"
namespace ROCKSDB_NAMESPACE {
bool DBImpl::EnoughRoomForCompaction(
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
ColumnFamilyData* cfd, const std::vector<CompactionInputFiles>& inputs,
bool* sfm_reserved_compact_space, LogBuffer* log_buffer) {
// Check if we have enough room to do the compaction
bool enough_room = true;
#ifndef ROCKSDB_LITE
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
if (sfm) {
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// Pass the current bg_error_ to SFM so it can decide what checks to
// perform. If this DB instance hasn't seen any error yet, the SFM can be
// optimistic and not do disk space checks
Status bg_error = error_handler_.GetBGError();
enough_room = sfm->EnoughRoomForCompaction(cfd, inputs, bg_error);
bg_error.PermitUncheckedError(); // bg_error is just a copy of the Status
// from the error_handler_
if (enough_room) {
*sfm_reserved_compact_space = true;
}
}
#else
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
(void)cfd;
(void)inputs;
(void)sfm_reserved_compact_space;
#endif // ROCKSDB_LITE
if (!enough_room) {
// Just in case tests want to change the value of enough_room
TEST_SYNC_POINT_CALLBACK(
"DBImpl::BackgroundCompaction():CancelledCompaction", &enough_room);
ROCKS_LOG_BUFFER(log_buffer,
"Cancelled compaction because not enough room");
RecordTick(stats_, COMPACTION_CANCELLED, 1);
}
return enough_room;
}
bool DBImpl::RequestCompactionToken(ColumnFamilyData* cfd, bool force,
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
std::unique_ptr<TaskLimiterToken>* token,
LogBuffer* log_buffer) {
assert(*token == nullptr);
auto limiter = static_cast<ConcurrentTaskLimiterImpl*>(
cfd->ioptions()->compaction_thread_limiter.get());
if (limiter == nullptr) {
return true;
}
*token = limiter->GetToken(force);
if (*token != nullptr) {
ROCKS_LOG_BUFFER(log_buffer,
"Thread limiter [%s] increase [%s] compaction task, "
"force: %s, tasks after: %d",
limiter->GetName().c_str(), cfd->GetName().c_str(),
force ? "true" : "false", limiter->GetOutstandingTask());
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
return true;
}
return false;
}
IOStatus DBImpl::SyncClosedLogs(JobContext* job_context) {
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Start");
mutex_.AssertHeld();
autovector<log::Writer*, 1> logs_to_sync;
uint64_t current_log_number = logfile_number_;
while (logs_.front().number < current_log_number &&
logs_.front().IsSyncing()) {
log_sync_cv_.Wait();
}
for (auto it = logs_.begin();
it != logs_.end() && it->number < current_log_number; ++it) {
auto& log = *it;
log.PrepareForSync();
logs_to_sync.push_back(log.writer);
}
IOStatus io_s;
if (!logs_to_sync.empty()) {
mutex_.Unlock();
assert(job_context);
for (log::Writer* log : logs_to_sync) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[JOB %d] Syncing log #%" PRIu64, job_context->job_id,
log->get_log_number());
io_s = log->file()->Sync(immutable_db_options_.use_fsync);
if (!io_s.ok()) {
break;
}
if (immutable_db_options_.recycle_log_file_num > 0) {
io_s = log->Close();
if (!io_s.ok()) {
break;
}
}
}
if (io_s.ok()) {
io_s = directories_.GetWalDir()->FsyncWithDirOptions(
IOOptions(), nullptr,
DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
}
TEST_SYNC_POINT_CALLBACK("DBImpl::SyncClosedLogs:BeforeReLock",
/*arg=*/nullptr);
mutex_.Lock();
// "number <= current_log_number - 1" is equivalent to
// "number < current_log_number".
if (io_s.ok()) {
io_s = status_to_io_status(MarkLogsSynced(current_log_number - 1, true));
} else {
MarkLogsNotSynced(current_log_number - 1);
}
if (!io_s.ok()) {
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Failed");
return io_s;
}
}
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:end");
return io_s;
}
Status DBImpl::FlushMemTableToOutputFile(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
bool* made_progress, JobContext* job_context,
SuperVersionContext* superversion_context,
std::vector<SequenceNumber>& snapshot_seqs,
SequenceNumber earliest_write_conflict_snapshot,
SnapshotChecker* snapshot_checker, LogBuffer* log_buffer,
Env::Priority thread_pri) {
mutex_.AssertHeld();
assert(cfd);
assert(cfd->imm());
assert(cfd->imm()->NumNotFlushed() != 0);
assert(cfd->imm()->IsFlushPending());
assert(versions_);
assert(versions_->GetColumnFamilySet());
// If there are more than one column families, we need to make sure that
// all the log files except the most recent one are synced. Otherwise if
// the host crashes after flushing and before WAL is persistent, the
// flushed SST may contain data from write batches whose updates to
// other (unflushed) column families are missing.
const bool needs_to_sync_closed_wals =
logfile_number_ > 0 &&
versions_->GetColumnFamilySet()->NumberOfColumnFamilies() > 1;
// If needs_to_sync_closed_wals is true, we need to record the current
// maximum memtable ID of this column family so that a later PickMemtables()
// call will not pick memtables whose IDs are higher. This is due to the fact
// that SyncClosedLogs() may release the db mutex, and memtable switch can
// happen for this column family in the meantime. The newly created memtables
// have their data backed by unsynced WALs, thus they cannot be included in
// this flush job.
// Another reason why we must record the current maximum memtable ID of this
// column family: SyncClosedLogs() may release db mutex, thus it's possible
// for application to continue to insert into memtables increasing db's
// sequence number. The application may take a snapshot, but this snapshot is
// not included in `snapshot_seqs` which will be passed to flush job because
// `snapshot_seqs` has already been computed before this function starts.
// Recording the max memtable ID ensures that the flush job does not flush
// a memtable without knowing such snapshot(s).
uint64_t max_memtable_id = needs_to_sync_closed_wals
? cfd->imm()->GetLatestMemTableID()
: std::numeric_limits<uint64_t>::max();
// If needs_to_sync_closed_wals is false, then the flush job will pick ALL
// existing memtables of the column family when PickMemTable() is called
// later. Although we won't call SyncClosedLogs() in this case, we may still
// call the callbacks of the listeners, i.e. NotifyOnFlushBegin() which also
// releases and re-acquires the db mutex. In the meantime, the application
// can still insert into the memtables and increase the db's sequence number.
// The application can take a snapshot, hoping that the latest visible state
// to this snapshto is preserved. This is hard to guarantee since db mutex
// not held. This newly-created snapshot is not included in `snapshot_seqs`
// and the flush job is unaware of its presence. Consequently, the flush job
// may drop certain keys when generating the L0, causing incorrect data to be
// returned for snapshot read using this snapshot.
// To address this, we make sure NotifyOnFlushBegin() executes after memtable
// picking so that no new snapshot can be taken between the two functions.
FlushJob flush_job(
dbname_, cfd, immutable_db_options_, mutable_cf_options, max_memtable_id,
file_options_for_compaction_, versions_.get(), &mutex_, &shutting_down_,
snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
job_context, log_buffer, directories_.GetDbDir(), GetDataDir(cfd, 0U),
GetCompressionFlush(*cfd->ioptions(), mutable_cf_options), stats_,
&event_logger_, mutable_cf_options.report_bg_io_stats,
true /* sync_output_directory */, true /* write_manifest */, thread_pri,
io_tracer_, seqno_time_mapping_, db_id_, db_session_id_,
cfd->GetFullHistoryTsLow(), &blob_callback_);
FileMetaData file_meta;
Status s;
bool need_cancel = false;
IOStatus log_io_s = IOStatus::OK();
if (needs_to_sync_closed_wals) {
// SyncClosedLogs() may unlock and re-lock the db_mutex.
log_io_s = SyncClosedLogs(job_context);
if (!log_io_s.ok() && !log_io_s.IsShutdownInProgress() &&
!log_io_s.IsColumnFamilyDropped()) {
error_handler_.SetBGError(log_io_s, BackgroundErrorReason::kFlush);
}
} else {
TEST_SYNC_POINT("DBImpl::SyncClosedLogs:Skip");
}
s = log_io_s;
// If the log sync failed, we do not need to pick memtable. Otherwise,
// num_flush_not_started_ needs to be rollback.
TEST_SYNC_POINT("DBImpl::FlushMemTableToOutputFile:BeforePickMemtables");
if (s.ok()) {
flush_job.PickMemTable();
need_cancel = true;
}
TEST_SYNC_POINT_CALLBACK(
"DBImpl::FlushMemTableToOutputFile:AfterPickMemtables", &flush_job);
#ifndef ROCKSDB_LITE
// may temporarily unlock and lock the mutex.
NotifyOnFlushBegin(cfd, &file_meta, mutable_cf_options, job_context->job_id);
#endif // ROCKSDB_LITE
bool switched_to_mempurge = false;
// Within flush_job.Run, rocksdb may call event listener to notify
// file creation and deletion.
//
// Note that flush_job.Run will unlock and lock the db_mutex,
// and EventListener callback will be called when the db_mutex
// is unlocked by the current thread.
if (s.ok()) {
s = flush_job.Run(&logs_with_prep_tracker_, &file_meta,
&switched_to_mempurge);
need_cancel = false;
}
if (!s.ok() && need_cancel) {
flush_job.Cancel();
}
if (s.ok()) {
InstallSuperVersionAndScheduleWork(cfd, superversion_context,
mutable_cf_options);
if (made_progress) {
*made_progress = true;
}
const std::string& column_family_name = cfd->GetName();
Version* const current = cfd->current();
assert(current);
const VersionStorageInfo* const storage_info = current->storage_info();
assert(storage_info);
VersionStorageInfo::LevelSummaryStorage tmp;
ROCKS_LOG_BUFFER(log_buffer, "[%s] Level summary: %s\n",
column_family_name.c_str(),
storage_info->LevelSummary(&tmp));
const auto& blob_files = storage_info->GetBlobFiles();
if (!blob_files.empty()) {
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
3 years ago
assert(blob_files.front());
assert(blob_files.back());
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Blob file summary: head=%" PRIu64 ", tail=%" PRIu64 "\n",
column_family_name.c_str(), blob_files.front()->GetBlobFileNumber(),
blob_files.back()->GetBlobFileNumber());
}
}
if (!s.ok() && !s.IsShutdownInProgress() && !s.IsColumnFamilyDropped()) {
if (log_io_s.ok()) {
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
4 years ago
// Error while writing to MANIFEST.
// In fact, versions_->io_status() can also be the result of renaming
// CURRENT file. With current code, it's just difficult to tell. So just
// be pessimistic and try write to a new MANIFEST.
// TODO: distinguish between MANIFEST write and CURRENT renaming
if (!versions_->io_status().ok()) {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the Manifest write will be map to soft error.
// TODO: kManifestWriteNoWAL and kFlushNoWAL are misleading. Refactor is
// needed.
error_handler_.SetBGError(s,
BackgroundErrorReason::kManifestWriteNoWAL);
} else {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the other SST file write errors will be set as
// kFlushNoWAL.
error_handler_.SetBGError(s, BackgroundErrorReason::kFlushNoWAL);
}
} else {
assert(s == log_io_s);
Status new_bg_error = s;
error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
}
}
// If flush ran smoothly and no mempurge happened
// install new SST file path.
if (s.ok() && (!switched_to_mempurge)) {
#ifndef ROCKSDB_LITE
// may temporarily unlock and lock the mutex.
NotifyOnFlushCompleted(cfd, mutable_cf_options,
flush_job.GetCommittedFlushJobsInfo());
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
if (sfm) {
// Notify sst_file_manager that a new file was added
std::string file_path = MakeTableFileName(
cfd->ioptions()->cf_paths[0].path, file_meta.fd.GetNumber());
// TODO (PR7798). We should only add the file to the FileManager if it
// exists. Otherwise, some tests may fail. Ignore the error in the
// interim.
sfm->OnAddFile(file_path).PermitUncheckedError();
if (sfm->IsMaxAllowedSpaceReached()) {
Status new_bg_error =
Status::SpaceLimit("Max allowed space was reached");
TEST_SYNC_POINT_CALLBACK(
"DBImpl::FlushMemTableToOutputFile:MaxAllowedSpaceReached",
&new_bg_error);
error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
}
}
#endif // ROCKSDB_LITE
}
TEST_SYNC_POINT("DBImpl::FlushMemTableToOutputFile:Finish");
return s;
}
Status DBImpl::FlushMemTablesToOutputFiles(
const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
JobContext* job_context, LogBuffer* log_buffer, Env::Priority thread_pri) {
if (immutable_db_options_.atomic_flush) {
return AtomicFlushMemTablesToOutputFiles(
bg_flush_args, made_progress, job_context, log_buffer, thread_pri);
}
assert(bg_flush_args.size() == 1);
std::vector<SequenceNumber> snapshot_seqs;
SequenceNumber earliest_write_conflict_snapshot;
SnapshotChecker* snapshot_checker;
GetSnapshotContext(job_context, &snapshot_seqs,
&earliest_write_conflict_snapshot, &snapshot_checker);
const auto& bg_flush_arg = bg_flush_args[0];
ColumnFamilyData* cfd = bg_flush_arg.cfd_;
// intentional infrequent copy for each flush
MutableCFOptions mutable_cf_options_copy = *cfd->GetLatestMutableCFOptions();
SuperVersionContext* superversion_context =
bg_flush_arg.superversion_context_;
Status s = FlushMemTableToOutputFile(
cfd, mutable_cf_options_copy, made_progress, job_context,
superversion_context, snapshot_seqs, earliest_write_conflict_snapshot,
snapshot_checker, log_buffer, thread_pri);
return s;
}
/*
* Atomically flushes multiple column families.
*
* For each column family, all memtables with ID smaller than or equal to the
* ID specified in bg_flush_args will be flushed. Only after all column
* families finish flush will this function commit to MANIFEST. If any of the
* column families are not flushed successfully, this function does not have
* any side-effect on the state of the database.
*/
Status DBImpl::AtomicFlushMemTablesToOutputFiles(
const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
JobContext* job_context, LogBuffer* log_buffer, Env::Priority thread_pri) {
mutex_.AssertHeld();
autovector<ColumnFamilyData*> cfds;
for (const auto& arg : bg_flush_args) {
cfds.emplace_back(arg.cfd_);
}
#ifndef NDEBUG
for (const auto cfd : cfds) {
assert(cfd->imm()->NumNotFlushed() != 0);
assert(cfd->imm()->IsFlushPending());
assert(cfd->GetFlushReason() == cfds[0]->GetFlushReason());
}
#endif /* !NDEBUG */
std::vector<SequenceNumber> snapshot_seqs;
SequenceNumber earliest_write_conflict_snapshot;
SnapshotChecker* snapshot_checker;
GetSnapshotContext(job_context, &snapshot_seqs,
&earliest_write_conflict_snapshot, &snapshot_checker);
autovector<FSDirectory*> distinct_output_dirs;
autovector<std::string> distinct_output_dir_paths;
std::vector<std::unique_ptr<FlushJob>> jobs;
std::vector<MutableCFOptions> all_mutable_cf_options;
int num_cfs = static_cast<int>(cfds.size());
all_mutable_cf_options.reserve(num_cfs);
for (int i = 0; i < num_cfs; ++i) {
auto cfd = cfds[i];
FSDirectory* data_dir = GetDataDir(cfd, 0U);
const std::string& curr_path = cfd->ioptions()->cf_paths[0].path;
// Add to distinct output directories if eligible. Use linear search. Since
// the number of elements in the vector is not large, performance should be
// tolerable.
bool found = false;
for (const auto& path : distinct_output_dir_paths) {
if (path == curr_path) {
found = true;
break;
}
}
if (!found) {
distinct_output_dir_paths.emplace_back(curr_path);
distinct_output_dirs.emplace_back(data_dir);
}
all_mutable_cf_options.emplace_back(*cfd->GetLatestMutableCFOptions());
const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.back();
uint64_t max_memtable_id = bg_flush_args[i].max_memtable_id_;
jobs.emplace_back(new FlushJob(
dbname_, cfd, immutable_db_options_, mutable_cf_options,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
max_memtable_id, file_options_for_compaction_, versions_.get(), &mutex_,
&shutting_down_, snapshot_seqs, earliest_write_conflict_snapshot,
snapshot_checker, job_context, log_buffer, directories_.GetDbDir(),
data_dir, GetCompressionFlush(*cfd->ioptions(), mutable_cf_options),
stats_, &event_logger_, mutable_cf_options.report_bg_io_stats,
false /* sync_output_directory */, false /* write_manifest */,
thread_pri, io_tracer_, seqno_time_mapping_, db_id_, db_session_id_,
cfd->GetFullHistoryTsLow(), &blob_callback_));
}
std::vector<FileMetaData> file_meta(num_cfs);
// Use of deque<bool> because vector<bool>
// is specific and doesn't allow &v[i].
std::deque<bool> switched_to_mempurge(num_cfs, false);
Status s;
IOStatus log_io_s = IOStatus::OK();
assert(num_cfs == static_cast<int>(jobs.size()));
#ifndef ROCKSDB_LITE
for (int i = 0; i != num_cfs; ++i) {
const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.at(i);
// may temporarily unlock and lock the mutex.
NotifyOnFlushBegin(cfds[i], &file_meta[i], mutable_cf_options,
job_context->job_id);
}
#endif /* !ROCKSDB_LITE */
if (logfile_number_ > 0) {
// TODO (yanqin) investigate whether we should sync the closed logs for
// single column family case.
log_io_s = SyncClosedLogs(job_context);
if (!log_io_s.ok() && !log_io_s.IsShutdownInProgress() &&
!log_io_s.IsColumnFamilyDropped()) {
if (total_log_size_ > 0) {
error_handler_.SetBGError(log_io_s, BackgroundErrorReason::kFlush);
} else {
// If the WAL is empty, we use different error reason
error_handler_.SetBGError(log_io_s, BackgroundErrorReason::kFlushNoWAL);
}
}
}
s = log_io_s;
// exec_status stores the execution status of flush_jobs as
// <bool /* executed */, Status /* status code */>
autovector<std::pair<bool, Status>> exec_status;
std::vector<bool> pick_status;
for (int i = 0; i != num_cfs; ++i) {
// Initially all jobs are not executed, with status OK.
exec_status.emplace_back(false, Status::OK());
pick_status.push_back(false);
}
if (s.ok()) {
for (int i = 0; i != num_cfs; ++i) {
jobs[i]->PickMemTable();
pick_status[i] = true;
}
}
if (s.ok()) {
assert(switched_to_mempurge.size() ==
static_cast<long unsigned int>(num_cfs));
// TODO (yanqin): parallelize jobs with threads.
for (int i = 1; i != num_cfs; ++i) {
exec_status[i].second =
jobs[i]->Run(&logs_with_prep_tracker_, &file_meta[i],
&(switched_to_mempurge.at(i)));
exec_status[i].first = true;
}
if (num_cfs > 1) {
TEST_SYNC_POINT(
"DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:1");
TEST_SYNC_POINT(
"DBImpl::AtomicFlushMemTablesToOutputFiles:SomeFlushJobsComplete:2");
}
assert(exec_status.size() > 0);
assert(!file_meta.empty());
exec_status[0].second = jobs[0]->Run(
&logs_with_prep_tracker_, file_meta.data() /* &file_meta[0] */,
switched_to_mempurge.empty() ? nullptr : &(switched_to_mempurge.at(0)));
exec_status[0].first = true;
Status error_status;
for (const auto& e : exec_status) {
if (!e.second.ok()) {
s = e.second;
if (!e.second.IsShutdownInProgress() &&
!e.second.IsColumnFamilyDropped()) {
// If a flush job did not return OK, and the CF is not dropped, and
// the DB is not shutting down, then we have to return this result to
// caller later.
error_status = e.second;
}
}
}
s = error_status.ok() ? s : error_status;
}
if (s.IsColumnFamilyDropped()) {
s = Status::OK();
}
if (s.ok() || s.IsShutdownInProgress()) {
// Sync on all distinct output directories.
for (auto dir : distinct_output_dirs) {
if (dir != nullptr) {
Status error_status = dir->FsyncWithDirOptions(
IOOptions(), nullptr,
DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
if (!error_status.ok()) {
s = error_status;
break;
}
}
}
} else {
// Need to undo atomic flush if something went wrong, i.e. s is not OK and
// it is not because of CF drop.
// Have to cancel the flush jobs that have NOT executed because we need to
// unref the versions.
for (int i = 0; i != num_cfs; ++i) {
if (pick_status[i] && !exec_status[i].first) {
jobs[i]->Cancel();
}
}
for (int i = 0; i != num_cfs; ++i) {
if (exec_status[i].second.ok() && exec_status[i].first) {
auto& mems = jobs[i]->GetMemTables();
cfds[i]->imm()->RollbackMemtableFlush(mems,
file_meta[i].fd.GetNumber());
}
}
}
if (s.ok()) {
const auto wait_to_install_func =
[&]() -> std::pair<Status, bool /*continue to wait*/> {
if (!versions_->io_status().ok()) {
// Something went wrong elsewhere, we cannot count on waiting for our
// turn to write/sync to MANIFEST or CURRENT. Just return.
return std::make_pair(versions_->io_status(), false);
} else if (shutting_down_.load(std::memory_order_acquire)) {
return std::make_pair(Status::ShutdownInProgress(), false);
}
bool ready = true;
for (size_t i = 0; i != cfds.size(); ++i) {
const auto& mems = jobs[i]->GetMemTables();
if (cfds[i]->IsDropped()) {
// If the column family is dropped, then do not wait.
continue;
} else if (!mems.empty() &&
cfds[i]->imm()->GetEarliestMemTableID() < mems[0]->GetID()) {
// If a flush job needs to install the flush result for mems and
// mems[0] is not the earliest memtable, it means another thread must
// be installing flush results for the same column family, then the
// current thread needs to wait.
ready = false;
break;
} else if (mems.empty() && cfds[i]->imm()->GetEarliestMemTableID() <=
bg_flush_args[i].max_memtable_id_) {
// If a flush job does not need to install flush results, then it has
// to wait until all memtables up to max_memtable_id_ (inclusive) are
// installed.
ready = false;
break;
}
}
return std::make_pair(Status::OK(), !ready);
};
bool resuming_from_bg_err =
error_handler_.IsDBStopped() ||
(cfds[0]->GetFlushReason() == FlushReason::kErrorRecovery ||
cfds[0]->GetFlushReason() == FlushReason::kErrorRecoveryRetryFlush);
while ((!resuming_from_bg_err || error_handler_.GetRecoveryError().ok())) {
std::pair<Status, bool> res = wait_to_install_func();
TEST_SYNC_POINT_CALLBACK(
"DBImpl::AtomicFlushMemTablesToOutputFiles:WaitToCommit", &res);
if (!res.first.ok()) {
s = res.first;
break;
} else if (!res.second) {
break;
}
atomic_flush_install_cv_.Wait();
resuming_from_bg_err =
error_handler_.IsDBStopped() ||
(cfds[0]->GetFlushReason() == FlushReason::kErrorRecovery ||
cfds[0]->GetFlushReason() == FlushReason::kErrorRecoveryRetryFlush);
}
if (!resuming_from_bg_err) {
// If not resuming from bg err, then we determine future action based on
// whether we hit background error.
if (s.ok()) {
s = error_handler_.GetBGError();
}
} else if (s.ok()) {
// If resuming from bg err, we still rely on wait_to_install_func()'s
// result to determine future action. If wait_to_install_func() returns
// non-ok already, then we should not proceed to flush result
// installation.
s = error_handler_.GetRecoveryError();
}
}
if (s.ok()) {
autovector<ColumnFamilyData*> tmp_cfds;
autovector<const autovector<MemTable*>*> mems_list;
autovector<const MutableCFOptions*> mutable_cf_options_list;
autovector<FileMetaData*> tmp_file_meta;
autovector<std::list<std::unique_ptr<FlushJobInfo>>*>
committed_flush_jobs_info;
for (int i = 0; i != num_cfs; ++i) {
const auto& mems = jobs[i]->GetMemTables();
if (!cfds[i]->IsDropped() && !mems.empty()) {
tmp_cfds.emplace_back(cfds[i]);
mems_list.emplace_back(&mems);
mutable_cf_options_list.emplace_back(&all_mutable_cf_options[i]);
tmp_file_meta.emplace_back(&file_meta[i]);
#ifndef ROCKSDB_LITE
committed_flush_jobs_info.emplace_back(
jobs[i]->GetCommittedFlushJobsInfo());
#endif //! ROCKSDB_LITE
}
}
s = InstallMemtableAtomicFlushResults(
nullptr /* imm_lists */, tmp_cfds, mutable_cf_options_list, mems_list,
versions_.get(), &logs_with_prep_tracker_, &mutex_, tmp_file_meta,
committed_flush_jobs_info, &job_context->memtables_to_free,
directories_.GetDbDir(), log_buffer);
}
if (s.ok()) {
assert(num_cfs ==
static_cast<int>(job_context->superversion_contexts.size()));
for (int i = 0; i != num_cfs; ++i) {
assert(cfds[i]);
if (cfds[i]->IsDropped()) {
continue;
}
InstallSuperVersionAndScheduleWork(cfds[i],
&job_context->superversion_contexts[i],
all_mutable_cf_options[i]);
const std::string& column_family_name = cfds[i]->GetName();
Version* const current = cfds[i]->current();
assert(current);
const VersionStorageInfo* const storage_info = current->storage_info();
assert(storage_info);
VersionStorageInfo::LevelSummaryStorage tmp;
ROCKS_LOG_BUFFER(log_buffer, "[%s] Level summary: %s\n",
column_family_name.c_str(),
storage_info->LevelSummary(&tmp));
const auto& blob_files = storage_info->GetBlobFiles();
if (!blob_files.empty()) {
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
3 years ago
assert(blob_files.front());
assert(blob_files.back());
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Blob file summary: head=%" PRIu64 ", tail=%" PRIu64 "\n",
column_family_name.c_str(), blob_files.front()->GetBlobFileNumber(),
blob_files.back()->GetBlobFileNumber());
}
}
if (made_progress) {
*made_progress = true;
}
#ifndef ROCKSDB_LITE
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
assert(all_mutable_cf_options.size() == static_cast<size_t>(num_cfs));
for (int i = 0; s.ok() && i != num_cfs; ++i) {
// If mempurge happened instead of Flush,
// no NotifyOnFlushCompleted call (no SST file created).
if (switched_to_mempurge[i]) {
continue;
}
if (cfds[i]->IsDropped()) {
continue;
}
NotifyOnFlushCompleted(cfds[i], all_mutable_cf_options[i],
jobs[i]->GetCommittedFlushJobsInfo());
if (sfm) {
std::string file_path = MakeTableFileName(
cfds[i]->ioptions()->cf_paths[0].path, file_meta[i].fd.GetNumber());
// TODO (PR7798). We should only add the file to the FileManager if it
// exists. Otherwise, some tests may fail. Ignore the error in the
// interim.
sfm->OnAddFile(file_path).PermitUncheckedError();
if (sfm->IsMaxAllowedSpaceReached() &&
error_handler_.GetBGError().ok()) {
Status new_bg_error =
Status::SpaceLimit("Max allowed space was reached");
error_handler_.SetBGError(new_bg_error,
BackgroundErrorReason::kFlush);
}
}
}
#endif // ROCKSDB_LITE
}
// Need to undo atomic flush if something went wrong, i.e. s is not OK and
// it is not because of CF drop.
if (!s.ok() && !s.IsColumnFamilyDropped()) {
if (log_io_s.ok()) {
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
4 years ago
// Error while writing to MANIFEST.
// In fact, versions_->io_status() can also be the result of renaming
// CURRENT file. With current code, it's just difficult to tell. So just
// be pessimistic and try write to a new MANIFEST.
// TODO: distinguish between MANIFEST write and CURRENT renaming
if (!versions_->io_status().ok()) {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the Manifest write will be map to soft error.
// TODO: kManifestWriteNoWAL and kFlushNoWAL are misleading. Refactor
// is needed.
error_handler_.SetBGError(s,
BackgroundErrorReason::kManifestWriteNoWAL);
} else {
// If WAL sync is successful (either WAL size is 0 or there is no IO
// error), all the other SST file write errors will be set as
// kFlushNoWAL.
error_handler_.SetBGError(s, BackgroundErrorReason::kFlushNoWAL);
}
} else {
assert(s == log_io_s);
Status new_bg_error = s;
error_handler_.SetBGError(new_bg_error, BackgroundErrorReason::kFlush);
}
}
return s;
}
void DBImpl::NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
const MutableCFOptions& mutable_cf_options,
int job_id) {
#ifndef ROCKSDB_LITE
if (immutable_db_options_.listeners.size() == 0U) {
return;
}
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return;
}
bool triggered_writes_slowdown =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_slowdown_writes_trigger);
bool triggered_writes_stop =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_stop_writes_trigger);
// release lock while notifying events
mutex_.Unlock();
{
FlushJobInfo info{};
info.cf_id = cfd->GetID();
info.cf_name = cfd->GetName();
// TODO(yhchiang): make db_paths dynamic in case flush does not
// go to L0 in the future.
const uint64_t file_number = file_meta->fd.GetNumber();
info.file_path =
MakeTableFileName(cfd->ioptions()->cf_paths[0].path, file_number);
info.file_number = file_number;
info.thread_id = env_->GetThreadID();
info.job_id = job_id;
info.triggered_writes_slowdown = triggered_writes_slowdown;
info.triggered_writes_stop = triggered_writes_stop;
info.smallest_seqno = file_meta->fd.smallest_seqno;
info.largest_seqno = file_meta->fd.largest_seqno;
info.flush_reason = cfd->GetFlushReason();
for (auto listener : immutable_db_options_.listeners) {
listener->OnFlushBegin(this, info);
}
}
mutex_.Lock();
// no need to signal bg_cv_ as it will be signaled at the end of the
// flush process.
#else
(void)cfd;
(void)file_meta;
(void)mutable_cf_options;
(void)job_id;
#endif // ROCKSDB_LITE
}
void DBImpl::NotifyOnFlushCompleted(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
std::list<std::unique_ptr<FlushJobInfo>>* flush_jobs_info) {
#ifndef ROCKSDB_LITE
assert(flush_jobs_info != nullptr);
if (immutable_db_options_.listeners.size() == 0U) {
return;
}
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return;
}
bool triggered_writes_slowdown =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_slowdown_writes_trigger);
bool triggered_writes_stop =
(cfd->current()->storage_info()->NumLevelFiles(0) >=
mutable_cf_options.level0_stop_writes_trigger);
// release lock while notifying events
mutex_.Unlock();
{
for (auto& info : *flush_jobs_info) {
info->triggered_writes_slowdown = triggered_writes_slowdown;
info->triggered_writes_stop = triggered_writes_stop;
for (auto listener : immutable_db_options_.listeners) {
listener->OnFlushCompleted(this, *info);
}
Fix TSAN data race in EventListenerTest.MultiCF (#9528) Summary: **Context:** `EventListenerTest.MultiCF` occasionally failed on TSAN data race as below: ``` WARNING: ThreadSanitizer: data race (pid=2047633) Read of size 8 at 0x7b6000001440 by main thread: #0 std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> >::size() const /usr/bin/../lib/gcc/x86_64-linux-gnu/9/../../../../include/c++/9/bits/stl_vector.h:916:40 (listener_test+0x52337c) https://github.com/facebook/rocksdb/issues/1 rocksdb::EventListenerTest_MultiCF_Test::TestBody() /home/circleci/project/db/listener_test.cc:384:7 (listener_test+0x52337c) Previous write of size 8 at 0x7b6000001440 by thread T2: #0 void std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> >::_M_realloc_insert<rocksdb::DB* const&>(__gnu_cxx::__normal_iterator<rocksdb::DB**, std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> > >, rocksdb::DB* const&) /usr/bin/../lib/gcc/x86_64-linux-gnu/9/../../../../include/c++/9/bits/vector.tcc:503:31 (listener_test+0x550654) https://github.com/facebook/rocksdb/issues/1 std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> >::push_back(rocksdb::DB* const&) /usr/bin/../lib/gcc/x86_64-linux-gnu/9/../../../../include/c++/9/bits/stl_vector.h:1195:4 (listener_test+0x550654) https://github.com/facebook/rocksdb/issues/2 rocksdb::TestFlushListener::OnFlushCompleted(rocksdb::DB*, rocksdb::FlushJobInfo const&) /home/circleci/project/db/listener_test.cc:255:18 (listener_test+0x550654) ``` After investigation, it is due to the following: (1) `ASSERT_OK(Flush(i));` before the read `std::vector::size()` is supposed to be [blocked on `DB::Impl::bg_cv_` for memtable flush to finish](https://github.com/facebook/rocksdb/blob/320d9a8e8a1b6998f92934f87fc71ad8bd6d4596/db/db_impl/db_impl_compaction_flush.cc#L2319) and get signaled [at the end of background flush ](https://github.com/facebook/rocksdb/blob/320d9a8e8a1b6998f92934f87fc71ad8bd6d4596/db/db_impl/db_impl_compaction_flush.cc#L2830), which happens after the write `std::vector::push_back()` . So the sequence of execution should have been synchronized as `call flush() -> write -> return from flush() -> read` and would not cause any TSAN data race. - The subsequent `ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());` serves a similar purpose based on [the previous attempt to deflake the test.](https://github.com/facebook/rocksdb/pull/9084) (2) However, there are multiple places in the code can signal this `DB::Impl::bg_cv_` and mistakenly wake up `ASSERT_OK(Flush(i));` (or `ASSERT_OK(dbfull()->TEST_WaitForFlushMemTable());`) too early (and with the lock available to them), resulting in non-synchronized read and write thus a TSAN data race. - Reproduced by the following, suggested by ajkr: ``` diff --git a/db/db_impl/db_impl_compaction_flush.cc b/db/db_impl/db_impl_compaction_flush.cc index 4ff87c1e4..52492e9cf 100644 --- a/db/db_impl/db_impl_compaction_flush.cc +++ b/db/db_impl/db_impl_compaction_flush.cc @@ -22,7 +22,7 @@ #include "test_util/sync_point.h" #include "util/cast_util.h" #include "util/concurrent_task_limiter_impl.h" namespace ROCKSDB_NAMESPACE { bool DBImpl::EnoughRoomForCompaction( @@ -855,6 +855,7 @@ void DBImpl::NotifyOnFlushCompleted( mutable_cf_options.level0_stop_writes_trigger); // release lock while notifying events mutex_.Unlock(); + bg_cv_.SignalAll(); ``` **Summary:** - Added synchornization between read and write by ` ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->LoadDependency()` mechanism Pull Request resolved: https://github.com/facebook/rocksdb/pull/9528 Test Plan: `./listener_test --gtest_filter=EventListenerTest.MultiCF --gtest_repeat=10` - pre-fix: ``` Repeating all tests (iteration 3) Note: Google Test filter = EventListenerTest.MultiCF [==========] Running 1 test from 1 test case. [----------] Global test environment set-up. [----------] 1 test from EventListenerTest [ RUN ] EventListenerTest.MultiCF ================== WARNING: ThreadSanitizer: data race (pid=3377137) Read of size 8 at 0x7b6000000840 by main thread: #0 std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> >::size() https://github.com/facebook/rocksdb/issues/1 rocksdb::EventListenerTest_MultiCF_Test::TestBody() db/listener_test.cc:384 (listener_test+0x4bb300) Previous write of size 8 at 0x7b6000000840 by thread T2: #0 void std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> >::_M_realloc_insert<rocksdb::DB* const&>(__gnu_cxx::__normal_iterator<rocksdb::DB**, std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> > >, rocksdb::DB* const&) https://github.com/facebook/rocksdb/issues/1 std::vector<rocksdb::DB*, std::allocator<rocksdb::DB*> >::push_back(rocksdb::DB* const&) https://github.com/facebook/rocksdb/issues/2 rocksdb::TestFlushListener::OnFlushCompleted(rocksdb::DB*, rocksdb::FlushJobInfo const&) db/listener_test.cc:255 (listener_test+0x4e820f) ``` - post-fix: `All passed` Reviewed By: ajkr Differential Revision: D34085791 Pulled By: hx235 fbshipit-source-id: f877aa687ea1d5cb6f31ef8c4772625d22868e8b
3 years ago
TEST_SYNC_POINT(
"DBImpl::NotifyOnFlushCompleted::PostAllOnFlushCompleted");
}
flush_jobs_info->clear();
}
mutex_.Lock();
// no need to signal bg_cv_ as it will be signaled at the end of the
// flush process.
#else
(void)cfd;
(void)mutable_cf_options;
(void)flush_jobs_info;
#endif // ROCKSDB_LITE
}
Status DBImpl::CompactRange(const CompactRangeOptions& options,
ColumnFamilyHandle* column_family,
const Slice* begin_without_ts,
const Slice* end_without_ts) {
if (manual_compaction_paused_.load(std::memory_order_acquire) > 0) {
return Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
if (options.canceled && options.canceled->load(std::memory_order_acquire)) {
return Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
const Comparator* const ucmp = column_family->GetComparator();
assert(ucmp);
size_t ts_sz = ucmp->timestamp_size();
if (ts_sz == 0) {
return CompactRangeInternal(options, column_family, begin_without_ts,
end_without_ts, "" /*trim_ts*/);
}
std::string begin_str;
std::string end_str;
// CompactRange compact all keys: [begin, end] inclusively. Add maximum
// timestamp to include all `begin` keys, and add minimal timestamp to include
// all `end` keys.
if (begin_without_ts != nullptr) {
AppendKeyWithMaxTimestamp(&begin_str, *begin_without_ts, ts_sz);
}
if (end_without_ts != nullptr) {
AppendKeyWithMinTimestamp(&end_str, *end_without_ts, ts_sz);
}
Slice begin(begin_str);
Slice end(end_str);
Slice* begin_with_ts = begin_without_ts ? &begin : nullptr;
Slice* end_with_ts = end_without_ts ? &end : nullptr;
return CompactRangeInternal(options, column_family, begin_with_ts,
end_with_ts, "" /*trim_ts*/);
}
Status DBImpl::IncreaseFullHistoryTsLow(ColumnFamilyHandle* column_family,
std::string ts_low) {
ColumnFamilyData* cfd = nullptr;
if (column_family == nullptr) {
cfd = default_cf_handle_->cfd();
} else {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
assert(cfh != nullptr);
cfd = cfh->cfd();
}
assert(cfd != nullptr && cfd->user_comparator() != nullptr);
if (cfd->user_comparator()->timestamp_size() == 0) {
return Status::InvalidArgument(
"Timestamp is not enabled in this column family");
}
if (cfd->user_comparator()->timestamp_size() != ts_low.size()) {
return Status::InvalidArgument("ts_low size mismatch");
}
return IncreaseFullHistoryTsLowImpl(cfd, ts_low);
}
Status DBImpl::IncreaseFullHistoryTsLowImpl(ColumnFamilyData* cfd,
std::string ts_low) {
VersionEdit edit;
edit.SetColumnFamily(cfd->GetID());
edit.SetFullHistoryTsLow(ts_low);
TEST_SYNC_POINT_CALLBACK("DBImpl::IncreaseFullHistoryTsLowImpl:BeforeEdit",
&edit);
InstrumentedMutexLock l(&mutex_);
std::string current_ts_low = cfd->GetFullHistoryTsLow();
const Comparator* ucmp = cfd->user_comparator();
assert(ucmp->timestamp_size() == ts_low.size() && !ts_low.empty());
if (!current_ts_low.empty() &&
ucmp->CompareTimestamp(ts_low, current_ts_low) < 0) {
return Status::InvalidArgument("Cannot decrease full_history_ts_low");
}
Status s = versions_->LogAndApply(cfd, *cfd->GetLatestMutableCFOptions(),
&edit, &mutex_);
if (!s.ok()) {
return s;
}
current_ts_low = cfd->GetFullHistoryTsLow();
if (!current_ts_low.empty() &&
ucmp->CompareTimestamp(current_ts_low, ts_low) > 0) {
std::stringstream oss;
oss << "full_history_ts_low: " << Slice(current_ts_low).ToString(true)
<< " is set to be higher than the requested "
"timestamp: "
<< Slice(ts_low).ToString(true) << std::endl;
return Status::TryAgain(oss.str());
}
return Status::OK();
}
Status DBImpl::CompactRangeInternal(const CompactRangeOptions& options,
ColumnFamilyHandle* column_family,
const Slice* begin, const Slice* end,
const std::string& trim_ts) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
auto cfd = cfh->cfd();
if (options.target_path_id >= cfd->ioptions()->cf_paths.size()) {
return Status::InvalidArgument("Invalid target path ID");
}
bool flush_needed = true;
// Update full_history_ts_low if it's set
if (options.full_history_ts_low != nullptr &&
!options.full_history_ts_low->empty()) {
std::string ts_low = options.full_history_ts_low->ToString();
if (begin != nullptr || end != nullptr) {
return Status::InvalidArgument(
"Cannot specify compaction range with full_history_ts_low");
}
Status s = IncreaseFullHistoryTsLowImpl(cfd, ts_low);
if (!s.ok()) {
LogFlush(immutable_db_options_.info_log);
return s;
}
}
Status s;
if (begin != nullptr && end != nullptr) {
// TODO(ajkr): We could also optimize away the flush in certain cases where
// one/both sides of the interval are unbounded. But it requires more
// changes to RangesOverlapWithMemtables.
Range range(*begin, *end);
SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
s = cfd->RangesOverlapWithMemtables(
{range}, super_version, immutable_db_options_.allow_data_in_errors,
&flush_needed);
CleanupSuperVersion(super_version);
}
if (s.ok() && flush_needed) {
FlushOptions fo;
fo.allow_write_stall = options.allow_write_stall;
if (immutable_db_options_.atomic_flush) {
autovector<ColumnFamilyData*> cfds;
mutex_.Lock();
SelectColumnFamiliesForAtomicFlush(&cfds);
mutex_.Unlock();
s = AtomicFlushMemTables(cfds, fo, FlushReason::kManualCompaction,
false /* writes_stopped */);
} else {
s = FlushMemTable(cfd, fo, FlushReason::kManualCompaction,
false /* writes_stopped*/);
}
if (!s.ok()) {
LogFlush(immutable_db_options_.info_log);
return s;
}
}
constexpr int kInvalidLevel = -1;
int final_output_level = kInvalidLevel;
bool exclusive = options.exclusive_manual_compaction;
if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal &&
cfd->NumberLevels() > 1) {
// Always compact all files together.
final_output_level = cfd->NumberLevels() - 1;
// if bottom most level is reserved
if (immutable_db_options_.allow_ingest_behind) {
final_output_level--;
}
s = RunManualCompaction(cfd, ColumnFamilyData::kCompactAllLevels,
final_output_level, options, begin, end, exclusive,
false, std::numeric_limits<uint64_t>::max(),
trim_ts);
} else {
int first_overlapped_level = kInvalidLevel;
int max_overlapped_level = kInvalidLevel;
{
SuperVersion* super_version = cfd->GetReferencedSuperVersion(this);
Version* current_version = super_version->current;
ReadOptions ro;
ro.total_order_seek = true;
bool overlap;
for (int level = 0;
level < current_version->storage_info()->num_non_empty_levels();
level++) {
overlap = true;
if (begin != nullptr && end != nullptr) {
Status status = current_version->OverlapWithLevelIterator(
ro, file_options_, *begin, *end, level, &overlap);
if (!status.ok()) {
overlap = current_version->storage_info()->OverlapInLevel(
level, begin, end);
}
} else {
overlap = current_version->storage_info()->OverlapInLevel(level,
begin, end);
}
if (overlap) {
if (first_overlapped_level == kInvalidLevel) {
first_overlapped_level = level;
}
max_overlapped_level = level;
}
}
CleanupSuperVersion(super_version);
}
if (s.ok() && first_overlapped_level != kInvalidLevel) {
// max_file_num_to_ignore can be used to filter out newly created SST
// files, useful for bottom level compaction in a manual compaction
uint64_t max_file_num_to_ignore = std::numeric_limits<uint64_t>::max();
uint64_t next_file_number = versions_->current_next_file_number();
final_output_level = max_overlapped_level;
int output_level;
for (int level = first_overlapped_level; level <= max_overlapped_level;
level++) {
bool disallow_trivial_move = false;
// in case the compaction is universal or if we're compacting the
// bottom-most level, the output level will be the same as input one.
// level 0 can never be the bottommost level (i.e. if all files are in
// level 0, we will compact to level 1)
if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
output_level = level;
} else if (level == max_overlapped_level && level > 0) {
if (options.bottommost_level_compaction ==
BottommostLevelCompaction::kSkip) {
// Skip bottommost level compaction
continue;
} else if (options.bottommost_level_compaction ==
BottommostLevelCompaction::kIfHaveCompactionFilter &&
cfd->ioptions()->compaction_filter == nullptr &&
cfd->ioptions()->compaction_filter_factory == nullptr) {
// Skip bottommost level compaction since we don't have a compaction
// filter
continue;
}
output_level = level;
// update max_file_num_to_ignore only for bottom level compaction
// because data in newly compacted files in middle levels may still
// need to be pushed down
max_file_num_to_ignore = next_file_number;
} else {
output_level = level + 1;
if (cfd->ioptions()->compaction_style == kCompactionStyleLevel &&
cfd->ioptions()->level_compaction_dynamic_level_bytes &&
level == 0) {
output_level = ColumnFamilyData::kCompactToBaseLevel;
}
// if it's a BottommostLevel compaction and `kForce*` compaction is
// set, disallow trivial move
if (level == max_overlapped_level &&
(options.bottommost_level_compaction ==
BottommostLevelCompaction::kForce ||
options.bottommost_level_compaction ==
BottommostLevelCompaction::kForceOptimized)) {
disallow_trivial_move = true;
}
}
// trim_ts need real compaction to remove latest record
if (!trim_ts.empty()) {
disallow_trivial_move = true;
}
s = RunManualCompaction(cfd, level, output_level, options, begin, end,
exclusive, disallow_trivial_move,
max_file_num_to_ignore, trim_ts);
if (!s.ok()) {
break;
}
if (output_level == ColumnFamilyData::kCompactToBaseLevel) {
final_output_level = cfd->NumberLevels() - 1;
} else if (output_level > final_output_level) {
final_output_level = output_level;
}
TEST_SYNC_POINT("DBImpl::RunManualCompaction()::1");
TEST_SYNC_POINT("DBImpl::RunManualCompaction()::2");
}
}
}
if (!s.ok() || final_output_level == kInvalidLevel) {
LogFlush(immutable_db_options_.info_log);
return s;
}
if (options.change_level) {
TEST_SYNC_POINT("DBImpl::CompactRange:BeforeRefit:1");
TEST_SYNC_POINT("DBImpl::CompactRange:BeforeRefit:2");
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[RefitLevel] waiting for background threads to stop");
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
DisableManualCompaction();
s = PauseBackgroundWork();
if (s.ok()) {
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
TEST_SYNC_POINT("DBImpl::CompactRange:PreRefitLevel");
s = ReFitLevel(cfd, final_output_level, options.target_level);
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
TEST_SYNC_POINT("DBImpl::CompactRange:PostRefitLevel");
// ContinueBackgroundWork always return Status::OK().
Status temp_s = ContinueBackgroundWork();
assert(temp_s.ok());
}
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
EnableManualCompaction();
Prevent corruption with parallel manual compactions and `change_level == true` (#9077) Summary: The bug can impact the following scenario. There must be two `CompactRange()`s, call them A and B. Compaction A must have `change_level=true`. Compactions A and B must run in parallel, and new data must be added while they run as well. Now, on to the details of the race condition. Compaction A must reach the refitting phase while B's next step is to trivial move new data (i.e., data that has been inserted behind A) down to the same level that A's refit targets (`CompactRangeOptions::target_level`). B must be unregistered (i.e., has not yet called `AddManualCompaction()` for the current `RunManualCompaction()`) while A invokes `DisableManualCompaction()`s to prepare for refitting. In the old code, B could still proceed to register a manual compaction, while A had disabled manual compaction. The next part of the race condition is B picks and schedules a trivial move while A has released the lock in refitting phase in order to persist the LSM state change (i.e., the log phase of `LogAndApply()`). That way, B does not see the refitted data when picking a trivial-move compaction. So it is susceptible to picking one that overlaps. Finally, B executes the picked trivial-move compaction. Trivial-move compactions are special in that they never check whether manual compaction is disabled. So the picked compaction causing overlap ends up being applied, leading to LSM corruption if `force_consistency_checks=false`, or entering read-only mode with `Status::Corruption` if `force_consistency_checks=true` (the default). The fix is just to prevent B from registering itself in `RunManualCompaction()` while manual compactions are disabled, consequently preventing any trivial move or other compaction from being picked/scheduled. Thanks to siying for finding the bug. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9077 Test Plan: The test does not go all the way in exposing the bug because it requires a compaction to be picked/scheduled while logging LSM state change for RefitLevel(). But the fix is to make such a compaction not picked/scheduled in the first place, so any repro of that scenario would end up hanging RefitLevel() logging. So instead I just verified no such compaction is registered in the scenario where `RefitLevel()` disables manual compactions. Reviewed By: siying Differential Revision: D31921908 Pulled By: ajkr fbshipit-source-id: 9bb5d0e847ad428211227f40830c685c209fbecb
3 years ago
TEST_SYNC_POINT(
"DBImpl::CompactRange:PostRefitLevel:ManualCompactionEnabled");
}
LogFlush(immutable_db_options_.info_log);
{
InstrumentedMutexLock l(&mutex_);
// an automatic compaction that has been scheduled might have been
// preempted by the manual compactions. Need to schedule it back.
MaybeScheduleFlushOrCompaction();
}
return s;
}
Status DBImpl::CompactFiles(const CompactionOptions& compact_options,
ColumnFamilyHandle* column_family,
const std::vector<std::string>& input_file_names,
const int output_level, const int output_path_id,
std::vector<std::string>* const output_file_names,
CompactionJobInfo* compaction_job_info) {
#ifdef ROCKSDB_LITE
(void)compact_options;
(void)column_family;
(void)input_file_names;
(void)output_level;
(void)output_path_id;
(void)output_file_names;
(void)compaction_job_info;
// not supported in lite version
return Status::NotSupported("Not supported in ROCKSDB LITE");
#else
if (column_family == nullptr) {
return Status::InvalidArgument("ColumnFamilyHandle must be non-null.");
}
auto cfd =
static_cast_with_check<ColumnFamilyHandleImpl>(column_family)->cfd();
assert(cfd);
Status s;
JobContext job_context(next_job_id_.fetch_add(1), true);
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
immutable_db_options_.info_log.get());
// Perform CompactFiles
TEST_SYNC_POINT("TestCompactFiles::IngestExternalFile2");
TEST_SYNC_POINT_CALLBACK(
"TestCompactFiles:PausingManualCompaction:3",
reinterpret_cast<void*>(
const_cast<std::atomic<int>*>(&manual_compaction_paused_)));
{
InstrumentedMutexLock l(&mutex_);
// This call will unlock/lock the mutex to wait for current running
// IngestExternalFile() calls to finish.
WaitForIngestFile();
// We need to get current after `WaitForIngestFile`, because
// `IngestExternalFile` may add files that overlap with `input_file_names`
auto* current = cfd->current();
current->Ref();
s = CompactFilesImpl(compact_options, cfd, current, input_file_names,
output_file_names, output_level, output_path_id,
&job_context, &log_buffer, compaction_job_info);
current->Unref();
}
// Find and delete obsolete files
{
InstrumentedMutexLock l(&mutex_);
// If !s.ok(), this means that Compaction failed. In that case, we want
// to delete all obsolete files we might have created and we force
// FindObsoleteFiles(). This is because job_context does not
// catch all created files if compaction failed.
FindObsoleteFiles(&job_context, !s.ok());
} // release the mutex
// delete unnecessary files if any, this is done outside the mutex
if (job_context.HaveSomethingToClean() ||
job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
// Have to flush the info logs before bg_compaction_scheduled_--
// because if bg_flush_scheduled_ becomes 0 and the lock is
// released, the deconstructor of DB can kick in and destroy all the
// states of DB so info_log might not be available after that point.
// It also applies to access other states that DB owns.
log_buffer.FlushBufferToLog();
if (job_context.HaveSomethingToDelete()) {
// no mutex is locked here. No need to Unlock() and Lock() here.
PurgeObsoleteFiles(job_context);
}
job_context.Clean();
}
return s;
#endif // ROCKSDB_LITE
}
#ifndef ROCKSDB_LITE
Status DBImpl::CompactFilesImpl(
const CompactionOptions& compact_options, ColumnFamilyData* cfd,
Version* version, const std::vector<std::string>& input_file_names,
std::vector<std::string>* const output_file_names, const int output_level,
int output_path_id, JobContext* job_context, LogBuffer* log_buffer,
CompactionJobInfo* compaction_job_info) {
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return Status::ShutdownInProgress();
}
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
if (manual_compaction_paused_.load(std::memory_order_acquire) > 0) {
return Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
std::unordered_set<uint64_t> input_set;
for (const auto& file_name : input_file_names) {
input_set.insert(TableFileNameToNumber(file_name));
}
ColumnFamilyMetaData cf_meta;
// TODO(yhchiang): can directly use version here if none of the
// following functions call is pluggable to external developers.
version->GetColumnFamilyMetaData(&cf_meta);
if (output_path_id < 0) {
if (cfd->ioptions()->cf_paths.size() == 1U) {
output_path_id = 0;
} else {
return Status::NotSupported(
"Automatic output path selection is not "
"yet supported in CompactFiles()");
}
}
Status s = cfd->compaction_picker()->SanitizeCompactionInputFiles(
&input_set, cf_meta, output_level);
if (!s.ok()) {
return s;
}
std::vector<CompactionInputFiles> input_files;
s = cfd->compaction_picker()->GetCompactionInputsFromFileNumbers(
&input_files, &input_set, version->storage_info(), compact_options);
if (!s.ok()) {
return s;
}
for (const auto& inputs : input_files) {
if (cfd->compaction_picker()->AreFilesInCompaction(inputs.files)) {
return Status::Aborted(
"Some of the necessary compaction input "
"files are already being compacted");
}
}
bool sfm_reserved_compact_space = false;
// First check if we have enough room to do the compaction
bool enough_room = EnoughRoomForCompaction(
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
cfd, input_files, &sfm_reserved_compact_space, log_buffer);
if (!enough_room) {
// m's vars will get set properly at the end of this function,
// as long as status == CompactionTooLarge
return Status::CompactionTooLarge();
}
// At this point, CompactFiles will be run.
bg_compaction_scheduled_++;
std::unique_ptr<Compaction> c;
assert(cfd->compaction_picker());
c.reset(cfd->compaction_picker()->CompactFiles(
compact_options, input_files, output_level, version->storage_info(),
*cfd->GetLatestMutableCFOptions(), mutable_db_options_, output_path_id));
// we already sanitized the set of input files and checked for conflicts
// without releasing the lock, so we're guaranteed a compaction can be formed.
assert(c != nullptr);
c->SetInputVersion(version);
// deletion compaction currently not allowed in CompactFiles.
assert(!c->deletion_compaction());
std::vector<SequenceNumber> snapshot_seqs;
SequenceNumber earliest_write_conflict_snapshot;
SnapshotChecker* snapshot_checker;
GetSnapshotContext(job_context, &snapshot_seqs,
&earliest_write_conflict_snapshot, &snapshot_checker);
std::unique_ptr<std::list<uint64_t>::iterator> pending_outputs_inserted_elem(
new std::list<uint64_t>::iterator(
CaptureCurrentFileNumberInPendingOutputs()));
assert(is_snapshot_supported_ || snapshots_.empty());
CompactionJobStats compaction_job_stats;
CompactionJob compaction_job(
job_context->job_id, c.get(), immutable_db_options_, mutable_db_options_,
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
5 years ago
file_options_for_compaction_, versions_.get(), &shutting_down_,
log_buffer, directories_.GetDbDir(),
GetDataDir(c->column_family_data(), c->output_path_id()),
GetDataDir(c->column_family_data(), 0), stats_, &mutex_, &error_handler_,
snapshot_seqs, earliest_write_conflict_snapshot, snapshot_checker,
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
job_context, table_cache_, &event_logger_,
c->mutable_cf_options()->paranoid_file_checks,
c->mutable_cf_options()->report_bg_io_stats, dbname_,
&compaction_job_stats, Env::Priority::USER, io_tracer_,
kManualCompactionCanceledFalse_, db_id_, db_session_id_,
c->column_family_data()->GetFullHistoryTsLow(), c->trim_ts(),
&blob_callback_);
// Creating a compaction influences the compaction score because the score
// takes running compactions into account (by skipping files that are already
// being compacted). Since we just changed compaction score, we recalculate it
// here.
version->storage_info()->ComputeCompactionScore(*cfd->ioptions(),
*c->mutable_cf_options());
compaction_job.Prepare();
mutex_.Unlock();
TEST_SYNC_POINT("CompactFilesImpl:0");
TEST_SYNC_POINT("CompactFilesImpl:1");
// Ignore the status here, as it will be checked in the Install down below...
compaction_job.Run().PermitUncheckedError();
TEST_SYNC_POINT("CompactFilesImpl:2");
TEST_SYNC_POINT("CompactFilesImpl:3");
mutex_.Lock();
Status status = compaction_job.Install(*c->mutable_cf_options());
if (status.ok()) {
assert(compaction_job.io_status().ok());
InstallSuperVersionAndScheduleWork(c->column_family_data(),
&job_context->superversion_contexts[0],
*c->mutable_cf_options());
}
// status above captures any error during compaction_job.Install, so its ok
// not check compaction_job.io_status() explicitly if we're not calling
// SetBGError
compaction_job.io_status().PermitUncheckedError();
c->ReleaseCompactionFiles(s);
#ifndef ROCKSDB_LITE
// Need to make sure SstFileManager does its bookkeeping
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
if (sfm && sfm_reserved_compact_space) {
sfm->OnCompactionCompletion(c.get());
}
#endif // ROCKSDB_LITE
ReleaseFileNumberFromPendingOutputs(pending_outputs_inserted_elem);
if (compaction_job_info != nullptr) {
BuildCompactionJobInfo(cfd, c.get(), s, compaction_job_stats,
job_context->job_id, version, compaction_job_info);
}
if (status.ok()) {
// Done
} else if (status.IsColumnFamilyDropped() || status.IsShutdownInProgress()) {
// Ignore compaction errors found during shutting down
} else if (status.IsManualCompactionPaused()) {
// Don't report stopping manual compaction as error
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[%s] [JOB %d] Stopping manual compaction",
c->column_family_data()->GetName().c_str(),
job_context->job_id);
} else {
ROCKS_LOG_WARN(immutable_db_options_.info_log,
"[%s] [JOB %d] Compaction error: %s",
c->column_family_data()->GetName().c_str(),
job_context->job_id, status.ToString().c_str());
IOStatus io_s = compaction_job.io_status();
if (!io_s.ok()) {
error_handler_.SetBGError(io_s, BackgroundErrorReason::kCompaction);
} else {
error_handler_.SetBGError(status, BackgroundErrorReason::kCompaction);
}
}
if (output_file_names != nullptr) {
for (const auto& newf : c->edit()->GetNewFiles()) {
output_file_names->push_back(TableFileName(
c->immutable_options()->cf_paths, newf.second.fd.GetNumber(),
newf.second.fd.GetPathId()));
}
for (const auto& blob_file : c->edit()->GetBlobFileAdditions()) {
output_file_names->push_back(
BlobFileName(c->immutable_options()->cf_paths.front().path,
blob_file.GetBlobFileNumber()));
}
}
c.reset();
bg_compaction_scheduled_--;
if (bg_compaction_scheduled_ == 0) {
bg_cv_.SignalAll();
}
MaybeScheduleFlushOrCompaction();
TEST_SYNC_POINT("CompactFilesImpl:End");
return status;
}
#endif // ROCKSDB_LITE
Status DBImpl::PauseBackgroundWork() {
InstrumentedMutexLock guard_lock(&mutex_);
bg_compaction_paused_++;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
while (bg_bottom_compaction_scheduled_ > 0 || bg_compaction_scheduled_ > 0 ||
bg_flush_scheduled_ > 0) {
bg_cv_.Wait();
}
bg_work_paused_++;
return Status::OK();
}
Status DBImpl::ContinueBackgroundWork() {
InstrumentedMutexLock guard_lock(&mutex_);
if (bg_work_paused_ == 0) {
return Status::InvalidArgument();
}
assert(bg_work_paused_ > 0);
assert(bg_compaction_paused_ > 0);
bg_compaction_paused_--;
bg_work_paused_--;
// It's sufficient to check just bg_work_paused_ here since
// bg_work_paused_ is always no greater than bg_compaction_paused_
if (bg_work_paused_ == 0) {
MaybeScheduleFlushOrCompaction();
}
return Status::OK();
}
void DBImpl::NotifyOnCompactionBegin(ColumnFamilyData* cfd, Compaction* c,
const Status& st,
const CompactionJobStats& job_stats,
int job_id) {
#ifndef ROCKSDB_LITE
if (immutable_db_options_.listeners.empty()) {
return;
}
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return;
}
if (c->is_manual_compaction() &&
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
manual_compaction_paused_.load(std::memory_order_acquire) > 0) {
return;
}
c->SetNotifyOnCompactionCompleted();
Version* current = cfd->current();
current->Ref();
// release lock while notifying events
mutex_.Unlock();
TEST_SYNC_POINT("DBImpl::NotifyOnCompactionBegin::UnlockMutex");
{
CompactionJobInfo info{};
BuildCompactionJobInfo(cfd, c, st, job_stats, job_id, current, &info);
for (auto listener : immutable_db_options_.listeners) {
listener->OnCompactionBegin(this, info);
}
info.status.PermitUncheckedError();
}
mutex_.Lock();
current->Unref();
#else
(void)cfd;
(void)c;
(void)st;
(void)job_stats;
(void)job_id;
#endif // ROCKSDB_LITE
}
void DBImpl::NotifyOnCompactionCompleted(
ColumnFamilyData* cfd, Compaction* c, const Status& st,
const CompactionJobStats& compaction_job_stats, const int job_id) {
#ifndef ROCKSDB_LITE
if (immutable_db_options_.listeners.size() == 0U) {
return;
}
mutex_.AssertHeld();
if (shutting_down_.load(std::memory_order_acquire)) {
return;
}
if (c->ShouldNotifyOnCompactionCompleted() == false) {
return;
}
Version* current = cfd->current();
current->Ref();
// release lock while notifying events
mutex_.Unlock();
TEST_SYNC_POINT("DBImpl::NotifyOnCompactionCompleted::UnlockMutex");
{
CompactionJobInfo info{};
BuildCompactionJobInfo(cfd, c, st, compaction_job_stats, job_id, current,
&info);
for (auto listener : immutable_db_options_.listeners) {
listener->OnCompactionCompleted(this, info);
}
}
mutex_.Lock();
current->Unref();
// no need to signal bg_cv_ as it will be signaled at the end of the
// flush process.
#else
(void)cfd;
(void)c;
(void)st;
(void)compaction_job_stats;
(void)job_id;
#endif // ROCKSDB_LITE
}
// REQUIREMENT: block all background work by calling PauseBackgroundWork()
// before calling this function
Status DBImpl::ReFitLevel(ColumnFamilyData* cfd, int level, int target_level) {
assert(level < cfd->NumberLevels());
if (target_level >= cfd->NumberLevels()) {
return Status::InvalidArgument("Target level exceeds number of levels");
}
SuperVersionContext sv_context(/* create_superversion */ true);
InstrumentedMutexLock guard_lock(&mutex_);
// only allow one thread refitting
if (refitting_level_) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[ReFitLevel] another thread is refitting");
return Status::NotSupported("another thread is refitting");
}
refitting_level_ = true;
const MutableCFOptions mutable_cf_options = *cfd->GetLatestMutableCFOptions();
// move to a smaller level
int to_level = target_level;
if (target_level < 0) {
to_level = FindMinimumEmptyLevelFitting(cfd, mutable_cf_options, level);
}
auto* vstorage = cfd->current()->storage_info();
if (to_level != level) {
if (to_level > level) {
if (level == 0) {
refitting_level_ = false;
return Status::NotSupported(
"Cannot change from level 0 to other levels.");
}
// Check levels are empty for a trivial move
for (int l = level + 1; l <= to_level; l++) {
if (vstorage->NumLevelFiles(l) > 0) {
refitting_level_ = false;
return Status::NotSupported(
"Levels between source and target are not empty for a move.");
}
}
} else {
// to_level < level
// Check levels are empty for a trivial move
for (int l = to_level; l < level; l++) {
if (vstorage->NumLevelFiles(l) > 0) {
refitting_level_ = false;
return Status::NotSupported(
"Levels between source and target are not empty for a move.");
}
}
}
ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
"[%s] Before refitting:\n%s", cfd->GetName().c_str(),
cfd->current()->DebugString().data());
VersionEdit edit;
edit.SetColumnFamily(cfd->GetID());
for (const auto& f : vstorage->LevelFiles(level)) {
edit.DeleteFile(level, f->fd.GetNumber());
edit.AddFile(
to_level, f->fd.GetNumber(), f->fd.GetPathId(), f->fd.GetFileSize(),
f->smallest, f->largest, f->fd.smallest_seqno, f->fd.largest_seqno,
f->marked_for_compaction, f->temperature, f->oldest_blob_file_number,
f->oldest_ancester_time, f->file_creation_time, f->file_checksum,
f->file_checksum_func_name, f->min_timestamp, f->max_timestamp,
f->unique_id);
}
ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
"[%s] Apply version edit:\n%s", cfd->GetName().c_str(),
edit.DebugString().data());
Status status = versions_->LogAndApply(cfd, mutable_cf_options, &edit,
&mutex_, directories_.GetDbDir());
InstallSuperVersionAndScheduleWork(cfd, &sv_context, mutable_cf_options);
ROCKS_LOG_DEBUG(immutable_db_options_.info_log, "[%s] LogAndApply: %s\n",
cfd->GetName().c_str(), status.ToString().data());
if (status.ok()) {
ROCKS_LOG_DEBUG(immutable_db_options_.info_log,
"[%s] After refitting:\n%s", cfd->GetName().c_str(),
cfd->current()->DebugString().data());
}
sv_context.Clean();
refitting_level_ = false;
return status;
}
refitting_level_ = false;
return Status::OK();
}
int DBImpl::NumberLevels(ColumnFamilyHandle* column_family) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
return cfh->cfd()->NumberLevels();
}
int DBImpl::MaxMemCompactionLevel(ColumnFamilyHandle* /*column_family*/) {
return 0;
}
int DBImpl::Level0StopWriteTrigger(ColumnFamilyHandle* column_family) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
InstrumentedMutexLock l(&mutex_);
return cfh->cfd()
->GetSuperVersion()
->mutable_cf_options.level0_stop_writes_trigger;
}
Status DBImpl::Flush(const FlushOptions& flush_options,
ColumnFamilyHandle* column_family) {
auto cfh = static_cast_with_check<ColumnFamilyHandleImpl>(column_family);
ROCKS_LOG_INFO(immutable_db_options_.info_log, "[%s] Manual flush start.",
cfh->GetName().c_str());
Status s;
if (immutable_db_options_.atomic_flush) {
s = AtomicFlushMemTables({cfh->cfd()}, flush_options,
FlushReason::kManualFlush);
} else {
s = FlushMemTable(cfh->cfd(), flush_options, FlushReason::kManualFlush);
}
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[%s] Manual flush finished, status: %s\n",
cfh->GetName().c_str(), s.ToString().c_str());
return s;
}
Status DBImpl::Flush(const FlushOptions& flush_options,
const std::vector<ColumnFamilyHandle*>& column_families) {
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
Status s;
if (!immutable_db_options_.atomic_flush) {
for (auto cfh : column_families) {
s = Flush(flush_options, cfh);
if (!s.ok()) {
break;
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
}
} else {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Manual atomic flush start.\n"
"=====Column families:=====");
for (auto cfh : column_families) {
auto cfhi = static_cast<ColumnFamilyHandleImpl*>(cfh);
ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s",
cfhi->GetName().c_str());
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
}
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"=====End of column families list=====");
autovector<ColumnFamilyData*> cfds;
std::for_each(column_families.begin(), column_families.end(),
[&cfds](ColumnFamilyHandle* elem) {
auto cfh = static_cast<ColumnFamilyHandleImpl*>(elem);
cfds.emplace_back(cfh->cfd());
});
s = AtomicFlushMemTables(cfds, flush_options, FlushReason::kManualFlush);
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Manual atomic flush finished, status: %s\n"
"=====Column families:=====",
s.ToString().c_str());
for (auto cfh : column_families) {
auto cfhi = static_cast<ColumnFamilyHandleImpl*>(cfh);
ROCKS_LOG_INFO(immutable_db_options_.info_log, "%s",
cfhi->GetName().c_str());
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
}
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"=====End of column families list=====");
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
}
return s;
}
Status DBImpl::RunManualCompaction(
ColumnFamilyData* cfd, int input_level, int output_level,
const CompactRangeOptions& compact_range_options, const Slice* begin,
const Slice* end, bool exclusive, bool disallow_trivial_move,
uint64_t max_file_num_to_ignore, const std::string& trim_ts) {
assert(input_level == ColumnFamilyData::kCompactAllLevels ||
input_level >= 0);
InternalKey begin_storage, end_storage;
CompactionArg* ca = nullptr;
bool scheduled = false;
bool unscheduled = false;
Env::Priority thread_pool_priority = Env::Priority::TOTAL;
bool manual_conflict = false;
ManualCompactionState manual(
cfd, input_level, output_level, compact_range_options.target_path_id,
exclusive, disallow_trivial_move, compact_range_options.canceled);
// For universal compaction, we enforce every manual compaction to compact
// all files.
if (begin == nullptr ||
cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
manual.begin = nullptr;
} else {
begin_storage.SetMinPossibleForUserKey(*begin);
manual.begin = &begin_storage;
}
if (end == nullptr ||
cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
manual.end = nullptr;
} else {
end_storage.SetMaxPossibleForUserKey(*end);
manual.end = &end_storage;
}
TEST_SYNC_POINT("DBImpl::RunManualCompaction:0");
TEST_SYNC_POINT("DBImpl::RunManualCompaction:1");
InstrumentedMutexLock l(&mutex_);
Prevent corruption with parallel manual compactions and `change_level == true` (#9077) Summary: The bug can impact the following scenario. There must be two `CompactRange()`s, call them A and B. Compaction A must have `change_level=true`. Compactions A and B must run in parallel, and new data must be added while they run as well. Now, on to the details of the race condition. Compaction A must reach the refitting phase while B's next step is to trivial move new data (i.e., data that has been inserted behind A) down to the same level that A's refit targets (`CompactRangeOptions::target_level`). B must be unregistered (i.e., has not yet called `AddManualCompaction()` for the current `RunManualCompaction()`) while A invokes `DisableManualCompaction()`s to prepare for refitting. In the old code, B could still proceed to register a manual compaction, while A had disabled manual compaction. The next part of the race condition is B picks and schedules a trivial move while A has released the lock in refitting phase in order to persist the LSM state change (i.e., the log phase of `LogAndApply()`). That way, B does not see the refitted data when picking a trivial-move compaction. So it is susceptible to picking one that overlaps. Finally, B executes the picked trivial-move compaction. Trivial-move compactions are special in that they never check whether manual compaction is disabled. So the picked compaction causing overlap ends up being applied, leading to LSM corruption if `force_consistency_checks=false`, or entering read-only mode with `Status::Corruption` if `force_consistency_checks=true` (the default). The fix is just to prevent B from registering itself in `RunManualCompaction()` while manual compactions are disabled, consequently preventing any trivial move or other compaction from being picked/scheduled. Thanks to siying for finding the bug. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9077 Test Plan: The test does not go all the way in exposing the bug because it requires a compaction to be picked/scheduled while logging LSM state change for RefitLevel(). But the fix is to make such a compaction not picked/scheduled in the first place, so any repro of that scenario would end up hanging RefitLevel() logging. So instead I just verified no such compaction is registered in the scenario where `RefitLevel()` disables manual compactions. Reviewed By: siying Differential Revision: D31921908 Pulled By: ajkr fbshipit-source-id: 9bb5d0e847ad428211227f40830c685c209fbecb
3 years ago
if (manual_compaction_paused_ > 0) {
// Does not make sense to `AddManualCompaction()` in this scenario since
// `DisableManualCompaction()` just waited for the manual compaction queue
// to drain. So return immediately.
TEST_SYNC_POINT("DBImpl::RunManualCompaction:PausedAtStart");
manual.status =
Prevent corruption with parallel manual compactions and `change_level == true` (#9077) Summary: The bug can impact the following scenario. There must be two `CompactRange()`s, call them A and B. Compaction A must have `change_level=true`. Compactions A and B must run in parallel, and new data must be added while they run as well. Now, on to the details of the race condition. Compaction A must reach the refitting phase while B's next step is to trivial move new data (i.e., data that has been inserted behind A) down to the same level that A's refit targets (`CompactRangeOptions::target_level`). B must be unregistered (i.e., has not yet called `AddManualCompaction()` for the current `RunManualCompaction()`) while A invokes `DisableManualCompaction()`s to prepare for refitting. In the old code, B could still proceed to register a manual compaction, while A had disabled manual compaction. The next part of the race condition is B picks and schedules a trivial move while A has released the lock in refitting phase in order to persist the LSM state change (i.e., the log phase of `LogAndApply()`). That way, B does not see the refitted data when picking a trivial-move compaction. So it is susceptible to picking one that overlaps. Finally, B executes the picked trivial-move compaction. Trivial-move compactions are special in that they never check whether manual compaction is disabled. So the picked compaction causing overlap ends up being applied, leading to LSM corruption if `force_consistency_checks=false`, or entering read-only mode with `Status::Corruption` if `force_consistency_checks=true` (the default). The fix is just to prevent B from registering itself in `RunManualCompaction()` while manual compactions are disabled, consequently preventing any trivial move or other compaction from being picked/scheduled. Thanks to siying for finding the bug. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9077 Test Plan: The test does not go all the way in exposing the bug because it requires a compaction to be picked/scheduled while logging LSM state change for RefitLevel(). But the fix is to make such a compaction not picked/scheduled in the first place, so any repro of that scenario would end up hanging RefitLevel() logging. So instead I just verified no such compaction is registered in the scenario where `RefitLevel()` disables manual compactions. Reviewed By: siying Differential Revision: D31921908 Pulled By: ajkr fbshipit-source-id: 9bb5d0e847ad428211227f40830c685c209fbecb
3 years ago
Status::Incomplete(Status::SubCode::kManualCompactionPaused);
manual.done = true;
return manual.status;
Prevent corruption with parallel manual compactions and `change_level == true` (#9077) Summary: The bug can impact the following scenario. There must be two `CompactRange()`s, call them A and B. Compaction A must have `change_level=true`. Compactions A and B must run in parallel, and new data must be added while they run as well. Now, on to the details of the race condition. Compaction A must reach the refitting phase while B's next step is to trivial move new data (i.e., data that has been inserted behind A) down to the same level that A's refit targets (`CompactRangeOptions::target_level`). B must be unregistered (i.e., has not yet called `AddManualCompaction()` for the current `RunManualCompaction()`) while A invokes `DisableManualCompaction()`s to prepare for refitting. In the old code, B could still proceed to register a manual compaction, while A had disabled manual compaction. The next part of the race condition is B picks and schedules a trivial move while A has released the lock in refitting phase in order to persist the LSM state change (i.e., the log phase of `LogAndApply()`). That way, B does not see the refitted data when picking a trivial-move compaction. So it is susceptible to picking one that overlaps. Finally, B executes the picked trivial-move compaction. Trivial-move compactions are special in that they never check whether manual compaction is disabled. So the picked compaction causing overlap ends up being applied, leading to LSM corruption if `force_consistency_checks=false`, or entering read-only mode with `Status::Corruption` if `force_consistency_checks=true` (the default). The fix is just to prevent B from registering itself in `RunManualCompaction()` while manual compactions are disabled, consequently preventing any trivial move or other compaction from being picked/scheduled. Thanks to siying for finding the bug. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9077 Test Plan: The test does not go all the way in exposing the bug because it requires a compaction to be picked/scheduled while logging LSM state change for RefitLevel(). But the fix is to make such a compaction not picked/scheduled in the first place, so any repro of that scenario would end up hanging RefitLevel() logging. So instead I just verified no such compaction is registered in the scenario where `RefitLevel()` disables manual compactions. Reviewed By: siying Differential Revision: D31921908 Pulled By: ajkr fbshipit-source-id: 9bb5d0e847ad428211227f40830c685c209fbecb
3 years ago
}
// When a manual compaction arrives, temporarily disable scheduling of
// non-manual compactions and wait until the number of scheduled compaction
// jobs drops to zero. This used to be needed to ensure that this manual
// compaction can compact any range of keys/files. Now it is optional
// (see `CompactRangeOptions::exclusive_manual_compaction`). The use case for
// `exclusive_manual_compaction=true` (the default) is unclear beyond not
// trusting the new code.
//
// HasPendingManualCompaction() is true when at least one thread is inside
// RunManualCompaction(), i.e. during that time no other compaction will
// get scheduled (see MaybeScheduleFlushOrCompaction).
//
// Note that the following loop doesn't stop more that one thread calling
// RunManualCompaction() from getting to the second while loop below.
// However, only one of them will actually schedule compaction, while
// others will wait on a condition variable until it completes.
AddManualCompaction(&manual);
TEST_SYNC_POINT_CALLBACK("DBImpl::RunManualCompaction:NotScheduled", &mutex_);
if (exclusive) {
// Limitation: there's no way to wake up the below loop when user sets
// `*manual.canceled`. So `CompactRangeOptions::exclusive_manual_compaction`
// and `CompactRangeOptions::canceled` might not work well together.
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
while (bg_bottom_compaction_scheduled_ > 0 ||
bg_compaction_scheduled_ > 0) {
if (manual_compaction_paused_ > 0 || manual.canceled == true) {
// Pretend the error came from compaction so the below cleanup/error
// handling code can process it.
manual.done = true;
manual.status =
Status::Incomplete(Status::SubCode::kManualCompactionPaused);
break;
}
TEST_SYNC_POINT("DBImpl::RunManualCompaction:WaitScheduled");
ROCKS_LOG_INFO(
immutable_db_options_.info_log,
"[%s] Manual compaction waiting for all other scheduled background "
"compactions to finish",
cfd->GetName().c_str());
bg_cv_.Wait();
}
}
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
immutable_db_options_.info_log.get());
ROCKS_LOG_BUFFER(&log_buffer, "[%s] Manual compaction starting",
cfd->GetName().c_str());
// We don't check bg_error_ here, because if we get the error in compaction,
// the compaction will set manual.status to bg_error_ and set manual.done to
// true.
while (!manual.done) {
assert(HasPendingManualCompaction());
manual_conflict = false;
Compaction* compaction = nullptr;
if (ShouldntRunManualCompaction(&manual) || (manual.in_progress == true) ||
scheduled ||
(((manual.manual_end = &manual.tmp_storage1) != nullptr) &&
((compaction = manual.cfd->CompactRange(
*manual.cfd->GetLatestMutableCFOptions(), mutable_db_options_,
manual.input_level, manual.output_level, compact_range_options,
manual.begin, manual.end, &manual.manual_end, &manual_conflict,
max_file_num_to_ignore, trim_ts)) == nullptr &&
manual_conflict))) {
// exclusive manual compactions should not see a conflict during
// CompactRange
assert(!exclusive || !manual_conflict);
// Running either this or some other manual compaction
bg_cv_.Wait();
if (manual_compaction_paused_ > 0 && scheduled && !unscheduled) {
assert(thread_pool_priority != Env::Priority::TOTAL);
// unschedule all manual compactions
auto unscheduled_task_num = env_->UnSchedule(
GetTaskTag(TaskType::kManualCompaction), thread_pool_priority);
if (unscheduled_task_num > 0) {
ROCKS_LOG_INFO(
immutable_db_options_.info_log,
"[%s] Unscheduled %d number of manual compactions from the "
"thread-pool",
cfd->GetName().c_str(), unscheduled_task_num);
// it may unschedule other manual compactions, notify others.
bg_cv_.SignalAll();
}
unscheduled = true;
TEST_SYNC_POINT("DBImpl::RunManualCompaction:Unscheduled");
}
if (scheduled && manual.incomplete == true) {
assert(!manual.in_progress);
scheduled = false;
manual.incomplete = false;
}
} else if (!scheduled) {
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
if (compaction == nullptr) {
manual.done = true;
bg_cv_.SignalAll();
continue;
}
ca = new CompactionArg;
ca->db = this;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
ca->prepicked_compaction = new PrepickedCompaction;
ca->prepicked_compaction->manual_compaction_state = &manual;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
ca->prepicked_compaction->compaction = compaction;
if (!RequestCompactionToken(
cfd, true, &ca->prepicked_compaction->task_token, &log_buffer)) {
// Don't throttle manual compaction, only count outstanding tasks.
assert(false);
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
}
manual.incomplete = false;
if (compaction->bottommost_level() &&
env_->GetBackgroundThreads(Env::Priority::BOTTOM) > 0) {
bg_bottom_compaction_scheduled_++;
ca->compaction_pri_ = Env::Priority::BOTTOM;
env_->Schedule(&DBImpl::BGWorkBottomCompaction, ca,
Env::Priority::BOTTOM,
GetTaskTag(TaskType::kManualCompaction),
&DBImpl::UnscheduleCompactionCallback);
thread_pool_priority = Env::Priority::BOTTOM;
} else {
bg_compaction_scheduled_++;
ca->compaction_pri_ = Env::Priority::LOW;
env_->Schedule(&DBImpl::BGWorkCompaction, ca, Env::Priority::LOW,
GetTaskTag(TaskType::kManualCompaction),
&DBImpl::UnscheduleCompactionCallback);
thread_pool_priority = Env::Priority::LOW;
}
scheduled = true;
TEST_SYNC_POINT("DBImpl::RunManualCompaction:Scheduled");
}
}
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
log_buffer.FlushBufferToLog();
assert(!manual.in_progress);
assert(HasPendingManualCompaction());
RemoveManualCompaction(&manual);
// if the manual job is unscheduled, try schedule other jobs in case there's
// any unscheduled compaction job which was blocked by exclusive manual
// compaction.
if (manual.status.IsIncomplete() &&
manual.status.subcode() == Status::SubCode::kManualCompactionPaused) {
MaybeScheduleFlushOrCompaction();
}
bg_cv_.SignalAll();
return manual.status;
}
void DBImpl::GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
FlushRequest* req) {
assert(req != nullptr);
req->reserve(cfds.size());
for (const auto cfd : cfds) {
if (nullptr == cfd) {
// cfd may be null, see DBImpl::ScheduleFlushes
continue;
}
uint64_t max_memtable_id = cfd->imm()->GetLatestMemTableID();
req->emplace_back(cfd, max_memtable_id);
}
}
Status DBImpl::FlushMemTable(ColumnFamilyData* cfd,
const FlushOptions& flush_options,
FlushReason flush_reason, bool writes_stopped) {
// This method should not be called if atomic_flush is true.
assert(!immutable_db_options_.atomic_flush);
Status s;
if (!flush_options.allow_write_stall) {
bool flush_needed = true;
s = WaitUntilFlushWouldNotStallWrites(cfd, &flush_needed);
TEST_SYNC_POINT("DBImpl::FlushMemTable:StallWaitDone");
if (!s.ok() || !flush_needed) {
return s;
}
}
autovector<FlushRequest> flush_reqs;
autovector<uint64_t> memtable_ids_to_wait;
{
WriteContext context;
InstrumentedMutexLock guard_lock(&mutex_);
WriteThread::Writer w;
WriteThread::Writer nonmem_w;
if (!writes_stopped) {
write_thread_.EnterUnbatched(&w, &mutex_);
if (two_write_queues_) {
nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
}
}
WaitForPendingWrites();
if (flush_reason != FlushReason::kErrorRecoveryRetryFlush &&
(!cfd->mem()->IsEmpty() || !cached_recoverable_state_empty_.load())) {
// Note that, when flush reason is kErrorRecoveryRetryFlush, during the
// auto retry resume, we want to avoid creating new small memtables.
// Therefore, SwitchMemtable will not be called. Also, since ResumeImpl
// will iterate through all the CFs and call FlushMemtable during auto
// retry resume, it is possible that in some CFs,
// cfd->imm()->NumNotFlushed() = 0. In this case, so no flush request will
// be created and scheduled, status::OK() will be returned.
s = SwitchMemtable(cfd, &context);
}
const uint64_t flush_memtable_id = std::numeric_limits<uint64_t>::max();
if (s.ok()) {
if (cfd->imm()->NumNotFlushed() != 0 || !cfd->mem()->IsEmpty() ||
!cached_recoverable_state_empty_.load()) {
FlushRequest req{{cfd, flush_memtable_id}};
flush_reqs.emplace_back(std::move(req));
memtable_ids_to_wait.emplace_back(cfd->imm()->GetLatestMemTableID());
}
if (immutable_db_options_.persist_stats_to_disk &&
flush_reason != FlushReason::kErrorRecoveryRetryFlush) {
ColumnFamilyData* cfd_stats =
versions_->GetColumnFamilySet()->GetColumnFamily(
kPersistentStatsColumnFamilyName);
if (cfd_stats != nullptr && cfd_stats != cfd &&
!cfd_stats->mem()->IsEmpty()) {
// only force flush stats CF when it will be the only CF lagging
// behind after the current flush
bool stats_cf_flush_needed = true;
for (auto* loop_cfd : *versions_->GetColumnFamilySet()) {
if (loop_cfd == cfd_stats || loop_cfd == cfd) {
continue;
}
if (loop_cfd->GetLogNumber() <= cfd_stats->GetLogNumber()) {
stats_cf_flush_needed = false;
}
}
if (stats_cf_flush_needed) {
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"Force flushing stats CF with manual flush of %s "
"to avoid holding old logs",
cfd->GetName().c_str());
s = SwitchMemtable(cfd_stats, &context);
FlushRequest req{{cfd_stats, flush_memtable_id}};
flush_reqs.emplace_back(std::move(req));
memtable_ids_to_wait.emplace_back(
cfd->imm()->GetLatestMemTableID());
}
}
}
}
if (s.ok() && !flush_reqs.empty()) {
for (const auto& req : flush_reqs) {
assert(req.size() == 1);
ColumnFamilyData* loop_cfd = req[0].first;
loop_cfd->imm()->FlushRequested();
}
// If the caller wants to wait for this flush to complete, it indicates
// that the caller expects the ColumnFamilyData not to be free'ed by
// other threads which may drop the column family concurrently.
// Therefore, we increase the cfd's ref count.
if (flush_options.wait) {
for (const auto& req : flush_reqs) {
assert(req.size() == 1);
ColumnFamilyData* loop_cfd = req[0].first;
loop_cfd->Ref();
}
}
for (const auto& req : flush_reqs) {
SchedulePendingFlush(req, flush_reason);
}
MaybeScheduleFlushOrCompaction();
}
if (!writes_stopped) {
write_thread_.ExitUnbatched(&w);
if (two_write_queues_) {
nonmem_write_thread_.ExitUnbatched(&nonmem_w);
}
}
}
TEST_SYNC_POINT("DBImpl::FlushMemTable:AfterScheduleFlush");
TEST_SYNC_POINT("DBImpl::FlushMemTable:BeforeWaitForBgFlush");
if (s.ok() && flush_options.wait) {
autovector<ColumnFamilyData*> cfds;
autovector<const uint64_t*> flush_memtable_ids;
assert(flush_reqs.size() == memtable_ids_to_wait.size());
for (size_t i = 0; i < flush_reqs.size(); ++i) {
assert(flush_reqs[i].size() == 1);
cfds.push_back(flush_reqs[i][0].first);
flush_memtable_ids.push_back(&(memtable_ids_to_wait[i]));
}
s = WaitForFlushMemTables(
cfds, flush_memtable_ids,
(flush_reason == FlushReason::kErrorRecovery ||
flush_reason == FlushReason::kErrorRecoveryRetryFlush));
InstrumentedMutexLock lock_guard(&mutex_);
for (auto* tmp_cfd : cfds) {
tmp_cfd->UnrefAndTryDelete();
}
}
TEST_SYNC_POINT("DBImpl::FlushMemTable:FlushMemTableFinished");
return s;
}
// Flush all elements in 'column_family_datas'
// and atomically record the result to the MANIFEST.
Status DBImpl::AtomicFlushMemTables(
const autovector<ColumnFamilyData*>& column_family_datas,
const FlushOptions& flush_options, FlushReason flush_reason,
bool writes_stopped) {
Status s;
if (!flush_options.allow_write_stall) {
int num_cfs_to_flush = 0;
for (auto cfd : column_family_datas) {
bool flush_needed = true;
s = WaitUntilFlushWouldNotStallWrites(cfd, &flush_needed);
if (!s.ok()) {
return s;
} else if (flush_needed) {
++num_cfs_to_flush;
}
}
if (0 == num_cfs_to_flush) {
return s;
}
}
FlushRequest flush_req;
autovector<ColumnFamilyData*> cfds;
{
WriteContext context;
InstrumentedMutexLock guard_lock(&mutex_);
WriteThread::Writer w;
WriteThread::Writer nonmem_w;
if (!writes_stopped) {
write_thread_.EnterUnbatched(&w, &mutex_);
if (two_write_queues_) {
nonmem_write_thread_.EnterUnbatched(&nonmem_w, &mutex_);
}
}
WaitForPendingWrites();
for (auto cfd : column_family_datas) {
if (cfd->IsDropped()) {
continue;
}
if (cfd->imm()->NumNotFlushed() != 0 || !cfd->mem()->IsEmpty() ||
!cached_recoverable_state_empty_.load()) {
cfds.emplace_back(cfd);
}
}
for (auto cfd : cfds) {
if ((cfd->mem()->IsEmpty() && cached_recoverable_state_empty_.load()) ||
flush_reason == FlushReason::kErrorRecoveryRetryFlush) {
continue;
}
cfd->Ref();
s = SwitchMemtable(cfd, &context);
cfd->UnrefAndTryDelete();
if (!s.ok()) {
break;
}
}
if (s.ok()) {
AssignAtomicFlushSeq(cfds);
for (auto cfd : cfds) {
cfd->imm()->FlushRequested();
}
// If the caller wants to wait for this flush to complete, it indicates
// that the caller expects the ColumnFamilyData not to be free'ed by
// other threads which may drop the column family concurrently.
// Therefore, we increase the cfd's ref count.
if (flush_options.wait) {
for (auto cfd : cfds) {
cfd->Ref();
}
}
GenerateFlushRequest(cfds, &flush_req);
SchedulePendingFlush(flush_req, flush_reason);
MaybeScheduleFlushOrCompaction();
}
if (!writes_stopped) {
write_thread_.ExitUnbatched(&w);
if (two_write_queues_) {
nonmem_write_thread_.ExitUnbatched(&nonmem_w);
}
}
}
TEST_SYNC_POINT("DBImpl::AtomicFlushMemTables:AfterScheduleFlush");
TEST_SYNC_POINT("DBImpl::AtomicFlushMemTables:BeforeWaitForBgFlush");
if (s.ok() && flush_options.wait) {
autovector<const uint64_t*> flush_memtable_ids;
for (auto& iter : flush_req) {
flush_memtable_ids.push_back(&(iter.second));
}
s = WaitForFlushMemTables(
cfds, flush_memtable_ids,
(flush_reason == FlushReason::kErrorRecovery ||
flush_reason == FlushReason::kErrorRecoveryRetryFlush));
InstrumentedMutexLock lock_guard(&mutex_);
for (auto* cfd : cfds) {
cfd->UnrefAndTryDelete();
}
}
return s;
}
// Calling FlushMemTable(), whether from DB::Flush() or from Backup Engine, can
// cause write stall, for example if one memtable is being flushed already.
// This method tries to avoid write stall (similar to CompactRange() behavior)
// it emulates how the SuperVersion / LSM would change if flush happens, checks
// it against various constrains and delays flush if it'd cause write stall.
// Caller should check status and flush_needed to see if flush already happened.
Status DBImpl::WaitUntilFlushWouldNotStallWrites(ColumnFamilyData* cfd,
bool* flush_needed) {
{
*flush_needed = true;
InstrumentedMutexLock l(&mutex_);
uint64_t orig_active_memtable_id = cfd->mem()->GetID();
WriteStallCondition write_stall_condition = WriteStallCondition::kNormal;
do {
if (write_stall_condition != WriteStallCondition::kNormal) {
// Same error handling as user writes: Don't wait if there's a
// background error, even if it's a soft error. We might wait here
// indefinitely as the pending flushes/compactions may never finish
// successfully, resulting in the stall condition lasting indefinitely
if (error_handler_.IsBGWorkStopped()) {
return error_handler_.GetBGError();
}
TEST_SYNC_POINT("DBImpl::WaitUntilFlushWouldNotStallWrites:StallWait");
ROCKS_LOG_INFO(immutable_db_options_.info_log,
"[%s] WaitUntilFlushWouldNotStallWrites"
" waiting on stall conditions to clear",
cfd->GetName().c_str());
bg_cv_.Wait();
}
if (cfd->IsDropped()) {
return Status::ColumnFamilyDropped();
}
if (shutting_down_.load(std::memory_order_acquire)) {
return Status::ShutdownInProgress();
}
uint64_t earliest_memtable_id =
std::min(cfd->mem()->GetID(), cfd->imm()->GetEarliestMemTableID());
if (earliest_memtable_id > orig_active_memtable_id) {
// We waited so long that the memtable we were originally waiting on was
// flushed.
*flush_needed = false;
return Status::OK();
}
const auto& mutable_cf_options = *cfd->GetLatestMutableCFOptions();
const auto* vstorage = cfd->current()->storage_info();
// Skip stalling check if we're below auto-flush and auto-compaction
// triggers. If it stalled in these conditions, that'd mean the stall
// triggers are so low that stalling is needed for any background work. In
// that case we shouldn't wait since background work won't be scheduled.
if (cfd->imm()->NumNotFlushed() <
cfd->ioptions()->min_write_buffer_number_to_merge &&
vstorage->l0_delay_trigger_count() <
mutable_cf_options.level0_file_num_compaction_trigger) {
break;
}
// check whether one extra immutable memtable or an extra L0 file would
// cause write stalling mode to be entered. It could still enter stall
// mode due to pending compaction bytes, but that's less common
Fix checkpoint stuck (#7921) Summary: ## 1. Bug description: When RocksDB Checkpoint, it may be stuck in `WaitUntilFlushWouldNotStallWrites` method. ## 2. Simple analysis of the reasons: ### 2.1 Configuration parameters: ```yaml Compaction Style : Universal max_write_buffer_number : 4 min_write_buffer_number_to_merge : 3 ``` Checkpoint is usually very fast. When the Checkpoint is executed, `WaitUntilFlushWouldNotStallWrites` is called. If there are 2 Immutable MemTables, which are less than `min_write_buffer_number_to_merge`, they will not be flushed. But will enter this code. ```c++ // method: GetWriteStallConditionAndCause if (mutable_cf_options.max_write_buffer_number> 3 && num_unflushed_memtables >= mutable_cf_options.max_write_buffer_number-1) { return {WriteStallCondition::kDelayed, WriteStallCause::kMemtableLimit}; } ``` code link: https://github.com/facebook/rocksdb/blob/fbed72f03c3d9e4fdca3e5993587ef2559ba6ab9/db/column_family.cc#L847 Checkpoint thought there was a FlushJob, but it didn't. So will always wait. ### 2.2 solution: Increase the restriction: the `number of Immutable MemTable` >= `min_write_buffer_number_to_merge will wait`. If there are other better solutions, you can correct me. ### 2.3 Code that can reproduce the problem: https://github.com/1996fanrui/fanrui-learning/blob/flink-1.12/module-java/src/main/java/com/dream/rocksdb/RocksDBCheckpointStuck.java ## 3. Interesting point This bug will be triggered only when `the number of sorted runs >= level0_file_num_compaction_trigger`. Because there is a break in WaitUntilFlushWouldNotStallWrites. ```c++ if (cfd->imm()->NumNotFlushed() < cfd->ioptions()->min_write_buffer_number_to_merge && vstorage->l0_delay_trigger_count() < mutable_cf_options.level0_file_num_compaction_trigger) { break; } ``` code link: https://github.com/facebook/rocksdb/blob/fbed72f03c3d9e4fdca3e5993587ef2559ba6ab9/db/db_impl/db_impl_compaction_flush.cc#L1974 Universal may have `l0_delay_trigger_count() >= level0_file_num_compaction_trigger`, so this bug is triggered. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7921 Reviewed By: jay-zhuang Differential Revision: D26900559 Pulled By: ajkr fbshipit-source-id: 133c1252dad7393753f04a47590b68c7d8e670df
4 years ago
write_stall_condition = ColumnFamilyData::GetWriteStallConditionAndCause(
cfd->imm()->NumNotFlushed() + 1,
vstorage->l0_delay_trigger_count() + 1,
vstorage->estimated_compaction_needed_bytes(),
mutable_cf_options, *cfd->ioptions())
.first;
} while (write_stall_condition != WriteStallCondition::kNormal);
}
return Status::OK();
}
// Wait for memtables to be flushed for multiple column families.
// let N = cfds.size()
// for i in [0, N),
// 1) if flush_memtable_ids[i] is not null, then the memtables with lower IDs
// have to be flushed for THIS column family;
// 2) if flush_memtable_ids[i] is null, then all memtables in THIS column
// family have to be flushed.
// Finish waiting when ALL column families finish flushing memtables.
// resuming_from_bg_err indicates whether the caller is trying to resume from
// background error or in normal processing.
Status DBImpl::WaitForFlushMemTables(
const autovector<ColumnFamilyData*>& cfds,
const autovector<const uint64_t*>& flush_memtable_ids,
bool resuming_from_bg_err) {
int num = static_cast<int>(cfds.size());
// Wait until the compaction completes
InstrumentedMutexLock l(&mutex_);
Status s;
// If the caller is trying to resume from bg error, then
// error_handler_.IsDBStopped() is true.
while (resuming_from_bg_err || !error_handler_.IsDBStopped()) {
if (shutting_down_.load(std::memory_order_acquire)) {
s = Status::ShutdownInProgress();
return s;
}
// If an error has occurred during resumption, then no need to wait.
// But flush operation may fail because of this error, so need to
// return the status.
if (!error_handler_.GetRecoveryError().ok()) {
s = error_handler_.GetRecoveryError();
break;
}
// If BGWorkStopped, which indicate that there is a BG error and
// 1) soft error but requires no BG work, 2) no in auto_recovery_
if (!resuming_from_bg_err && error_handler_.IsBGWorkStopped() &&
error_handler_.GetBGError().severity() < Status::Severity::kHardError) {
s = error_handler_.GetBGError();
return s;
}
// Number of column families that have been dropped.
int num_dropped = 0;
// Number of column families that have finished flush.
int num_finished = 0;
for (int i = 0; i < num; ++i) {
if (cfds[i]->IsDropped()) {
++num_dropped;
} else if (cfds[i]->imm()->NumNotFlushed() == 0 ||
(flush_memtable_ids[i] != nullptr &&
cfds[i]->imm()->GetEarliestMemTableID() >
*flush_memtable_ids[i])) {
++num_finished;
}
}
if (1 == num_dropped && 1 == num) {
s = Status::ColumnFamilyDropped();
return s;
}
// Column families involved in this flush request have either been dropped
// or finished flush. Then it's time to finish waiting.
if (num_dropped + num_finished == num) {
break;
}
bg_cv_.Wait();
}
// If not resuming from bg error, and an error has caused the DB to stop,
// then report the bg error to caller.
if (!resuming_from_bg_err && error_handler_.IsDBStopped()) {
s = error_handler_.GetBGError();
}
return s;
}
Status DBImpl::EnableAutoCompaction(
const std::vector<ColumnFamilyHandle*>& column_family_handles) {
Status s;
for (auto cf_ptr : column_family_handles) {
Status status =
this->SetOptions(cf_ptr, {{"disable_auto_compactions", "false"}});
if (!status.ok()) {
s = status;
}
}
return s;
}
// NOTE: Calling DisableManualCompaction() may overwrite the
// user-provided canceled variable in CompactRangeOptions
void DBImpl::DisableManualCompaction() {
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
InstrumentedMutexLock l(&mutex_);
manual_compaction_paused_.fetch_add(1, std::memory_order_release);
// Mark the canceled as true when the cancellation is triggered by
// manual_compaction_paused (may overwrite user-provided `canceled`)
for (const auto& manual_compaction : manual_compaction_dequeue_) {
manual_compaction->canceled = true;
}
// Wake up manual compactions waiting to start.
bg_cv_.SignalAll();
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
// Wait for any pending manual compactions to finish (typically through
// failing with `Status::Incomplete`) prior to returning. This way we are
// guaranteed no pending manual compaction will commit while manual
// compactions are "disabled".
while (HasPendingManualCompaction()) {
bg_cv_.Wait();
}
}
// NOTE: In contrast to DisableManualCompaction(), calling
// EnableManualCompaction() does NOT overwrite the user-provided *canceled
// variable to be false since there is NO CHANCE a canceled compaction
// is uncanceled. In other words, a canceled compaction must have been
// dropped out of the manual compaction queue, when we disable it.
void DBImpl::EnableManualCompaction() {
Disable manual compaction during `ReFitLevel()` (#7250) Summary: Manual compaction with `CompactRangeOptions::change_levels` set could refit to a level targeted by another manual compaction. If force_consistency_checks were disabled, it could be possible for overlapping files to be written at that target level. This PR prevents the possibility by calling `DisableManualCompaction()` prior to `ReFitLevel()`. It also improves the manual compaction disabling mechanism to wait for pending manual compactions to complete before returning, and support disabling from multiple threads. Fixes https://github.com/facebook/rocksdb/issues/6432. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7250 Test Plan: crash test command that repro'd the bug reliably: ``` $ TEST_TMPDIR=/dev/shm python tools/db_crashtest.py blackbox --simple -target_file_size_base=524288 -write_buffer_size=1048576 -clear_column_family_one_in=0 -reopen=0 -max_key=10000000 -column_families=1 -max_background_compactions=8 -compact_range_one_in=100000 -compression_type=none -compaction_style=1 -num_levels=5 -universal_min_merge_width=4 -universal_max_merge_width=8 -level0_file_num_compaction_trigger=12 -rate_limiter_bytes_per_sec=1048576000 -universal_max_size_amplification_percent=100 --duration=3600 --interval=60 --use_direct_io_for_flush_and_compaction=0 --use_direct_reads=0 --enable_compaction_filter=0 ``` Reviewed By: ltamasi Differential Revision: D23090800 Pulled By: ajkr fbshipit-source-id: afcbcd51b42ce76789fdb907d8b9ada790709c13
4 years ago
InstrumentedMutexLock l(&mutex_);
assert(manual_compaction_paused_ > 0);
manual_compaction_paused_.fetch_sub(1, std::memory_order_release);
}
void DBImpl::MaybeScheduleFlushOrCompaction() {
mutex_.AssertHeld();
if (!opened_successfully_) {
// Compaction may introduce data race to DB open
return;
}
if (bg_work_paused_ > 0) {
// we paused the background work
return;
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
} else if (error_handler_.IsBGWorkStopped() &&
!error_handler_.IsRecoveryInProgress()) {
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// There has been a hard error and this call is not part of the recovery
// sequence. Bail out here so we don't get into an endless loop of
// scheduling BG work which will again call this function
return;
} else if (shutting_down_.load(std::memory_order_acquire)) {
// DB is being deleted; no more background compactions
return;
}
auto bg_job_limits = GetBGJobLimits();
bool is_flush_pool_empty =
env_->GetBackgroundThreads(Env::Priority::HIGH) == 0;
while (!is_flush_pool_empty && unscheduled_flushes_ > 0 &&
bg_flush_scheduled_ < bg_job_limits.max_flushes) {
bg_flush_scheduled_++;
FlushThreadArg* fta = new FlushThreadArg;
fta->db_ = this;
fta->thread_pri_ = Env::Priority::HIGH;
env_->Schedule(&DBImpl::BGWorkFlush, fta, Env::Priority::HIGH, this,
&DBImpl::UnscheduleFlushCallback);
--unscheduled_flushes_;
TEST_SYNC_POINT_CALLBACK(
"DBImpl::MaybeScheduleFlushOrCompaction:AfterSchedule:0",
&unscheduled_flushes_);
}
// special case -- if high-pri (flush) thread pool is empty, then schedule
// flushes in low-pri (compaction) thread pool.
if (is_flush_pool_empty) {
while (unscheduled_flushes_ > 0 &&
bg_flush_scheduled_ + bg_compaction_scheduled_ <
bg_job_limits.max_flushes) {
bg_flush_scheduled_++;
FlushThreadArg* fta = new FlushThreadArg;
fta->db_ = this;
fta->thread_pri_ = Env::Priority::LOW;
env_->Schedule(&DBImpl::BGWorkFlush, fta, Env::Priority::LOW, this,
&DBImpl::UnscheduleFlushCallback);
--unscheduled_flushes_;
}
}
if (bg_compaction_paused_ > 0) {
// we paused the background compaction
return;
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
} else if (error_handler_.IsBGWorkStopped()) {
// Compaction is not part of the recovery sequence from a hard error. We
// might get here because recovery might do a flush and install a new
// super version, which will try to schedule pending compactions. Bail
// out here and let the higher level recovery handle compactions
return;
}
if (HasExclusiveManualCompaction()) {
// only manual compactions are allowed to run. don't schedule automatic
// compactions
TEST_SYNC_POINT("DBImpl::MaybeScheduleFlushOrCompaction:Conflict");
return;
}
while (bg_compaction_scheduled_ + bg_bottom_compaction_scheduled_ <
bg_job_limits.max_compactions &&
unscheduled_compactions_ > 0) {
CompactionArg* ca = new CompactionArg;
ca->db = this;
ca->compaction_pri_ = Env::Priority::LOW;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
ca->prepicked_compaction = nullptr;
bg_compaction_scheduled_++;
unscheduled_compactions_--;
env_->Schedule(&DBImpl::BGWorkCompaction, ca, Env::Priority::LOW, this,
&DBImpl::UnscheduleCompactionCallback);
}
}
DBImpl::BGJobLimits DBImpl::GetBGJobLimits() const {
mutex_.AssertHeld();
return GetBGJobLimits(mutable_db_options_.max_background_flushes,
mutable_db_options_.max_background_compactions,
mutable_db_options_.max_background_jobs,
write_controller_.NeedSpeedupCompaction());
}
DBImpl::BGJobLimits DBImpl::GetBGJobLimits(int max_background_flushes,
int max_background_compactions,
int max_background_jobs,
bool parallelize_compactions) {
BGJobLimits res;
if (max_background_flushes == -1 && max_background_compactions == -1) {
// for our first stab implementing max_background_jobs, simply allocate a
// quarter of the threads to flushes.
res.max_flushes = std::max(1, max_background_jobs / 4);
res.max_compactions = std::max(1, max_background_jobs - res.max_flushes);
} else {
// compatibility code in case users haven't migrated to max_background_jobs,
// which automatically computes flush/compaction limits
res.max_flushes = std::max(1, max_background_flushes);
res.max_compactions = std::max(1, max_background_compactions);
}
if (!parallelize_compactions) {
// throttle background compactions until we deem necessary
res.max_compactions = 1;
}
return res;
}
void DBImpl::AddToCompactionQueue(ColumnFamilyData* cfd) {
assert(!cfd->queued_for_compaction());
cfd->Ref();
compaction_queue_.push_back(cfd);
cfd->set_queued_for_compaction(true);
}
ColumnFamilyData* DBImpl::PopFirstFromCompactionQueue() {
assert(!compaction_queue_.empty());
auto cfd = *compaction_queue_.begin();
compaction_queue_.pop_front();
assert(cfd->queued_for_compaction());
cfd->set_queued_for_compaction(false);
return cfd;
}
DBImpl::FlushRequest DBImpl::PopFirstFromFlushQueue() {
assert(!flush_queue_.empty());
FlushRequest flush_req = flush_queue_.front();
flush_queue_.pop_front();
if (!immutable_db_options_.atomic_flush) {
assert(flush_req.size() == 1);
}
for (const auto& elem : flush_req) {
if (!immutable_db_options_.atomic_flush) {
ColumnFamilyData* cfd = elem.first;
assert(cfd);
assert(cfd->queued_for_flush());
cfd->set_queued_for_flush(false);
}
}
// TODO: need to unset flush reason?
return flush_req;
}
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
ColumnFamilyData* DBImpl::PickCompactionFromQueue(
std::unique_ptr<TaskLimiterToken>* token, LogBuffer* log_buffer) {
assert(!compaction_queue_.empty());
assert(*token == nullptr);
autovector<ColumnFamilyData*> throttled_candidates;
ColumnFamilyData* cfd = nullptr;
while (!compaction_queue_.empty()) {
auto first_cfd = *compaction_queue_.begin();
compaction_queue_.pop_front();
assert(first_cfd->queued_for_compaction());
if (!RequestCompactionToken(first_cfd, false, token, log_buffer)) {
throttled_candidates.push_back(first_cfd);
continue;
}
cfd = first_cfd;
cfd->set_queued_for_compaction(false);
break;
}
// Add throttled compaction candidates back to queue in the original order.
for (auto iter = throttled_candidates.rbegin();
iter != throttled_candidates.rend(); ++iter) {
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
compaction_queue_.push_front(*iter);
}
return cfd;
}
void DBImpl::SchedulePendingFlush(const FlushRequest& flush_req,
FlushReason flush_reason) {
mutex_.AssertHeld();
if (flush_req.empty()) {
return;
}
if (!immutable_db_options_.atomic_flush) {
// For the non-atomic flush case, we never schedule multiple column
// families in the same flush request.
assert(flush_req.size() == 1);
ColumnFamilyData* cfd = flush_req[0].first;
assert(cfd);
Dynamically changeable `MemPurge` option (#10011) Summary: **Summary** Make the mempurge option flag a Mutable Column Family option flag. Therefore, the mempurge feature can be dynamically toggled. **Motivation** RocksDB users prefer having the ability to switch features on and off without having to close and reopen the DB. This is particularly important if the feature causes issues and needs to be turned off. Dynamically changing a DB option flag does not seem currently possible. Moreover, with this new change, the MemPurge feature can be toggled on or off independently between column families, which we see as a major improvement. **Content of this PR** This PR includes removal of the `experimental_mempurge_threshold` flag as a DB option flag, and its re-introduction as a `MutableCFOption` flag. I updated the code to handle dynamic changes of the flag (in particular inside the `FlushJob` file). Additionally, this PR includes a new test to demonstrate the capacity of the code to toggle the MemPurge feature on and off, as well as the addition in the `db_stress` module of 2 different mempurge threshold values (0.0 and 1.0) that can be randomly changed with the `set_option_one_in` flag. This is useful to stress test the dynamic changes. **Benchmarking** I will add numbers to prove that there is no performance impact within the next 12 hours. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10011 Reviewed By: pdillinger Differential Revision: D36462357 Pulled By: bjlemaire fbshipit-source-id: 5e3d63bdadf085c0572ecc2349e7dd9729ce1802
2 years ago
if (!cfd->queued_for_flush() && cfd->imm()->IsFlushPending()) {
cfd->Ref();
cfd->set_queued_for_flush(true);
cfd->SetFlushReason(flush_reason);
++unscheduled_flushes_;
flush_queue_.push_back(flush_req);
}
} else {
for (auto& iter : flush_req) {
ColumnFamilyData* cfd = iter.first;
cfd->Ref();
cfd->SetFlushReason(flush_reason);
}
++unscheduled_flushes_;
flush_queue_.push_back(flush_req);
}
}
void DBImpl::SchedulePendingCompaction(ColumnFamilyData* cfd) {
mutex_.AssertHeld();
if (!cfd->queued_for_compaction() && cfd->NeedsCompaction()) {
AddToCompactionQueue(cfd);
++unscheduled_compactions_;
}
}
void DBImpl::SchedulePendingPurge(std::string fname, std::string dir_to_sync,
FileType type, uint64_t number, int job_id) {
mutex_.AssertHeld();
PurgeFileInfo file_info(fname, dir_to_sync, type, number, job_id);
purge_files_.insert({{number, std::move(file_info)}});
}
void DBImpl::BGWorkFlush(void* arg) {
FlushThreadArg fta = *(reinterpret_cast<FlushThreadArg*>(arg));
delete reinterpret_cast<FlushThreadArg*>(arg);
IOSTATS_SET_THREAD_POOL_ID(fta.thread_pri_);
TEST_SYNC_POINT("DBImpl::BGWorkFlush");
static_cast_with_check<DBImpl>(fta.db_)->BackgroundCallFlush(fta.thread_pri_);
TEST_SYNC_POINT("DBImpl::BGWorkFlush:done");
}
void DBImpl::BGWorkCompaction(void* arg) {
CompactionArg ca = *(reinterpret_cast<CompactionArg*>(arg));
delete reinterpret_cast<CompactionArg*>(arg);
IOSTATS_SET_THREAD_POOL_ID(Env::Priority::LOW);
TEST_SYNC_POINT("DBImpl::BGWorkCompaction");
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
auto prepicked_compaction =
static_cast<PrepickedCompaction*>(ca.prepicked_compaction);
static_cast_with_check<DBImpl>(ca.db)->BackgroundCallCompaction(
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
prepicked_compaction, Env::Priority::LOW);
delete prepicked_compaction;
}
void DBImpl::BGWorkBottomCompaction(void* arg) {
CompactionArg ca = *(static_cast<CompactionArg*>(arg));
delete static_cast<CompactionArg*>(arg);
IOSTATS_SET_THREAD_POOL_ID(Env::Priority::BOTTOM);
TEST_SYNC_POINT("DBImpl::BGWorkBottomCompaction");
auto* prepicked_compaction = ca.prepicked_compaction;
assert(prepicked_compaction && prepicked_compaction->compaction);
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
ca.db->BackgroundCallCompaction(prepicked_compaction, Env::Priority::BOTTOM);
delete prepicked_compaction;
}
void DBImpl::BGWorkPurge(void* db) {
IOSTATS_SET_THREAD_POOL_ID(Env::Priority::HIGH);
TEST_SYNC_POINT("DBImpl::BGWorkPurge:start");
reinterpret_cast<DBImpl*>(db)->BackgroundCallPurge();
TEST_SYNC_POINT("DBImpl::BGWorkPurge:end");
}
void DBImpl::UnscheduleCompactionCallback(void* arg) {
CompactionArg* ca_ptr = reinterpret_cast<CompactionArg*>(arg);
Env::Priority compaction_pri = ca_ptr->compaction_pri_;
if (Env::Priority::BOTTOM == compaction_pri) {
// Decrement bg_bottom_compaction_scheduled_ if priority is BOTTOM
ca_ptr->db->bg_bottom_compaction_scheduled_--;
} else if (Env::Priority::LOW == compaction_pri) {
// Decrement bg_compaction_scheduled_ if priority is LOW
ca_ptr->db->bg_compaction_scheduled_--;
}
CompactionArg ca = *(ca_ptr);
delete reinterpret_cast<CompactionArg*>(arg);
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
if (ca.prepicked_compaction != nullptr) {
// if it's a manual compaction, set status to ManualCompactionPaused
if (ca.prepicked_compaction->manual_compaction_state) {
ca.prepicked_compaction->manual_compaction_state->done = true;
ca.prepicked_compaction->manual_compaction_state->status =
Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
if (ca.prepicked_compaction->compaction != nullptr) {
ca.prepicked_compaction->compaction->ReleaseCompactionFiles(
Status::Incomplete(Status::SubCode::kManualCompactionPaused));
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
delete ca.prepicked_compaction->compaction;
}
delete ca.prepicked_compaction;
}
TEST_SYNC_POINT("DBImpl::UnscheduleCompactionCallback");
}
void DBImpl::UnscheduleFlushCallback(void* arg) {
// Decrement bg_flush_scheduled_ in flush callback
reinterpret_cast<FlushThreadArg*>(arg)->db_->bg_flush_scheduled_--;
Env::Priority flush_pri = reinterpret_cast<FlushThreadArg*>(arg)->thread_pri_;
if (Env::Priority::LOW == flush_pri) {
TEST_SYNC_POINT("DBImpl::UnscheduleLowFlushCallback");
} else if (Env::Priority::HIGH == flush_pri) {
TEST_SYNC_POINT("DBImpl::UnscheduleHighFlushCallback");
}
delete reinterpret_cast<FlushThreadArg*>(arg);
TEST_SYNC_POINT("DBImpl::UnscheduleFlushCallback");
}
Status DBImpl::BackgroundFlush(bool* made_progress, JobContext* job_context,
LogBuffer* log_buffer, FlushReason* reason,
Env::Priority thread_pri) {
mutex_.AssertHeld();
Status status;
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
*reason = FlushReason::kOthers;
// If BG work is stopped due to an error, but a recovery is in progress,
// that means this flush is part of the recovery. So allow it to go through
if (!error_handler_.IsBGWorkStopped()) {
if (shutting_down_.load(std::memory_order_acquire)) {
status = Status::ShutdownInProgress();
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
} else if (!error_handler_.IsRecoveryInProgress()) {
status = error_handler_.GetBGError();
}
if (!status.ok()) {
return status;
}
autovector<BGFlushArg> bg_flush_args;
std::vector<SuperVersionContext>& superversion_contexts =
job_context->superversion_contexts;
autovector<ColumnFamilyData*> column_families_not_to_flush;
while (!flush_queue_.empty()) {
// This cfd is already referenced
const FlushRequest& flush_req = PopFirstFromFlushQueue();
superversion_contexts.clear();
superversion_contexts.reserve(flush_req.size());
for (const auto& iter : flush_req) {
ColumnFamilyData* cfd = iter.first;
Dynamically changeable `MemPurge` option (#10011) Summary: **Summary** Make the mempurge option flag a Mutable Column Family option flag. Therefore, the mempurge feature can be dynamically toggled. **Motivation** RocksDB users prefer having the ability to switch features on and off without having to close and reopen the DB. This is particularly important if the feature causes issues and needs to be turned off. Dynamically changing a DB option flag does not seem currently possible. Moreover, with this new change, the MemPurge feature can be toggled on or off independently between column families, which we see as a major improvement. **Content of this PR** This PR includes removal of the `experimental_mempurge_threshold` flag as a DB option flag, and its re-introduction as a `MutableCFOption` flag. I updated the code to handle dynamic changes of the flag (in particular inside the `FlushJob` file). Additionally, this PR includes a new test to demonstrate the capacity of the code to toggle the MemPurge feature on and off, as well as the addition in the `db_stress` module of 2 different mempurge threshold values (0.0 and 1.0) that can be randomly changed with the `set_option_one_in` flag. This is useful to stress test the dynamic changes. **Benchmarking** I will add numbers to prove that there is no performance impact within the next 12 hours. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10011 Reviewed By: pdillinger Differential Revision: D36462357 Pulled By: bjlemaire fbshipit-source-id: 5e3d63bdadf085c0572ecc2349e7dd9729ce1802
2 years ago
if (cfd->GetMempurgeUsed()) {
// If imm() contains silent memtables (e.g.: because
// MemPurge was activated), requesting a flush will
// mark the imm_needed as true.
cfd->imm()->FlushRequested();
}
Dynamically changeable `MemPurge` option (#10011) Summary: **Summary** Make the mempurge option flag a Mutable Column Family option flag. Therefore, the mempurge feature can be dynamically toggled. **Motivation** RocksDB users prefer having the ability to switch features on and off without having to close and reopen the DB. This is particularly important if the feature causes issues and needs to be turned off. Dynamically changing a DB option flag does not seem currently possible. Moreover, with this new change, the MemPurge feature can be toggled on or off independently between column families, which we see as a major improvement. **Content of this PR** This PR includes removal of the `experimental_mempurge_threshold` flag as a DB option flag, and its re-introduction as a `MutableCFOption` flag. I updated the code to handle dynamic changes of the flag (in particular inside the `FlushJob` file). Additionally, this PR includes a new test to demonstrate the capacity of the code to toggle the MemPurge feature on and off, as well as the addition in the `db_stress` module of 2 different mempurge threshold values (0.0 and 1.0) that can be randomly changed with the `set_option_one_in` flag. This is useful to stress test the dynamic changes. **Benchmarking** I will add numbers to prove that there is no performance impact within the next 12 hours. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10011 Reviewed By: pdillinger Differential Revision: D36462357 Pulled By: bjlemaire fbshipit-source-id: 5e3d63bdadf085c0572ecc2349e7dd9729ce1802
2 years ago
if (cfd->IsDropped() || !cfd->imm()->IsFlushPending()) {
// can't flush this CF, try next one
column_families_not_to_flush.push_back(cfd);
continue;
}
superversion_contexts.emplace_back(SuperVersionContext(true));
bg_flush_args.emplace_back(cfd, iter.second,
&(superversion_contexts.back()));
}
if (!bg_flush_args.empty()) {
break;
}
}
if (!bg_flush_args.empty()) {
auto bg_job_limits = GetBGJobLimits();
for (const auto& arg : bg_flush_args) {
ColumnFamilyData* cfd = arg.cfd_;
ROCKS_LOG_BUFFER(
log_buffer,
"Calling FlushMemTableToOutputFile with column "
"family [%s], flush slots available %d, compaction slots available "
"%d, "
"flush slots scheduled %d, compaction slots scheduled %d",
cfd->GetName().c_str(), bg_job_limits.max_flushes,
bg_job_limits.max_compactions, bg_flush_scheduled_,
bg_compaction_scheduled_);
}
status = FlushMemTablesToOutputFiles(bg_flush_args, made_progress,
job_context, log_buffer, thread_pri);
TEST_SYNC_POINT("DBImpl::BackgroundFlush:BeforeFlush");
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// All the CFDs in the FlushReq must have the same flush reason, so just
// grab the first one
*reason = bg_flush_args[0].cfd_->GetFlushReason();
for (auto& arg : bg_flush_args) {
ColumnFamilyData* cfd = arg.cfd_;
if (cfd->UnrefAndTryDelete()) {
arg.cfd_ = nullptr;
}
}
}
for (auto cfd : column_families_not_to_flush) {
cfd->UnrefAndTryDelete();
}
return status;
}
void DBImpl::BackgroundCallFlush(Env::Priority thread_pri) {
bool made_progress = false;
JobContext job_context(next_job_id_.fetch_add(1), true);
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCallFlush:start", nullptr);
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
immutable_db_options_.info_log.get());
TEST_SYNC_POINT("DBImpl::BackgroundCallFlush:Start:1");
TEST_SYNC_POINT("DBImpl::BackgroundCallFlush:Start:2");
{
InstrumentedMutexLock l(&mutex_);
assert(bg_flush_scheduled_);
num_running_flushes_++;
std::unique_ptr<std::list<uint64_t>::iterator>
pending_outputs_inserted_elem(new std::list<uint64_t>::iterator(
CaptureCurrentFileNumberInPendingOutputs()));
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
FlushReason reason;
Status s = BackgroundFlush(&made_progress, &job_context, &log_buffer,
&reason, thread_pri);
if (!s.ok() && !s.IsShutdownInProgress() && !s.IsColumnFamilyDropped() &&
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
reason != FlushReason::kErrorRecovery) {
// Wait a little bit before retrying background flush in
// case this is an environmental problem and we do not want to
// chew up resources for failed flushes for the duration of
// the problem.
uint64_t error_cnt =
default_cf_internal_stats_->BumpAndGetBackgroundErrorCount();
bg_cv_.SignalAll(); // In case a waiter can proceed despite the error
mutex_.Unlock();
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Waiting after background flush error: %s"
"Accumulated background error counts: %" PRIu64,
s.ToString().c_str(), error_cnt);
log_buffer.FlushBufferToLog();
LogFlush(immutable_db_options_.info_log);
immutable_db_options_.clock->SleepForMicroseconds(1000000);
mutex_.Lock();
}
TEST_SYNC_POINT("DBImpl::BackgroundCallFlush:FlushFinish:0");
ReleaseFileNumberFromPendingOutputs(pending_outputs_inserted_elem);
// If flush failed, we want to delete all temporary files that we might have
// created. Thus, we force full scan in FindObsoleteFiles()
FindObsoleteFiles(&job_context, !s.ok() && !s.IsShutdownInProgress() &&
!s.IsColumnFamilyDropped());
// delete unnecessary files if any, this is done outside the mutex
if (job_context.HaveSomethingToClean() ||
job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
mutex_.Unlock();
TEST_SYNC_POINT("DBImpl::BackgroundCallFlush:FilesFound");
// Have to flush the info logs before bg_flush_scheduled_--
// because if bg_flush_scheduled_ becomes 0 and the lock is
// released, the deconstructor of DB can kick in and destroy all the
// states of DB so info_log might not be available after that point.
// It also applies to access other states that DB owns.
log_buffer.FlushBufferToLog();
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
}
job_context.Clean();
mutex_.Lock();
}
TEST_SYNC_POINT("DBImpl::BackgroundCallFlush:ContextCleanedUp");
assert(num_running_flushes_ > 0);
num_running_flushes_--;
bg_flush_scheduled_--;
// See if there's more work to be done
MaybeScheduleFlushOrCompaction();
atomic_flush_install_cv_.SignalAll();
bg_cv_.SignalAll();
// IMPORTANT: there should be no code after calling SignalAll. This call may
// signal the DB destructor that it's OK to proceed with destruction. In
// that case, all DB variables will be dealloacated and referencing them
// will cause trouble.
}
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
Env::Priority bg_thread_pri) {
bool made_progress = false;
JobContext job_context(next_job_id_.fetch_add(1), true);
TEST_SYNC_POINT("BackgroundCallCompaction:0");
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
immutable_db_options_.info_log.get());
{
InstrumentedMutexLock l(&mutex_);
// This call will unlock/lock the mutex to wait for current running
// IngestExternalFile() calls to finish.
WaitForIngestFile();
num_running_compactions_++;
std::unique_ptr<std::list<uint64_t>::iterator>
pending_outputs_inserted_elem(new std::list<uint64_t>::iterator(
CaptureCurrentFileNumberInPendingOutputs()));
assert((bg_thread_pri == Env::Priority::BOTTOM &&
bg_bottom_compaction_scheduled_) ||
(bg_thread_pri == Env::Priority::LOW && bg_compaction_scheduled_));
Status s = BackgroundCompaction(&made_progress, &job_context, &log_buffer,
prepicked_compaction, bg_thread_pri);
TEST_SYNC_POINT("BackgroundCallCompaction:1");
if (s.IsBusy()) {
bg_cv_.SignalAll(); // In case a waiter can proceed despite the error
mutex_.Unlock();
immutable_db_options_.clock->SleepForMicroseconds(
10000); // prevent hot loop
mutex_.Lock();
} else if (!s.ok() && !s.IsShutdownInProgress() &&
!s.IsManualCompactionPaused() && !s.IsColumnFamilyDropped()) {
// Wait a little bit before retrying background compaction in
// case this is an environmental problem and we do not want to
// chew up resources for failed compactions for the duration of
// the problem.
uint64_t error_cnt =
default_cf_internal_stats_->BumpAndGetBackgroundErrorCount();
bg_cv_.SignalAll(); // In case a waiter can proceed despite the error
mutex_.Unlock();
log_buffer.FlushBufferToLog();
ROCKS_LOG_ERROR(immutable_db_options_.info_log,
"Waiting after background compaction error: %s, "
"Accumulated background error counts: %" PRIu64,
s.ToString().c_str(), error_cnt);
LogFlush(immutable_db_options_.info_log);
immutable_db_options_.clock->SleepForMicroseconds(1000000);
mutex_.Lock();
} else if (s.IsManualCompactionPaused()) {
assert(prepicked_compaction);
ManualCompactionState* m = prepicked_compaction->manual_compaction_state;
assert(m);
ROCKS_LOG_BUFFER(&log_buffer, "[%s] [JOB %d] Manual compaction paused",
m->cfd->GetName().c_str(), job_context.job_id);
}
ReleaseFileNumberFromPendingOutputs(pending_outputs_inserted_elem);
// If compaction failed, we want to delete all temporary files that we
// might have created (they might not be all recorded in job_context in
// case of a failure). Thus, we force full scan in FindObsoleteFiles()
FindObsoleteFiles(&job_context, !s.ok() && !s.IsShutdownInProgress() &&
!s.IsManualCompactionPaused() &&
!s.IsColumnFamilyDropped() &&
!s.IsBusy());
TEST_SYNC_POINT("DBImpl::BackgroundCallCompaction:FoundObsoleteFiles");
// delete unnecessary files if any, this is done outside the mutex
if (job_context.HaveSomethingToClean() ||
job_context.HaveSomethingToDelete() || !log_buffer.IsEmpty()) {
mutex_.Unlock();
// Have to flush the info logs before bg_compaction_scheduled_--
// because if bg_flush_scheduled_ becomes 0 and the lock is
// released, the deconstructor of DB can kick in and destroy all the
// states of DB so info_log might not be available after that point.
// It also applies to access other states that DB owns.
log_buffer.FlushBufferToLog();
if (job_context.HaveSomethingToDelete()) {
PurgeObsoleteFiles(job_context);
TEST_SYNC_POINT("DBImpl::BackgroundCallCompaction:PurgedObsoleteFiles");
}
job_context.Clean();
mutex_.Lock();
}
assert(num_running_compactions_ > 0);
num_running_compactions_--;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
if (bg_thread_pri == Env::Priority::LOW) {
bg_compaction_scheduled_--;
} else {
assert(bg_thread_pri == Env::Priority::BOTTOM);
bg_bottom_compaction_scheduled_--;
}
// See if there's more work to be done
MaybeScheduleFlushOrCompaction();
Fix ConcurrentTaskLimiter token release for shutdown (#8253) Summary: Previously the shutdown process did not properly wait for all `compaction_thread_limiter` tokens to be released before proceeding to delete the DB's C++ objects. When this happened, we saw tests like "DBCompactionTest.CompactionLimiter" flake with the following error: ``` virtual rocksdb::ConcurrentTaskLimiterImpl::~ConcurrentTaskLimiterImpl(): Assertion `outstanding_tasks_ == 0' failed. ``` There is a case where a token can still be alive even after the shutdown process has waited for BG work to complete. In particular, this happens because the shutdown process only waits for flush/compaction scheduled/unscheduled counters to all reach zero. These counters are decremented in `BackgroundCallCompaction()` functions. However, tokens are released in `BGWork*Compaction()` functions, which actually wrap the `BackgroundCallCompaction()` function. A simple sleep could repro the race condition: ``` $ diff --git a/db/db_impl/db_impl_compaction_flush.cc b/db/db_impl/db_impl_compaction_flush.cc index 806bc548a..ba59efa89 100644 --- a/db/db_impl/db_impl_compaction_flush.cc +++ b/db/db_impl/db_impl_compaction_flush.cc @@ -2442,6 +2442,7 @@ void DBImpl::BGWorkCompaction(void* arg) { static_cast<PrepickedCompaction*>(ca.prepicked_compaction); static_cast_with_check<DBImpl>(ca.db)->BackgroundCallCompaction( prepicked_compaction, Env::Priority::LOW); + sleep(1); delete prepicked_compaction; } $ ./db_compaction_test --gtest_filter=DBCompactionTest.CompactionLimiter db_compaction_test: util/concurrent_task_limiter_impl.cc:24: virtual rocksdb::ConcurrentTaskLimiterImpl::~ConcurrentTaskLimiterImpl(): Assertion `outstanding_tasks_ == 0' failed. Received signal 6 (Aborted) #0 /usr/local/fbcode/platform007/lib/libc.so.6(gsignal+0xcf) [0x7f02673c30ff] ?? ??:0 https://github.com/facebook/rocksdb/issues/1 /usr/local/fbcode/platform007/lib/libc.so.6(abort+0x134) [0x7f02673ac934] ?? ??:0 ... ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8253 Test Plan: sleeps to expose race conditions Reviewed By: akankshamahajan15 Differential Revision: D28168064 Pulled By: ajkr fbshipit-source-id: 9e5167c74398d323e7975980c5cc00f450631160
4 years ago
if (prepicked_compaction != nullptr &&
prepicked_compaction->task_token != nullptr) {
// Releasing task tokens affects (and asserts on) the DB state, so
// must be done before we potentially signal the DB close process to
// proceed below.
prepicked_compaction->task_token.reset();
Fix ConcurrentTaskLimiter token release for shutdown (#8253) Summary: Previously the shutdown process did not properly wait for all `compaction_thread_limiter` tokens to be released before proceeding to delete the DB's C++ objects. When this happened, we saw tests like "DBCompactionTest.CompactionLimiter" flake with the following error: ``` virtual rocksdb::ConcurrentTaskLimiterImpl::~ConcurrentTaskLimiterImpl(): Assertion `outstanding_tasks_ == 0' failed. ``` There is a case where a token can still be alive even after the shutdown process has waited for BG work to complete. In particular, this happens because the shutdown process only waits for flush/compaction scheduled/unscheduled counters to all reach zero. These counters are decremented in `BackgroundCallCompaction()` functions. However, tokens are released in `BGWork*Compaction()` functions, which actually wrap the `BackgroundCallCompaction()` function. A simple sleep could repro the race condition: ``` $ diff --git a/db/db_impl/db_impl_compaction_flush.cc b/db/db_impl/db_impl_compaction_flush.cc index 806bc548a..ba59efa89 100644 --- a/db/db_impl/db_impl_compaction_flush.cc +++ b/db/db_impl/db_impl_compaction_flush.cc @@ -2442,6 +2442,7 @@ void DBImpl::BGWorkCompaction(void* arg) { static_cast<PrepickedCompaction*>(ca.prepicked_compaction); static_cast_with_check<DBImpl>(ca.db)->BackgroundCallCompaction( prepicked_compaction, Env::Priority::LOW); + sleep(1); delete prepicked_compaction; } $ ./db_compaction_test --gtest_filter=DBCompactionTest.CompactionLimiter db_compaction_test: util/concurrent_task_limiter_impl.cc:24: virtual rocksdb::ConcurrentTaskLimiterImpl::~ConcurrentTaskLimiterImpl(): Assertion `outstanding_tasks_ == 0' failed. Received signal 6 (Aborted) #0 /usr/local/fbcode/platform007/lib/libc.so.6(gsignal+0xcf) [0x7f02673c30ff] ?? ??:0 https://github.com/facebook/rocksdb/issues/1 /usr/local/fbcode/platform007/lib/libc.so.6(abort+0x134) [0x7f02673ac934] ?? ??:0 ... ``` Pull Request resolved: https://github.com/facebook/rocksdb/pull/8253 Test Plan: sleeps to expose race conditions Reviewed By: akankshamahajan15 Differential Revision: D28168064 Pulled By: ajkr fbshipit-source-id: 9e5167c74398d323e7975980c5cc00f450631160
4 years ago
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
if (made_progress ||
(bg_compaction_scheduled_ == 0 &&
bg_bottom_compaction_scheduled_ == 0) ||
HasPendingManualCompaction() || unscheduled_compactions_ == 0) {
// signal if
// * made_progress -- need to wakeup DelayWrite
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
// * bg_{bottom,}_compaction_scheduled_ == 0 -- need to wakeup ~DBImpl
// * HasPendingManualCompaction -- need to wakeup RunManualCompaction
// If none of this is true, there is no need to signal since nobody is
// waiting for it
bg_cv_.SignalAll();
}
// IMPORTANT: there should be no code after calling SignalAll. This call may
// signal the DB destructor that it's OK to proceed with destruction. In
// that case, all DB variables will be dealloacated and referencing them
// will cause trouble.
}
}
Status DBImpl::BackgroundCompaction(bool* made_progress,
JobContext* job_context,
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
LogBuffer* log_buffer,
PrepickedCompaction* prepicked_compaction,
Env::Priority thread_pri) {
ManualCompactionState* manual_compaction =
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
prepicked_compaction == nullptr
? nullptr
: prepicked_compaction->manual_compaction_state;
*made_progress = false;
mutex_.AssertHeld();
TEST_SYNC_POINT("DBImpl::BackgroundCompaction:Start");
bool is_manual = (manual_compaction != nullptr);
std::unique_ptr<Compaction> c;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
if (prepicked_compaction != nullptr &&
prepicked_compaction->compaction != nullptr) {
c.reset(prepicked_compaction->compaction);
}
bool is_prepicked = is_manual || c;
// (manual_compaction->in_progress == false);
bool trivial_move_disallowed =
is_manual && manual_compaction->disallow_trivial_move;
CompactionJobStats compaction_job_stats;
Status status;
if (!error_handler_.IsBGWorkStopped()) {
if (shutting_down_.load(std::memory_order_acquire)) {
status = Status::ShutdownInProgress();
} else if (is_manual &&
manual_compaction->canceled.load(std::memory_order_acquire)) {
status = Status::Incomplete(Status::SubCode::kManualCompactionPaused);
}
} else {
status = error_handler_.GetBGError();
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// If we get here, it means a hard error happened after this compaction
// was scheduled by MaybeScheduleFlushOrCompaction(), but before it got
// a chance to execute. Since we didn't pop a cfd from the compaction
// queue, increment unscheduled_compactions_
unscheduled_compactions_++;
}
if (!status.ok()) {
if (is_manual) {
manual_compaction->status = status;
manual_compaction->done = true;
manual_compaction->in_progress = false;
manual_compaction = nullptr;
}
if (c) {
c->ReleaseCompactionFiles(status);
c.reset();
}
return status;
}
if (is_manual) {
// another thread cannot pick up the same work
manual_compaction->in_progress = true;
}
TEST_SYNC_POINT("DBImpl::BackgroundCompaction:InProgress");
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
std::unique_ptr<TaskLimiterToken> task_token;
// InternalKey manual_end_storage;
// InternalKey* manual_end = &manual_end_storage;
bool sfm_reserved_compact_space = false;
if (is_manual) {
ManualCompactionState* m = manual_compaction;
assert(m->in_progress);
if (!c) {
m->done = true;
m->manual_end = nullptr;
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Manual compaction from level-%d from %s .. "
"%s; nothing to do\n",
m->cfd->GetName().c_str(), m->input_level,
(m->begin ? m->begin->DebugString(true).c_str() : "(begin)"),
(m->end ? m->end->DebugString(true).c_str() : "(end)"));
} else {
// First check if we have enough room to do the compaction
bool enough_room = EnoughRoomForCompaction(
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
m->cfd, *(c->inputs()), &sfm_reserved_compact_space, log_buffer);
if (!enough_room) {
// Then don't do the compaction
c->ReleaseCompactionFiles(status);
c.reset();
// m's vars will get set properly at the end of this function,
// as long as status == CompactionTooLarge
status = Status::CompactionTooLarge();
} else {
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Manual compaction from level-%d to level-%d from %s .. "
"%s; will stop at %s\n",
m->cfd->GetName().c_str(), m->input_level, c->output_level(),
(m->begin ? m->begin->DebugString(true).c_str() : "(begin)"),
(m->end ? m->end->DebugString(true).c_str() : "(end)"),
((m->done || m->manual_end == nullptr)
? "(end)"
: m->manual_end->DebugString(true).c_str()));
}
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
} else if (!is_prepicked && !compaction_queue_.empty()) {
if (HasExclusiveManualCompaction()) {
// Can't compact right now, but try again later
TEST_SYNC_POINT("DBImpl::BackgroundCompaction()::Conflict");
// Stay in the compaction queue.
unscheduled_compactions_++;
return Status::OK();
}
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
auto cfd = PickCompactionFromQueue(&task_token, log_buffer);
if (cfd == nullptr) {
// Can't find any executable task from the compaction queue.
// All tasks have been throttled by compaction thread limiter.
++unscheduled_compactions_;
return Status::Busy();
}
// We unreference here because the following code will take a Ref() on
// this cfd if it is going to use it (Compaction class holds a
// reference).
// This will all happen under a mutex so we don't have to be afraid of
// somebody else deleting it.
if (cfd->UnrefAndTryDelete()) {
// This was the last reference of the column family, so no need to
// compact.
return Status::OK();
}
// Pick up latest mutable CF Options and use it throughout the
// compaction job
// Compaction makes a copy of the latest MutableCFOptions. It should be used
// throughout the compaction procedure to make sure consistency. It will
// eventually be installed into SuperVersion
auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
if (!mutable_cf_options->disable_auto_compactions && !cfd->IsDropped()) {
// NOTE: try to avoid unnecessary copy of MutableCFOptions if
// compaction is not necessary. Need to make sure mutex is held
// until we make a copy in the following code
TEST_SYNC_POINT("DBImpl::BackgroundCompaction():BeforePickCompaction");
c.reset(cfd->PickCompaction(*mutable_cf_options, mutable_db_options_,
log_buffer));
TEST_SYNC_POINT("DBImpl::BackgroundCompaction():AfterPickCompaction");
if (c != nullptr) {
bool enough_room = EnoughRoomForCompaction(
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
cfd, *(c->inputs()), &sfm_reserved_compact_space, log_buffer);
if (!enough_room) {
// Then don't do the compaction
c->ReleaseCompactionFiles(status);
c->column_family_data()
->current()
->storage_info()
->ComputeCompactionScore(*(c->immutable_options()),
*(c->mutable_cf_options()));
AddToCompactionQueue(cfd);
++unscheduled_compactions_;
c.reset();
// Don't need to sleep here, because BackgroundCallCompaction
// will sleep if !s.ok()
status = Status::CompactionTooLarge();
} else {
// update statistics
size_t num_files = 0;
for (auto& each_level : *c->inputs()) {
num_files += each_level.files.size();
}
RecordInHistogram(stats_, NUM_FILES_IN_SINGLE_COMPACTION, num_files);
// There are three things that can change compaction score:
// 1) When flush or compaction finish. This case is covered by
// InstallSuperVersionAndScheduleWork
// 2) When MutableCFOptions changes. This case is also covered by
// InstallSuperVersionAndScheduleWork, because this is when the new
// options take effect.
// 3) When we Pick a new compaction, we "remove" those files being
// compacted from the calculation, which then influences compaction
// score. Here we check if we need the new compaction even without the
// files that are currently being compacted. If we need another
// compaction, we might be able to execute it in parallel, so we add
// it to the queue and schedule a new thread.
if (cfd->NeedsCompaction()) {
// Yes, we need more compactions!
AddToCompactionQueue(cfd);
++unscheduled_compactions_;
MaybeScheduleFlushOrCompaction();
}
}
}
}
}
IOStatus io_s;
if (!c) {
// Nothing to do
ROCKS_LOG_BUFFER(log_buffer, "Compaction nothing to do");
} else if (c->deletion_compaction()) {
// TODO(icanadi) Do we want to honor snapshots here? i.e. not delete old
// file if there is alive snapshot pointing to it
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
c->column_family_data());
assert(c->num_input_files(1) == 0);
assert(c->level() == 0);
assert(c->column_family_data()->ioptions()->compaction_style ==
kCompactionStyleFIFO);
compaction_job_stats.num_input_files = c->num_input_files(0);
NotifyOnCompactionBegin(c->column_family_data(), c.get(), status,
compaction_job_stats, job_context->job_id);
for (const auto& f : *c->inputs(0)) {
c->edit()->DeleteFile(c->level(), f->fd.GetNumber());
}
status = versions_->LogAndApply(c->column_family_data(),
*c->mutable_cf_options(), c->edit(),
&mutex_, directories_.GetDbDir());
io_s = versions_->io_status();
InstallSuperVersionAndScheduleWork(c->column_family_data(),
&job_context->superversion_contexts[0],
*c->mutable_cf_options());
ROCKS_LOG_BUFFER(log_buffer, "[%s] Deleted %d files\n",
c->column_family_data()->GetName().c_str(),
c->num_input_files(0));
*made_progress = true;
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
c->column_family_data());
} else if (!trivial_move_disallowed && c->IsTrivialMove()) {
TEST_SYNC_POINT("DBImpl::BackgroundCompaction:TrivialMove");
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
c->column_family_data());
// Instrument for event update
// TODO(yhchiang): add op details for showing trivial-move.
ThreadStatusUtil::SetColumnFamily(
c->column_family_data(), c->column_family_data()->ioptions()->env,
immutable_db_options_.enable_thread_tracking);
ThreadStatusUtil::SetThreadOperation(ThreadStatus::OP_COMPACTION);
compaction_job_stats.num_input_files = c->num_input_files(0);
NotifyOnCompactionBegin(c->column_family_data(), c.get(), status,
compaction_job_stats, job_context->job_id);
// Move files to next level
int32_t moved_files = 0;
int64_t moved_bytes = 0;
for (unsigned int l = 0; l < c->num_input_levels(); l++) {
if (c->level(l) == c->output_level()) {
continue;
}
for (size_t i = 0; i < c->num_input_files(l); i++) {
FileMetaData* f = c->input(l, i);
c->edit()->DeleteFile(c->level(l), f->fd.GetNumber());
c->edit()->AddFile(
c->output_level(), f->fd.GetNumber(), f->fd.GetPathId(),
f->fd.GetFileSize(), f->smallest, f->largest, f->fd.smallest_seqno,
f->fd.largest_seqno, f->marked_for_compaction, f->temperature,
f->oldest_blob_file_number, f->oldest_ancester_time,
f->file_creation_time, f->file_checksum, f->file_checksum_func_name,
f->min_timestamp, f->max_timestamp, f->unique_id);
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Moving #%" PRIu64 " to level-%d %" PRIu64 " bytes\n",
c->column_family_data()->GetName().c_str(), f->fd.GetNumber(),
c->output_level(), f->fd.GetFileSize());
++moved_files;
moved_bytes += f->fd.GetFileSize();
}
}
Add basic kRoundRobin compaction policy (#10107) Summary: Add `kRoundRobin` as a compaction priority. The implementation is as follows. - Define a cursor as the smallest Internal key in the successor of the selected file. Add `vector<InternalKey> compact_cursor_` into `VersionStorageInfo` where each element (`InternalKey`) in `compact_cursor_` represents a cursor. In round-robin compaction policy, we just need to select the first file (assuming files are sorted) and also has the smallest InternalKey larger than/equal to the cursor. After a file is chosen, we create a new `Fsize` vector which puts the selected file is placed at the first position in `temp`, the next cursor is then updated as the smallest InternalKey in successor of the selected file (the above logic is implemented in `SortFileByRoundRobin`). - After a compaction succeeds, typically `InstallCompactionResults()`, we choose the next cursor for the input level and save it to `edit`. When calling `LogAndApply`, we save the next cursor with its level into some local variable and finally apply the change to `vstorage` in `SaveTo` function. - Cursors are persist pair by pair (<level, InternalKey>) in `EncodeTo` so that they can be reconstructed when reopening. An empty cursor will not be encoded to MANIFEST Pull Request resolved: https://github.com/facebook/rocksdb/pull/10107 Test Plan: add unit test (`CompactionPriRoundRobin`) in `compaction_picker_test`, add `kRoundRobin` priority in `CompactionPriTest` from `db_compaction_test`, and add `PersistRoundRobinCompactCursor` in `db_compaction_test` Reviewed By: ajkr Differential Revision: D37316037 Pulled By: littlepig2013 fbshipit-source-id: 9f481748190ace416079139044e00df2968fb1ee
2 years ago
if (c->compaction_reason() == CompactionReason::kLevelMaxLevelSize &&
c->immutable_options()->compaction_pri == kRoundRobin) {
int start_level = c->start_level();
if (start_level > 0) {
auto vstorage = c->input_version()->storage_info();
c->edit()->AddCompactCursor(
start_level, vstorage->GetNextCompactCursor(start_level));
}
}
status = versions_->LogAndApply(c->column_family_data(),
*c->mutable_cf_options(), c->edit(),
&mutex_, directories_.GetDbDir());
io_s = versions_->io_status();
// Use latest MutableCFOptions
InstallSuperVersionAndScheduleWork(c->column_family_data(),
&job_context->superversion_contexts[0],
*c->mutable_cf_options());
VersionStorageInfo::LevelSummaryStorage tmp;
c->column_family_data()->internal_stats()->IncBytesMoved(c->output_level(),
moved_bytes);
{
event_logger_.LogToBuffer(log_buffer)
<< "job" << job_context->job_id << "event"
<< "trivial_move"
<< "destination_level" << c->output_level() << "files" << moved_files
<< "total_files_size" << moved_bytes;
}
ROCKS_LOG_BUFFER(
log_buffer,
"[%s] Moved #%d files to level-%d %" PRIu64 " bytes %s: %s\n",
c->column_family_data()->GetName().c_str(), moved_files,
c->output_level(), moved_bytes, status.ToString().c_str(),
c->column_family_data()->current()->storage_info()->LevelSummary(&tmp));
*made_progress = true;
// Clear Instrument
ThreadStatusUtil::ResetThreadStatus();
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
c->column_family_data());
} else if (!is_prepicked && c->output_level() > 0 &&
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
c->output_level() ==
c->column_family_data()
->current()
->storage_info()
->MaxOutputLevel(
immutable_db_options_.allow_ingest_behind) &&
env_->GetBackgroundThreads(Env::Priority::BOTTOM) > 0) {
// Forward compactions involving last level to the bottom pool if it exists,
// such that compactions unlikely to contribute to write stalls can be
// delayed or deprioritized.
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
TEST_SYNC_POINT("DBImpl::BackgroundCompaction:ForwardToBottomPriPool");
CompactionArg* ca = new CompactionArg;
ca->db = this;
ca->compaction_pri_ = Env::Priority::BOTTOM;
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
ca->prepicked_compaction = new PrepickedCompaction;
ca->prepicked_compaction->compaction = c.release();
ca->prepicked_compaction->manual_compaction_state = nullptr;
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
// Transfer requested token, so it doesn't need to do it again.
ca->prepicked_compaction->task_token = std::move(task_token);
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
++bg_bottom_compaction_scheduled_;
env_->Schedule(&DBImpl::BGWorkBottomCompaction, ca, Env::Priority::BOTTOM,
this, &DBImpl::UnscheduleCompactionCallback);
} else {
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:BeforeCompaction",
c->column_family_data());
int output_level __attribute__((__unused__));
output_level = c->output_level();
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:NonTrivial",
&output_level);
std::vector<SequenceNumber> snapshot_seqs;
SequenceNumber earliest_write_conflict_snapshot;
SnapshotChecker* snapshot_checker;
GetSnapshotContext(job_context, &snapshot_seqs,
&earliest_write_conflict_snapshot, &snapshot_checker);
assert(is_snapshot_supported_ || snapshots_.empty());
CompactionJob compaction_job(
job_context->job_id, c.get(), immutable_db_options_,
mutable_db_options_, file_options_for_compaction_, versions_.get(),
&shutting_down_, log_buffer, directories_.GetDbDir(),
GetDataDir(c->column_family_data(), c->output_path_id()),
GetDataDir(c->column_family_data(), 0), stats_, &mutex_,
&error_handler_, snapshot_seqs, earliest_write_conflict_snapshot,
CompactionIterator sees consistent view of which keys are committed (#9830) Summary: **This PR does not affect the functionality of `DB` and write-committed transactions.** `CompactionIterator` uses `KeyCommitted(seq)` to determine if a key in the database is committed. As the name 'write-committed' implies, if write-committed policy is used, a key exists in the database only if it is committed. In fact, the implementation of `KeyCommitted()` is as follows: ``` inline bool KeyCommitted(SequenceNumber seq) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(seq, kMaxSequence) == SnapshotCheckerResult::kInSnapshot; } ``` With that being said, we focus on write-prepared/write-unprepared transactions. A few notes: - A key can exist in the db even if it's uncommitted. Therefore, we rely on `snapshot_checker_` to determine data visibility. We also require that all writes go through transaction API instead of the raw `WriteBatch` + `Write`, thus at most one uncommitted version of one user key can exist in the database. - `CompactionIterator` outputs a key as long as the key is uncommitted. Due to the above reasons, it is possible that `CompactionIterator` decides to output an uncommitted key without doing further checks on the key (`NextFromInput()`). By the time the key is being prepared for output, the key becomes committed because the `snapshot_checker_(seq, kMaxSequence)` becomes true in the implementation of `KeyCommitted()`. Then `CompactionIterator` will try to zero its sequence number and hit assertion error if the key is a tombstone. To fix this issue, we should make the `CompactionIterator` see a consistent view of the input keys. Note that for write-prepared/write-unprepared, the background flush/compaction jobs already take a "job snapshot" before starting processing keys. The job snapshot is released only after the entire flush/compaction finishes. We can use this snapshot to determine whether a key is committed or not with minor change to `KeyCommitted()`. ``` inline bool KeyCommitted(SequenceNumber sequence) { // For non-txn-db and write-committed, snapshot_checker_ is always nullptr. return snapshot_checker_ == nullptr || snapshot_checker_->CheckInSnapshot(sequence, job_snapshot_) == SnapshotCheckerResult::kInSnapshot; } ``` As a result, whether a key is committed or not will remain a constant throughout compaction, causing no trouble for `CompactionIterator`s assertions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9830 Test Plan: make check Reviewed By: ltamasi Differential Revision: D35561162 Pulled By: riversand963 fbshipit-source-id: 0e00d200c195240341cfe6d34cbc86798b315b9f
3 years ago
snapshot_checker, job_context, table_cache_, &event_logger_,
c->mutable_cf_options()->paranoid_file_checks,
c->mutable_cf_options()->report_bg_io_stats, dbname_,
&compaction_job_stats, thread_pri, io_tracer_,
is_manual ? manual_compaction->canceled
: kManualCompactionCanceledFalse_,
db_id_, db_session_id_, c->column_family_data()->GetFullHistoryTsLow(),
c->trim_ts(), &blob_callback_);
compaction_job.Prepare();
NotifyOnCompactionBegin(c->column_family_data(), c.get(), status,
compaction_job_stats, job_context->job_id);
mutex_.Unlock();
TEST_SYNC_POINT_CALLBACK(
"DBImpl::BackgroundCompaction:NonTrivial:BeforeRun", nullptr);
// Should handle erorr?
compaction_job.Run().PermitUncheckedError();
TEST_SYNC_POINT("DBImpl::BackgroundCompaction:NonTrivial:AfterRun");
mutex_.Lock();
status = compaction_job.Install(*c->mutable_cf_options());
io_s = compaction_job.io_status();
if (status.ok()) {
InstallSuperVersionAndScheduleWork(c->column_family_data(),
&job_context->superversion_contexts[0],
*c->mutable_cf_options());
}
*made_progress = true;
Concurrent task limiter for compaction thread control (#4332) Summary: The PR is targeting to resolve the issue of: https://github.com/facebook/rocksdb/issues/3972#issue-330771918 We have a rocksdb created with leveled-compaction with multiple column families (CFs), some of CFs are using HDD to store big and less frequently accessed data and others are using SSD. When there are continuously write traffics going on to all CFs, the compaction thread pool is mostly occupied by those slow HDD compactions, which blocks fully utilize SSD bandwidth. Since atomic write and transaction is needed across CFs, so splitting it to multiple rocksdb instance is not an option for us. With the compaction thread control, we got 30%+ HDD write throughput gain, and also a lot smooth SSD write since less write stall happening. ConcurrentTaskLimiter can be shared with multi-CFs across rocksdb instances, so the feature does not only work for multi-CFs scenarios, but also for multi-rocksdbs scenarios, who need disk IO resource control per tenant. The usage is straight forward: e.g.: // // Enable compaction thread limiter thru ColumnFamilyOptions // std::shared_ptr<ConcurrentTaskLimiter> ctl(NewConcurrentTaskLimiter("foo_limiter", 4)); Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = ctl; ... // // Compaction thread limiter can be tuned or disabled on-the-fly // ctl->SetMaxOutstandingTask(12); // enlarge to 12 tasks ... ctl->ResetMaxOutstandingTask(); // disable (bypass) thread limiter ctl->SetMaxOutstandingTask(-1); // Same as above ... ctl->SetMaxOutstandingTask(0); // full throttle (0 task) // // Sharing compaction thread limiter among CFs (to resolve multiple storage perf issue) // std::shared_ptr<ConcurrentTaskLimiter> ctl_ssd(NewConcurrentTaskLimiter("ssd_limiter", 8)); std::shared_ptr<ConcurrentTaskLimiter> ctl_hdd(NewConcurrentTaskLimiter("hdd_limiter", 4)); Options options; ColumnFamilyOptions cf_opt_ssd1(options); ColumnFamilyOptions cf_opt_ssd2(options); ColumnFamilyOptions cf_opt_hdd1(options); ColumnFamilyOptions cf_opt_hdd2(options); ColumnFamilyOptions cf_opt_hdd3(options); // SSD CFs cf_opt_ssd1.compaction_thread_limiter = ctl_ssd; cf_opt_ssd2.compaction_thread_limiter = ctl_ssd; // HDD CFs cf_opt_hdd1.compaction_thread_limiter = ctl_hdd; cf_opt_hdd2.compaction_thread_limiter = ctl_hdd; cf_opt_hdd3.compaction_thread_limiter = ctl_hdd; ... // // The limiter is disabled by default (or set to nullptr explicitly) // Options options; ColumnFamilyOptions cf_opt(options); cf_opt.compaction_thread_limiter = nullptr; Pull Request resolved: https://github.com/facebook/rocksdb/pull/4332 Differential Revision: D13226590 Pulled By: siying fbshipit-source-id: 14307aec55b8bd59c8223d04aa6db3c03d1b0c1d
6 years ago
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
c->column_family_data());
}
if (status.ok() && !io_s.ok()) {
status = io_s;
} else {
io_s.PermitUncheckedError();
}
if (c != nullptr) {
c->ReleaseCompactionFiles(status);
*made_progress = true;
#ifndef ROCKSDB_LITE
// Need to make sure SstFileManager does its bookkeeping
auto sfm = static_cast<SstFileManagerImpl*>(
immutable_db_options_.sst_file_manager.get());
if (sfm && sfm_reserved_compact_space) {
sfm->OnCompactionCompletion(c.get());
}
#endif // ROCKSDB_LITE
NotifyOnCompactionCompleted(c->column_family_data(), c.get(), status,
compaction_job_stats, job_context->job_id);
}
if (status.ok() || status.IsCompactionTooLarge() ||
status.IsManualCompactionPaused()) {
// Done
} else if (status.IsColumnFamilyDropped() || status.IsShutdownInProgress()) {
// Ignore compaction errors found during shutting down
} else {
ROCKS_LOG_WARN(immutable_db_options_.info_log, "Compaction error: %s",
status.ToString().c_str());
if (!io_s.ok()) {
First step towards handling MANIFEST write error (#6949) Summary: This PR provides preliminary support for handling IO error during MANIFEST write. File write/sync is not guaranteed to be atomic. If we encounter an IOError while writing/syncing to the MANIFEST file, we cannot be sure about the state of the MANIFEST file. The version edits may or may not have reached the file. During cleanup, if we delete the newly-generated SST files referenced by the pending version edit(s), but the version edit(s) actually are persistent in the MANIFEST, then next recovery attempt will process the version edits(s) and then fail since the SST files have already been deleted. One approach is to truncate the MANIFEST after write/sync error, so that it is safe to delete the SST files. However, file truncation may not be supported on certain file systems. Therefore, we take the following approach. If an IOError is detected during MANIFEST write/sync, we disable file deletions for the faulty database. Depending on whether the IOError is retryable (set by underlying file system), either RocksDB or application can call `DB::Resume()`, or simply shutdown and restart. During `Resume()`, RocksDB will try to switch to a new MANIFEST and write all existing in-memory version storage in the new file. If this succeeds, then RocksDB may proceed. If all recovery is completed, then file deletions will be re-enabled. Note that multiple threads can call `LogAndApply()` at the same time, though only one of them will be going through the process MANIFEST write, possibly batching the version edits of other threads. When the leading MANIFEST writer finishes, all of the MANIFEST writing threads in this batch will have the same IOError. They will all call `ErrorHandler::SetBGError()` in which file deletion will be disabled. Possible future directions: - Add an `ErrorContext` structure so that it is easier to pass more info to `ErrorHandler`. Currently, as in this example, a new `BackgroundErrorReason` has to be added. Test plan (dev server): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6949 Reviewed By: anand1976 Differential Revision: D22026020 Pulled By: riversand963 fbshipit-source-id: f3c68a2ef45d9b505d0d625c7c5e0c88495b91c8
4 years ago
// Error while writing to MANIFEST.
// In fact, versions_->io_status() can also be the result of renaming
// CURRENT file. With current code, it's just difficult to tell. So just
// be pessimistic and try write to a new MANIFEST.
// TODO: distinguish between MANIFEST write and CURRENT renaming
auto err_reason = versions_->io_status().ok()
? BackgroundErrorReason::kCompaction
: BackgroundErrorReason::kManifestWrite;
error_handler_.SetBGError(io_s, err_reason);
} else {
error_handler_.SetBGError(status, BackgroundErrorReason::kCompaction);
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
if (c != nullptr && !is_manual && !error_handler_.IsBGWorkStopped()) {
// Put this cfd back in the compaction queue so we can retry after some
// time
auto cfd = c->column_family_data();
assert(cfd != nullptr);
// Since this compaction failed, we need to recompute the score so it
// takes the original input files into account
c->column_family_data()
->current()
->storage_info()
->ComputeCompactionScore(*(c->immutable_options()),
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
*(c->mutable_cf_options()));
if (!cfd->queued_for_compaction()) {
AddToCompactionQueue(cfd);
++unscheduled_compactions_;
}
}
}
Auto recovery from out of space errors (#4164) Summary: This commit implements automatic recovery from a Status::NoSpace() error during background operations such as write callback, flush and compaction. The broad design is as follows - 1. Compaction errors are treated as soft errors and don't put the database in read-only mode. A compaction is delayed until enough free disk space is available to accomodate the compaction outputs, which is estimated based on the input size. This means that users can continue to write, and we rely on the WriteController to delay or stop writes if the compaction debt becomes too high due to persistent low disk space condition 2. Errors during write callback and flush are treated as hard errors, i.e the database is put in read-only mode and goes back to read-write only fater certain recovery actions are taken. 3. Both types of recovery rely on the SstFileManagerImpl to poll for sufficient disk space. We assume that there is a 1-1 mapping between an SFM and the underlying OS storage container. For cases where multiple DBs are hosted on a single storage container, the user is expected to allocate a single SFM instance and use the same one for all the DBs. If no SFM is specified by the user, DBImpl::Open() will allocate one, but this will be one per DB and each DB will recover independently. The recovery implemented by SFM is as follows - a) On the first occurance of an out of space error during compaction, subsequent compactions will be delayed until the disk free space check indicates enough available space. The required space is computed as the sum of input sizes. b) The free space check requirement will be removed once the amount of free space is greater than the size reserved by in progress compactions when the first error occured c) If the out of space error is a hard error, a background thread in SFM will poll for sufficient headroom before triggering the recovery of the database and putting it in write-only mode. The headroom is calculated as the sum of the write_buffer_size of all the DB instances associated with the SFM 4. EventListener callbacks will be called at the start and completion of automatic recovery. Users can disable the auto recov ery in the start callback, and later initiate it manually by calling DB::Resume() Todo: 1. More extensive testing 2. Add disk full condition to db_stress (follow-on PR) Pull Request resolved: https://github.com/facebook/rocksdb/pull/4164 Differential Revision: D9846378 Pulled By: anand1976 fbshipit-source-id: 80ea875dbd7f00205e19c82215ff6e37da10da4a
6 years ago
// this will unref its input_version and column_family_data
c.reset();
if (is_manual) {
ManualCompactionState* m = manual_compaction;
if (!status.ok()) {
m->status = status;
m->done = true;
}
// For universal compaction:
// Because universal compaction always happens at level 0, so one
// compaction will pick up all overlapped files. No files will be
// filtered out due to size limit and left for a successive compaction.
// So we can safely conclude the current compaction.
//
// Also note that, if we don't stop here, then the current compaction
// writes a new file back to level 0, which will be used in successive
// compaction. Hence the manual compaction will never finish.
//
// Stop the compaction if manual_end points to nullptr -- this means
// that we compacted the whole range. manual_end should always point
// to nullptr in case of universal compaction
if (m->manual_end == nullptr) {
m->done = true;
}
if (!m->done) {
// We only compacted part of the requested range. Update *m
// to the range that is left to be compacted.
// Universal and FIFO compactions should always compact the whole range
assert(m->cfd->ioptions()->compaction_style !=
kCompactionStyleUniversal ||
m->cfd->ioptions()->num_levels > 1);
assert(m->cfd->ioptions()->compaction_style != kCompactionStyleFIFO);
m->tmp_storage = *m->manual_end;
m->begin = &m->tmp_storage;
m->incomplete = true;
}
m->in_progress = false; // not being processed anymore
}
TEST_SYNC_POINT("DBImpl::BackgroundCompaction:Finish");
return status;
}
bool DBImpl::HasPendingManualCompaction() {
return (!manual_compaction_dequeue_.empty());
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
void DBImpl::AddManualCompaction(DBImpl::ManualCompactionState* m) {
Prevent corruption with parallel manual compactions and `change_level == true` (#9077) Summary: The bug can impact the following scenario. There must be two `CompactRange()`s, call them A and B. Compaction A must have `change_level=true`. Compactions A and B must run in parallel, and new data must be added while they run as well. Now, on to the details of the race condition. Compaction A must reach the refitting phase while B's next step is to trivial move new data (i.e., data that has been inserted behind A) down to the same level that A's refit targets (`CompactRangeOptions::target_level`). B must be unregistered (i.e., has not yet called `AddManualCompaction()` for the current `RunManualCompaction()`) while A invokes `DisableManualCompaction()`s to prepare for refitting. In the old code, B could still proceed to register a manual compaction, while A had disabled manual compaction. The next part of the race condition is B picks and schedules a trivial move while A has released the lock in refitting phase in order to persist the LSM state change (i.e., the log phase of `LogAndApply()`). That way, B does not see the refitted data when picking a trivial-move compaction. So it is susceptible to picking one that overlaps. Finally, B executes the picked trivial-move compaction. Trivial-move compactions are special in that they never check whether manual compaction is disabled. So the picked compaction causing overlap ends up being applied, leading to LSM corruption if `force_consistency_checks=false`, or entering read-only mode with `Status::Corruption` if `force_consistency_checks=true` (the default). The fix is just to prevent B from registering itself in `RunManualCompaction()` while manual compactions are disabled, consequently preventing any trivial move or other compaction from being picked/scheduled. Thanks to siying for finding the bug. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9077 Test Plan: The test does not go all the way in exposing the bug because it requires a compaction to be picked/scheduled while logging LSM state change for RefitLevel(). But the fix is to make such a compaction not picked/scheduled in the first place, so any repro of that scenario would end up hanging RefitLevel() logging. So instead I just verified no such compaction is registered in the scenario where `RefitLevel()` disables manual compactions. Reviewed By: siying Differential Revision: D31921908 Pulled By: ajkr fbshipit-source-id: 9bb5d0e847ad428211227f40830c685c209fbecb
3 years ago
assert(manual_compaction_paused_ == 0);
manual_compaction_dequeue_.push_back(m);
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
void DBImpl::RemoveManualCompaction(DBImpl::ManualCompactionState* m) {
// Remove from queue
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
std::deque<ManualCompactionState*>::iterator it =
manual_compaction_dequeue_.begin();
while (it != manual_compaction_dequeue_.end()) {
if (m == (*it)) {
it = manual_compaction_dequeue_.erase(it);
return;
}
++it;
}
assert(false);
return;
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
bool DBImpl::ShouldntRunManualCompaction(ManualCompactionState* m) {
if (num_running_ingest_file_ > 0) {
// We need to wait for other IngestExternalFile() calls to finish
// before running a manual compaction.
return true;
}
if (m->exclusive) {
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
return (bg_bottom_compaction_scheduled_ > 0 ||
bg_compaction_scheduled_ > 0);
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
std::deque<ManualCompactionState*>::iterator it =
manual_compaction_dequeue_.begin();
bool seen = false;
while (it != manual_compaction_dequeue_.end()) {
if (m == (*it)) {
++it;
seen = true;
continue;
} else if (MCOverlap(m, (*it)) && (!seen && !(*it)->in_progress)) {
// Consider the other manual compaction *it, conflicts if:
// overlaps with m
// and (*it) is ahead in the queue and is not yet in progress
return true;
}
++it;
}
return false;
}
bool DBImpl::HaveManualCompaction(ColumnFamilyData* cfd) {
// Remove from priority queue
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
std::deque<ManualCompactionState*>::iterator it =
manual_compaction_dequeue_.begin();
while (it != manual_compaction_dequeue_.end()) {
if ((*it)->exclusive) {
return true;
}
if ((cfd == (*it)->cfd) && (!((*it)->in_progress || (*it)->done))) {
// Allow automatic compaction if manual compaction is
// in progress
return true;
}
++it;
}
return false;
}
bool DBImpl::HasExclusiveManualCompaction() {
// Remove from priority queue
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
std::deque<ManualCompactionState*>::iterator it =
manual_compaction_dequeue_.begin();
while (it != manual_compaction_dequeue_.end()) {
if ((*it)->exclusive) {
return true;
}
++it;
}
return false;
}
Introduce bottom-pri thread pool for large universal compactions Summary: When we had a single thread pool for compactions, a thread could be busy for a long time (minutes) executing a compaction involving the bottom level. In multi-instance setups, the entire thread pool could be consumed by such bottom-level compactions. Then, top-level compactions (e.g., a few L0 files) would be blocked for a long time ("head-of-line blocking"). Such top-level compactions are critical to prevent compaction stalls as they can quickly reduce number of L0 files / sorted runs. This diff introduces a bottom-priority queue for universal compactions including the bottom level. This alleviates the head-of-line blocking situation for fast, top-level compactions. - Added `Env::Priority::BOTTOM` thread pool. This feature is only enabled if user explicitly configures it to have a positive number of threads. - Changed `ThreadPoolImpl`'s default thread limit from one to zero. This change is invisible to users as we call `IncBackgroundThreadsIfNeeded` on the low-pri/high-pri pools during `DB::Open` with values of at least one. It is necessary, though, for bottom-pri to start with zero threads so the feature is disabled by default. - Separated `ManualCompaction` into two parts in `PrepickedCompaction`. `PrepickedCompaction` is used for any compaction that's picked outside of its execution thread, either manual or automatic. - Forward universal compactions involving last level to the bottom pool (worker thread's entry point is `BGWorkBottomCompaction`). - Track `bg_bottom_compaction_scheduled_` so we can wait for bottom-level compactions to finish. We don't count them against the background jobs limits. So users of this feature will get an extra compaction for free. Closes https://github.com/facebook/rocksdb/pull/2580 Differential Revision: D5422916 Pulled By: ajkr fbshipit-source-id: a74bd11f1ea4933df3739b16808bb21fcd512333
7 years ago
bool DBImpl::MCOverlap(ManualCompactionState* m, ManualCompactionState* m1) {
if ((m->exclusive) || (m1->exclusive)) {
return true;
}
if (m->cfd != m1->cfd) {
return false;
}
return false;
}
#ifndef ROCKSDB_LITE
void DBImpl::BuildCompactionJobInfo(
const ColumnFamilyData* cfd, Compaction* c, const Status& st,
const CompactionJobStats& compaction_job_stats, const int job_id,
const Version* current, CompactionJobInfo* compaction_job_info) const {
assert(compaction_job_info != nullptr);
compaction_job_info->cf_id = cfd->GetID();
compaction_job_info->cf_name = cfd->GetName();
compaction_job_info->status = st;
compaction_job_info->thread_id = env_->GetThreadID();
compaction_job_info->job_id = job_id;
compaction_job_info->base_input_level = c->start_level();
compaction_job_info->output_level = c->output_level();
compaction_job_info->stats = compaction_job_stats;
compaction_job_info->table_properties = c->GetOutputTableProperties();
compaction_job_info->compaction_reason = c->compaction_reason();
compaction_job_info->compression = c->output_compression();
for (size_t i = 0; i < c->num_input_levels(); ++i) {
for (const auto fmd : *c->inputs(i)) {
const FileDescriptor& desc = fmd->fd;
const uint64_t file_number = desc.GetNumber();
auto fn = TableFileName(c->immutable_options()->cf_paths, file_number,
desc.GetPathId());
compaction_job_info->input_files.push_back(fn);
compaction_job_info->input_file_infos.push_back(CompactionFileInfo{
static_cast<int>(i), file_number, fmd->oldest_blob_file_number});
if (compaction_job_info->table_properties.count(fn) == 0) {
std::shared_ptr<const TableProperties> tp;
auto s = current->GetTableProperties(&tp, fmd, &fn);
if (s.ok()) {
compaction_job_info->table_properties[fn] = tp;
}
}
}
}
for (const auto& newf : c->edit()->GetNewFiles()) {
const FileMetaData& meta = newf.second;
const FileDescriptor& desc = meta.fd;
const uint64_t file_number = desc.GetNumber();
compaction_job_info->output_files.push_back(TableFileName(
c->immutable_options()->cf_paths, file_number, desc.GetPathId()));
compaction_job_info->output_file_infos.push_back(CompactionFileInfo{
newf.first, file_number, meta.oldest_blob_file_number});
}
compaction_job_info->blob_compression_type =
c->mutable_cf_options()->blob_compression_type;
// Update BlobFilesInfo.
for (const auto& blob_file : c->edit()->GetBlobFileAdditions()) {
BlobFileAdditionInfo blob_file_addition_info(
BlobFileName(c->immutable_options()->cf_paths.front().path,
blob_file.GetBlobFileNumber()) /*blob_file_path*/,
blob_file.GetBlobFileNumber(), blob_file.GetTotalBlobCount(),
blob_file.GetTotalBlobBytes());
compaction_job_info->blob_file_addition_infos.emplace_back(
std::move(blob_file_addition_info));
}
// Update BlobFilesGarbageInfo.
for (const auto& blob_file : c->edit()->GetBlobFileGarbages()) {
BlobFileGarbageInfo blob_file_garbage_info(
BlobFileName(c->immutable_options()->cf_paths.front().path,
blob_file.GetBlobFileNumber()) /*blob_file_path*/,
blob_file.GetBlobFileNumber(), blob_file.GetGarbageBlobCount(),
blob_file.GetGarbageBlobBytes());
compaction_job_info->blob_file_garbage_infos.emplace_back(
std::move(blob_file_garbage_info));
}
}
#endif
// SuperVersionContext gets created and destructed outside of the lock --
// we use this conveniently to:
// * malloc one SuperVersion() outside of the lock -- new_superversion
// * delete SuperVersion()s outside of the lock -- superversions_to_free
//
// However, if InstallSuperVersionAndScheduleWork() gets called twice with the
// same sv_context, we can't reuse the SuperVersion() that got
// malloced because
// first call already used it. In that rare case, we take a hit and create a
// new SuperVersion() inside of the mutex. We do similar thing
// for superversion_to_free
void DBImpl::InstallSuperVersionAndScheduleWork(
ColumnFamilyData* cfd, SuperVersionContext* sv_context,
Make mempurge a background process (equivalent to in-memory compaction). (#8505) Summary: In https://github.com/facebook/rocksdb/issues/8454, I introduced a new process baptized `MemPurge` (memtable garbage collection). This new PR is built upon this past mempurge prototype. In this PR, I made the `mempurge` process a background task, which provides superior performance since the mempurge process does not cling on the db_mutex anymore, and addresses severe restrictions from the past iteration (including a scenario where the past mempurge was failling, when a memtable was mempurged but was still referred to by an iterator/snapshot/...). Now the mempurge process ressembles an in-memory compaction process: the stack of immutable memtables is filtered out, and the useful payload is used to populate an output memtable. If the output memtable is filled at more than 60% capacity (arbitrary heuristic) the mempurge process is aborted and a regular flush process takes place, else the output memtable is kept in the immutable memtable stack. Note that adding this output memtable to the `imm()` memtable stack does not trigger another flush process, so that the flush thread can go to sleep at the end of a successful mempurge. MemPurge is activated by making the `experimental_allow_mempurge` flag `true`. When activated, the `MemPurge` process will always happen when the flush reason is `kWriteBufferFull`. The 3 unit tests confirm that this process supports `Put`, `Get`, `Delete`, `DeleteRange` operators and is compatible with `Iterators` and `CompactionFilters`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8505 Reviewed By: pdillinger Differential Revision: D29619283 Pulled By: bjlemaire fbshipit-source-id: 8a99bee76b63a8211bff1a00e0ae32360aaece95
3 years ago
const MutableCFOptions& mutable_cf_options) {
mutex_.AssertHeld();
// Update max_total_in_memory_state_
size_t old_memtable_size = 0;
auto* old_sv = cfd->GetSuperVersion();
if (old_sv) {
old_memtable_size = old_sv->mutable_cf_options.write_buffer_size *
old_sv->mutable_cf_options.max_write_buffer_number;
}
// this branch is unlikely to step in
if (UNLIKELY(sv_context->new_superversion == nullptr)) {
sv_context->NewSuperVersion();
}
Fix a race in ColumnFamilyData::UnrefAndTryDelete (#8605) Summary: The `ColumnFamilyData::UnrefAndTryDelete` code currently on the trunk unlocks the DB mutex before destroying the `ThreadLocalPtr` holding the per-thread `SuperVersion` pointers when the only remaining reference is the back reference from `super_version_`. The idea behind this was to break the circular dependency between `ColumnFamilyData` and `SuperVersion`: when the penultimate reference goes away, `ColumnFamilyData` can clean up the `SuperVersion`, which can in turn clean up `ColumnFamilyData`. (Assuming there is a `SuperVersion` and it is not referenced by anything else.) However, unlocking the mutex throws a wrench in this plan by making it possible for another thread to jump in and take another reference to the `ColumnFamilyData`, keeping the object alive in a zombie `ThreadLocalPtr`-less state. This can cause issues like https://github.com/facebook/rocksdb/issues/8440 , https://github.com/facebook/rocksdb/issues/8382 , and might also explain the `was_last_ref` assertion failures from the `ColumnFamilySet` destructor we sometimes observe during close in our stress tests. Digging through the archives, this unlocking goes way back to 2014 (or earlier). The original rationale was that `SuperVersionUnrefHandle` used to lock the mutex so it can call `SuperVersion::Cleanup`; however, this logic turned out to be deadlock-prone. https://github.com/facebook/rocksdb/pull/3510 fixed the deadlock but left the unlocking in place. https://github.com/facebook/rocksdb/pull/6147 then introduced the circular dependency and associated cleanup logic described above (in order to enable iterators to keep the `ColumnFamilyData` for dropped column families alive), and moved the unlocking-relocking snippet to its present location in `UnrefAndTryDelete`. Finally, https://github.com/facebook/rocksdb/pull/7749 fixed a memory leak but apparently exacerbated the race by (otherwise correctly) switching to `UnrefAndTryDelete` in `SuperVersion::Cleanup`. The patch simply eliminates the unlocking and relocking, which has been unnecessary ever since https://github.com/facebook/rocksdb/issues/3510 made `SuperVersionUnrefHandle` lock-free. This closes the window during which another thread could increase the reference count, and hopefully fixes the issues above. Pull Request resolved: https://github.com/facebook/rocksdb/pull/8605 Test Plan: Ran `make check` and stress tests locally. Reviewed By: pdillinger Differential Revision: D30051035 Pulled By: ltamasi fbshipit-source-id: 8fe559e4b4ad69fc142579f8bc393ef525918528
3 years ago
cfd->InstallSuperVersion(sv_context, mutable_cf_options);
// There may be a small data race here. The snapshot tricking bottommost
// compaction may already be released here. But assuming there will always be
// newer snapshot created and released frequently, the compaction will be
// triggered soon anyway.
bottommost_files_mark_threshold_ = kMaxSequenceNumber;
for (auto* my_cfd : *versions_->GetColumnFamilySet()) {
bottommost_files_mark_threshold_ = std::min(
bottommost_files_mark_threshold_,
my_cfd->current()->storage_info()->bottommost_files_mark_threshold());
}
// Whenever we install new SuperVersion, we might need to issue new flushes or
// compactions.
SchedulePendingCompaction(cfd);
MaybeScheduleFlushOrCompaction();
// Update max_total_in_memory_state_
max_total_in_memory_state_ = max_total_in_memory_state_ - old_memtable_size +
mutable_cf_options.write_buffer_size *
mutable_cf_options.max_write_buffer_number;
}
Fix race condition causing double deletion of ssts Summary: Possible interleaved execution of background compaction thread calling `FindObsoleteFiles (no full scan) / PurgeObsoleteFiles` and user thread calling `FindObsoleteFiles (full scan) / PurgeObsoleteFiles` can lead to race condition on which RocksDB attempts to delete a file twice. The second attempt will fail and return `IO error`. This may occur to other files, but this PR targets sst. Also add a unit test to verify that this PR fixes the issue. The newly added unit test `obsolete_files_test` has a test case for this scenario, implemented in `ObsoleteFilesTest#RaceForObsoleteFileDeletion`. `TestSyncPoint`s are used to coordinate the interleaving the `user_thread` and background compaction thread. They execute as follows ``` timeline user_thread background_compaction thread t1 | FindObsoleteFiles(full_scan=false) t2 | FindObsoleteFiles(full_scan=true) t3 | PurgeObsoleteFiles t4 | PurgeObsoleteFiles V ``` When `user_thread` invokes `FindObsoleteFiles` with full scan, it collects ALL files in RocksDB directory, including the ones that background compaction thread have collected in its job context. Then `user_thread` will see an IO error when trying to delete these files in `PurgeObsoleteFiles` because background compaction thread has already deleted the file in `PurgeObsoleteFiles`. To fix this, we make RocksDB remember which (SST) files have been found by threads after calling `FindObsoleteFiles` (see `DBImpl#files_grabbed_for_purge_`). Therefore, when another thread calls `FindObsoleteFiles` with full scan, it will not collect such files. ajkr could you take a look and comment? Thanks! Closes https://github.com/facebook/rocksdb/pull/3638 Differential Revision: D7384372 Pulled By: riversand963 fbshipit-source-id: 01489516d60012e722ee65a80e1449e589ce26d3
7 years ago
// ShouldPurge is called by FindObsoleteFiles when doing a full scan,
// and db mutex (mutex_) should already be held.
Fix race condition causing double deletion of ssts Summary: Possible interleaved execution of background compaction thread calling `FindObsoleteFiles (no full scan) / PurgeObsoleteFiles` and user thread calling `FindObsoleteFiles (full scan) / PurgeObsoleteFiles` can lead to race condition on which RocksDB attempts to delete a file twice. The second attempt will fail and return `IO error`. This may occur to other files, but this PR targets sst. Also add a unit test to verify that this PR fixes the issue. The newly added unit test `obsolete_files_test` has a test case for this scenario, implemented in `ObsoleteFilesTest#RaceForObsoleteFileDeletion`. `TestSyncPoint`s are used to coordinate the interleaving the `user_thread` and background compaction thread. They execute as follows ``` timeline user_thread background_compaction thread t1 | FindObsoleteFiles(full_scan=false) t2 | FindObsoleteFiles(full_scan=true) t3 | PurgeObsoleteFiles t4 | PurgeObsoleteFiles V ``` When `user_thread` invokes `FindObsoleteFiles` with full scan, it collects ALL files in RocksDB directory, including the ones that background compaction thread have collected in its job context. Then `user_thread` will see an IO error when trying to delete these files in `PurgeObsoleteFiles` because background compaction thread has already deleted the file in `PurgeObsoleteFiles`. To fix this, we make RocksDB remember which (SST) files have been found by threads after calling `FindObsoleteFiles` (see `DBImpl#files_grabbed_for_purge_`). Therefore, when another thread calls `FindObsoleteFiles` with full scan, it will not collect such files. ajkr could you take a look and comment? Thanks! Closes https://github.com/facebook/rocksdb/pull/3638 Differential Revision: D7384372 Pulled By: riversand963 fbshipit-source-id: 01489516d60012e722ee65a80e1449e589ce26d3
7 years ago
// Actually, the current implementation of FindObsoleteFiles with
// full_scan=true can issue I/O requests to obtain list of files in
// directories, e.g. env_->getChildren while holding db mutex.
bool DBImpl::ShouldPurge(uint64_t file_number) const {
return files_grabbed_for_purge_.find(file_number) ==
files_grabbed_for_purge_.end() &&
purge_files_.find(file_number) == purge_files_.end();
Fix race condition causing double deletion of ssts Summary: Possible interleaved execution of background compaction thread calling `FindObsoleteFiles (no full scan) / PurgeObsoleteFiles` and user thread calling `FindObsoleteFiles (full scan) / PurgeObsoleteFiles` can lead to race condition on which RocksDB attempts to delete a file twice. The second attempt will fail and return `IO error`. This may occur to other files, but this PR targets sst. Also add a unit test to verify that this PR fixes the issue. The newly added unit test `obsolete_files_test` has a test case for this scenario, implemented in `ObsoleteFilesTest#RaceForObsoleteFileDeletion`. `TestSyncPoint`s are used to coordinate the interleaving the `user_thread` and background compaction thread. They execute as follows ``` timeline user_thread background_compaction thread t1 | FindObsoleteFiles(full_scan=false) t2 | FindObsoleteFiles(full_scan=true) t3 | PurgeObsoleteFiles t4 | PurgeObsoleteFiles V ``` When `user_thread` invokes `FindObsoleteFiles` with full scan, it collects ALL files in RocksDB directory, including the ones that background compaction thread have collected in its job context. Then `user_thread` will see an IO error when trying to delete these files in `PurgeObsoleteFiles` because background compaction thread has already deleted the file in `PurgeObsoleteFiles`. To fix this, we make RocksDB remember which (SST) files have been found by threads after calling `FindObsoleteFiles` (see `DBImpl#files_grabbed_for_purge_`). Therefore, when another thread calls `FindObsoleteFiles` with full scan, it will not collect such files. ajkr could you take a look and comment? Thanks! Closes https://github.com/facebook/rocksdb/pull/3638 Differential Revision: D7384372 Pulled By: riversand963 fbshipit-source-id: 01489516d60012e722ee65a80e1449e589ce26d3
7 years ago
}
// MarkAsGrabbedForPurge is called by FindObsoleteFiles, and db mutex
// (mutex_) should already be held.
void DBImpl::MarkAsGrabbedForPurge(uint64_t file_number) {
files_grabbed_for_purge_.insert(file_number);
Fix race condition causing double deletion of ssts Summary: Possible interleaved execution of background compaction thread calling `FindObsoleteFiles (no full scan) / PurgeObsoleteFiles` and user thread calling `FindObsoleteFiles (full scan) / PurgeObsoleteFiles` can lead to race condition on which RocksDB attempts to delete a file twice. The second attempt will fail and return `IO error`. This may occur to other files, but this PR targets sst. Also add a unit test to verify that this PR fixes the issue. The newly added unit test `obsolete_files_test` has a test case for this scenario, implemented in `ObsoleteFilesTest#RaceForObsoleteFileDeletion`. `TestSyncPoint`s are used to coordinate the interleaving the `user_thread` and background compaction thread. They execute as follows ``` timeline user_thread background_compaction thread t1 | FindObsoleteFiles(full_scan=false) t2 | FindObsoleteFiles(full_scan=true) t3 | PurgeObsoleteFiles t4 | PurgeObsoleteFiles V ``` When `user_thread` invokes `FindObsoleteFiles` with full scan, it collects ALL files in RocksDB directory, including the ones that background compaction thread have collected in its job context. Then `user_thread` will see an IO error when trying to delete these files in `PurgeObsoleteFiles` because background compaction thread has already deleted the file in `PurgeObsoleteFiles`. To fix this, we make RocksDB remember which (SST) files have been found by threads after calling `FindObsoleteFiles` (see `DBImpl#files_grabbed_for_purge_`). Therefore, when another thread calls `FindObsoleteFiles` with full scan, it will not collect such files. ajkr could you take a look and comment? Thanks! Closes https://github.com/facebook/rocksdb/pull/3638 Differential Revision: D7384372 Pulled By: riversand963 fbshipit-source-id: 01489516d60012e722ee65a80e1449e589ce26d3
7 years ago
}
void DBImpl::SetSnapshotChecker(SnapshotChecker* snapshot_checker) {
InstrumentedMutexLock l(&mutex_);
// snapshot_checker_ should only set once. If we need to set it multiple
// times, we need to make sure the old one is not deleted while it is still
// using by a compaction job.
assert(!snapshot_checker_);
snapshot_checker_.reset(snapshot_checker);
}
void DBImpl::GetSnapshotContext(
JobContext* job_context, std::vector<SequenceNumber>* snapshot_seqs,
SequenceNumber* earliest_write_conflict_snapshot,
SnapshotChecker** snapshot_checker_ptr) {
mutex_.AssertHeld();
assert(job_context != nullptr);
assert(snapshot_seqs != nullptr);
assert(earliest_write_conflict_snapshot != nullptr);
assert(snapshot_checker_ptr != nullptr);
*snapshot_checker_ptr = snapshot_checker_.get();
if (use_custom_gc_ && *snapshot_checker_ptr == nullptr) {
*snapshot_checker_ptr = DisableGCSnapshotChecker::Instance();
}
if (*snapshot_checker_ptr != nullptr) {
// If snapshot_checker is used, that means the flush/compaction may
// contain values not visible to snapshot taken after
// flush/compaction job starts. Take a snapshot and it will appear
// in snapshot_seqs and force compaction iterator to consider such
// snapshots.
const Snapshot* job_snapshot =
GetSnapshotImpl(false /*write_conflict_boundary*/, false /*lock*/);
job_context->job_snapshot.reset(new ManagedSnapshot(this, job_snapshot));
}
*snapshot_seqs = snapshots_.GetAll(earliest_write_conflict_snapshot);
}
Status DBImpl::WaitForCompact(bool wait_unscheduled) {
// Wait until the compaction completes
InstrumentedMutexLock l(&mutex_);
while ((bg_bottom_compaction_scheduled_ || bg_compaction_scheduled_ ||
bg_flush_scheduled_ ||
(wait_unscheduled && unscheduled_compactions_)) &&
(error_handler_.GetBGError().ok())) {
bg_cv_.Wait();
}
return error_handler_.GetBGError();
}
} // namespace ROCKSDB_NAMESPACE