rocksdb/db/db_impl/db_impl.h

2208 lines
93 KiB
C
Raw Normal View History

// 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.
#pragma once
#include <atomic>
#include <deque>
#include <functional>
#include <limits>
#include <list>
#include <map>
#include <set>
#include <string>
#include <utility>
#include <vector>
#include "db/column_family.h"
#include "db/compaction/compaction_job.h"
#include "db/dbformat.h"
#include "db/error_handler.h"
#include "db/event_helpers.h"
#include "db/external_sst_file_ingestion_job.h"
#include "db/flush_job.h"
#include "db/flush_scheduler.h"
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
2019-07-17 21:22:21 +02:00
#include "db/import_column_family_job.h"
#include "db/internal_stats.h"
#include "db/log_writer.h"
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
#include "db/logs_with_prep_tracker.h"
#include "db/memtable_list.h"
#include "db/pre_release_callback.h"
#include "db/range_del_aggregator.h"
#include "db/read_callback.h"
#include "db/snapshot_checker.h"
#include "db/snapshot_impl.h"
Refactor trimming logic for immutable memtables (#5022) Summary: MyRocks currently sets `max_write_buffer_number_to_maintain` in order to maintain enough history for transaction conflict checking. The effectiveness of this approach depends on the size of memtables. When memtables are small, it may not keep enough history; when memtables are large, this may consume too much memory. We are proposing a new way to configure memtable list history: by limiting the memory usage of immutable memtables. The new option is `max_write_buffer_size_to_maintain` and it will take precedence over the old `max_write_buffer_number_to_maintain` if they are both set to non-zero values. The new option accounts for the total memory usage of flushed immutable memtables and mutable memtable. When the total usage exceeds the limit, RocksDB may start dropping immutable memtables (which is also called trimming history), starting from the oldest one. The semantics of the old option actually works both as an upper bound and lower bound. History trimming will start if number of immutable memtables exceeds the limit, but it will never go below (limit-1) due to history trimming. In order the mimic the behavior with the new option, history trimming will stop if dropping the next immutable memtable causes the total memory usage go below the size limit. For example, assuming the size limit is set to 64MB, and there are 3 immutable memtables with sizes of 20, 30, 30. Although the total memory usage is 80MB > 64MB, dropping the oldest memtable will reduce the memory usage to 60MB < 64MB, so in this case no memtable will be dropped. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5022 Differential Revision: D14394062 Pulled By: miasantreble fbshipit-source-id: 60457a509c6af89d0993f988c9b5c2aa9e45f5c5
2019-08-23 22:54:09 +02:00
#include "db/trim_history_scheduler.h"
#include "db/version_edit.h"
#include "db/wal_manager.h"
#include "db/write_controller.h"
#include "db/write_thread.h"
#include "logging/event_logger.h"
#include "monitoring/instrumented_mutex.h"
#include "options/db_options.h"
#include "port/port.h"
#include "rocksdb/db.h"
#include "rocksdb/env.h"
#include "rocksdb/memtablerep.h"
#include "rocksdb/status.h"
#include "rocksdb/trace_reader_writer.h"
#include "rocksdb/transaction_log.h"
#include "rocksdb/write_buffer_manager.h"
#include "table/scoped_arena_iterator.h"
#include "trace_replay/block_cache_tracer.h"
#include "trace_replay/io_tracer.h"
#include "trace_replay/trace_replay.h"
#include "util/autovector.h"
#include "util/hash.h"
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
2018-10-09 07:52:58 +02:00
#include "util/repeatable_thread.h"
#include "util/stop_watch.h"
#include "util/thread_local.h"
namespace ROCKSDB_NAMESPACE {
class Arena;
class ArenaWrappedDBIter;
class InMemoryStatsHistoryIterator;
class MemTable;
class PersistentStatsHistoryIterator;
class PeriodicWorkScheduler;
#ifndef NDEBUG
class PeriodicWorkTestScheduler;
#endif // !NDEBUG
class TableCache;
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
2018-12-13 22:16:04 +01:00
class TaskLimiterToken;
class Version;
class VersionEdit;
class VersionSet;
class WriteCallback;
struct JobContext;
struct ExternalSstFileInfo;
struct MemTableInfo;
// Class to maintain directories for all database paths other than main one.
class Directories {
public:
IOStatus SetDirectories(FileSystem* fs, const std::string& dbname,
const std::string& wal_dir,
const std::vector<DbPath>& data_paths);
FSDirectory* GetDataDir(size_t path_id) const {
assert(path_id < data_dirs_.size());
FSDirectory* ret_dir = data_dirs_[path_id].get();
if (ret_dir == nullptr) {
// Should use db_dir_
return db_dir_.get();
}
return ret_dir;
}
FSDirectory* GetWalDir() {
if (wal_dir_) {
return wal_dir_.get();
}
return db_dir_.get();
}
FSDirectory* GetDbDir() { return db_dir_.get(); }
private:
std::unique_ptr<FSDirectory> db_dir_;
std::vector<std::unique_ptr<FSDirectory>> data_dirs_;
std::unique_ptr<FSDirectory> wal_dir_;
};
// While DB is the public interface of RocksDB, and DBImpl is the actual
// class implementing it. It's the entrance of the core RocksdB engine.
// All other DB implementations, e.g. TransactionDB, BlobDB, etc, wrap a
// DBImpl internally.
// Other than functions implementing the DB interface, some public
// functions are there for other internal components to call. For
// example, TransactionDB directly calls DBImpl::WriteImpl() and
// BlobDB directly calls DBImpl::GetImpl(). Some other functions
// are for sub-components to call. For example, ColumnFamilyHandleImpl
// calls DBImpl::FindObsoleteFiles().
//
// Since it's a very large class, the definition of the functions is
// divided in several db_impl_*.cc files, besides db_impl.cc.
class DBImpl : public DB {
public:
DBImpl(const DBOptions& options, const std::string& dbname,
const bool seq_per_batch = false, const bool batch_per_txn = true);
// No copying allowed
DBImpl(const DBImpl&) = delete;
void operator=(const DBImpl&) = delete;
virtual ~DBImpl();
// ---- Implementations of the DB interface ----
using DB::Resume;
virtual Status Resume() override;
using DB::Put;
virtual Status Put(const WriteOptions& options,
ColumnFamilyHandle* column_family, const Slice& key,
const Slice& value) override;
using DB::Merge;
virtual Status Merge(const WriteOptions& options,
ColumnFamilyHandle* column_family, const Slice& key,
const Slice& value) override;
using DB::Delete;
virtual Status Delete(const WriteOptions& options,
ColumnFamilyHandle* column_family,
const Slice& key) override;
Support for SingleDelete() Summary: This patch fixes #7460559. It introduces SingleDelete as a new database operation. This operation can be used to delete keys that were never overwritten (no put following another put of the same key). If an overwritten key is single deleted the behavior is undefined. Single deletion of a non-existent key has no effect but multiple consecutive single deletions are not allowed (see limitations). In contrast to the conventional Delete() operation, the deletion entry is removed along with the value when the two are lined up in a compaction. Note: The semantics are similar to @igor's prototype that allowed to have this behavior on the granularity of a column family ( https://reviews.facebook.net/D42093 ). This new patch, however, is more aggressive when it comes to removing tombstones: It removes the SingleDelete together with the value whenever there is no snapshot between them while the older patch only did this when the sequence number of the deletion was older than the earliest snapshot. Most of the complex additions are in the Compaction Iterator, all other changes should be relatively straightforward. The patch also includes basic support for single deletions in db_stress and db_bench. Limitations: - Not compatible with cuckoo hash tables - Single deletions cannot be used in combination with merges and normal deletions on the same key (other keys are not affected by this) - Consecutive single deletions are currently not allowed (and older version of this patch supported this so it could be resurrected if needed) Test Plan: make all check Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor Reviewed By: igor Subscribers: maykov, dhruba, leveldb Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
using DB::SingleDelete;
virtual Status SingleDelete(const WriteOptions& options,
ColumnFamilyHandle* column_family,
const Slice& key) override;
using DB::Write;
virtual Status Write(const WriteOptions& options,
WriteBatch* updates) override;
using DB::Get;
virtual Status Get(const ReadOptions& options,
ColumnFamilyHandle* column_family, const Slice& key,
PinnableSlice* value) override;
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
2020-03-03 00:58:32 +01:00
virtual Status Get(const ReadOptions& options,
ColumnFamilyHandle* column_family, const Slice& key,
PinnableSlice* value, std::string* timestamp) override;
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
2019-08-06 23:22:34 +02:00
using DB::GetMergeOperands;
Status GetMergeOperands(const ReadOptions& options,
ColumnFamilyHandle* column_family, const Slice& key,
PinnableSlice* merge_operands,
GetMergeOperandsOptions* get_merge_operands_options,
int* number_of_operands) override {
GetImplOptions get_impl_options;
get_impl_options.column_family = column_family;
get_impl_options.merge_operands = merge_operands;
get_impl_options.get_merge_operands_options = get_merge_operands_options;
get_impl_options.number_of_operands = number_of_operands;
get_impl_options.get_value = false;
return GetImpl(options, key, get_impl_options);
}
using DB::MultiGet;
virtual std::vector<Status> MultiGet(
const ReadOptions& options,
const std::vector<ColumnFamilyHandle*>& column_family,
const std::vector<Slice>& keys,
std::vector<std::string>* values) override;
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
2020-03-24 19:21:10 +01:00
virtual std::vector<Status> MultiGet(
const ReadOptions& options,
const std::vector<ColumnFamilyHandle*>& column_family,
const std::vector<Slice>& keys, std::vector<std::string>* values,
std::vector<std::string>* timestamps) override;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
// This MultiGet is a batched version, which may be faster than calling Get
// multiple times, especially if the keys have some spatial locality that
// enables them to be queried in the same SST files/set of files. The larger
// the batch size, the more scope for batching and performance improvement
// The values and statuses parameters are arrays with number of elements
// equal to keys.size(). This allows the storage for those to be alloacted
// by the caller on the stack for small batches
virtual void MultiGet(const ReadOptions& options,
ColumnFamilyHandle* column_family,
const size_t num_keys, const Slice* keys,
PinnableSlice* values, Status* statuses,
const bool sorted_input = false) override;
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
2020-03-24 19:21:10 +01:00
virtual void MultiGet(const ReadOptions& options,
ColumnFamilyHandle* column_family,
const size_t num_keys, const Slice* keys,
PinnableSlice* values, std::string* timestamps,
Status* statuses,
const bool sorted_input = false) override;
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
virtual void MultiGet(const ReadOptions& options, const size_t num_keys,
ColumnFamilyHandle** column_families, const Slice* keys,
PinnableSlice* values, Status* statuses,
const bool sorted_input = false) override;
multiget support for timestamps (#6483) Summary: Add timestamp support for MultiGet(). timestamp from readoptions is honored, and timestamps can be returned along with values. MultiReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 17bef7d3a): multireadrandom : 104.173 micros/op 307167 ops/sec; (5462999 of 5462999 found) This PR: multireadrandom : 104.199 micros/op 307095 ops/sec; (5307999 of 5307999 found) .\db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=multireadrandom --use_existing_db=1 --num=25000000 --threads=32 --allow_concurrent_memtable_write=0 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6483 Reviewed By: anand1976 Differential Revision: D20498373 Pulled By: riversand963 fbshipit-source-id: 8505f22bc40fd791bc7dd05e48d7e67c91edb627
2020-03-24 19:21:10 +01:00
virtual void MultiGet(const ReadOptions& options, const size_t num_keys,
ColumnFamilyHandle** column_families, const Slice* keys,
PinnableSlice* values, std::string* timestamps,
Status* statuses,
const bool sorted_input = false) override;
virtual void MultiGetWithCallback(
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
const ReadOptions& options, ColumnFamilyHandle* column_family,
ReadCallback* callback,
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys);
Introduce a new MultiGet batching implementation (#5011) Summary: This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching. Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to - 1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch() 2. Bloom filter cachelines can be prefetched, hiding the cache miss latency The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress. Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32). Batch Sizes 1 | 2 | 4 | 8 | 16 | 32 Random pattern (Stride length 0) 4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get 4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching) 4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching) Good locality (Stride length 16) 4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753 4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781 4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135 Good locality (Stride length 256) 4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232 4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268 4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62 Medium locality (Stride length 4096) 4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555 4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465 4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891 dbbench command used (on a DB with 4 levels, 12 million keys)- TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4 Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011 Differential Revision: D14348703 Pulled By: anand1976 fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
virtual Status CreateColumnFamily(const ColumnFamilyOptions& cf_options,
const std::string& column_family,
ColumnFamilyHandle** handle) override;
virtual Status CreateColumnFamilies(
const ColumnFamilyOptions& cf_options,
const std::vector<std::string>& column_family_names,
std::vector<ColumnFamilyHandle*>* handles) override;
virtual Status CreateColumnFamilies(
const std::vector<ColumnFamilyDescriptor>& column_families,
std::vector<ColumnFamilyHandle*>* handles) override;
virtual Status DropColumnFamily(ColumnFamilyHandle* column_family) override;
virtual Status DropColumnFamilies(
const std::vector<ColumnFamilyHandle*>& column_families) override;
// Returns false if key doesn't exist in the database and true if it may.
// If value_found is not passed in as null, then return the value if found in
// memory. On return, if value was found, then value_found will be set to true
// , otherwise false.
using DB::KeyMayExist;
virtual bool KeyMayExist(const ReadOptions& options,
ColumnFamilyHandle* column_family, const Slice& key,
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
2020-03-03 00:58:32 +01:00
std::string* value, std::string* timestamp,
bool* value_found = nullptr) override;
using DB::NewIterator;
virtual Iterator* NewIterator(const ReadOptions& options,
ColumnFamilyHandle* column_family) override;
virtual Status NewIterators(
const ReadOptions& options,
const std::vector<ColumnFamilyHandle*>& column_families,
std::vector<Iterator*>* iterators) override;
virtual const Snapshot* GetSnapshot() override;
virtual void ReleaseSnapshot(const Snapshot* snapshot) override;
using DB::GetProperty;
virtual bool GetProperty(ColumnFamilyHandle* column_family,
const Slice& property, std::string* value) override;
using DB::GetMapProperty;
virtual bool GetMapProperty(
ColumnFamilyHandle* column_family, const Slice& property,
std::map<std::string, std::string>* value) override;
using DB::GetIntProperty;
virtual bool GetIntProperty(ColumnFamilyHandle* column_family,
const Slice& property, uint64_t* value) override;
using DB::GetAggregatedIntProperty;
virtual bool GetAggregatedIntProperty(const Slice& property,
uint64_t* aggregated_value) override;
using DB::GetApproximateSizes;
virtual Status GetApproximateSizes(const SizeApproximationOptions& options,
ColumnFamilyHandle* column_family,
const Range* range, int n,
uint64_t* sizes) override;
using DB::GetApproximateMemTableStats;
virtual void GetApproximateMemTableStats(ColumnFamilyHandle* column_family,
const Range& range,
uint64_t* const count,
uint64_t* const size) override;
using DB::CompactRange;
virtual Status CompactRange(const CompactRangeOptions& options,
ColumnFamilyHandle* column_family,
const Slice* begin, const Slice* end) override;
using DB::CompactFiles;
virtual Status 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 = -1,
std::vector<std::string>* const output_file_names = nullptr,
CompactionJobInfo* compaction_job_info = nullptr) override;
virtual Status PauseBackgroundWork() override;
virtual Status ContinueBackgroundWork() override;
virtual Status EnableAutoCompaction(
const std::vector<ColumnFamilyHandle*>& column_family_handles) override;
virtual void EnableManualCompaction() override;
virtual void DisableManualCompaction() override;
using DB::SetOptions;
Status SetOptions(
ColumnFamilyHandle* column_family,
const std::unordered_map<std::string, std::string>& options_map) override;
virtual Status SetDBOptions(
const std::unordered_map<std::string, std::string>& options_map) override;
using DB::NumberLevels;
virtual int NumberLevels(ColumnFamilyHandle* column_family) override;
using DB::MaxMemCompactionLevel;
virtual int MaxMemCompactionLevel(ColumnFamilyHandle* column_family) override;
using DB::Level0StopWriteTrigger;
virtual int Level0StopWriteTrigger(
ColumnFamilyHandle* column_family) override;
virtual const std::string& GetName() const override;
virtual Env* GetEnv() const override;
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
2019-12-13 23:47:08 +01:00
virtual FileSystem* GetFileSystem() const override;
using DB::GetOptions;
virtual Options GetOptions(ColumnFamilyHandle* column_family) const override;
using DB::GetDBOptions;
virtual DBOptions GetDBOptions() const override;
using DB::Flush;
virtual Status Flush(const FlushOptions& options,
ColumnFamilyHandle* column_family) override;
virtual Status Flush(
const FlushOptions& options,
const std::vector<ColumnFamilyHandle*>& column_families) override;
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
virtual Status FlushWAL(bool sync) override;
Expose DB methods to lock and unlock the WAL (#5146) Summary: Expose DB methods to lock and unlock the WAL. These methods are intended to use by MyRocks in order to obtain WAL coordinates in consistent way. Usage scenario is following: MySQL has performance_schema.log_status which provides information that enables a backup tool to copy the required log files without locking for the duration of copy. To populate this table MySQL does following: 1. Lock the binary log. Transactions are not allowed to commit now 2. Save the binary log coordinates 3. Walk through the storage engines and lock writes on each engine. For InnoDB, redo log is locked. For MyRocks, WAL should be locked. 4. Ask storage engines for their coordinates. InnoDB reports its current LSN and checkpoint LSN. MyRocks should report active WAL files names and sizes. 5. Release storage engine's locks 6. Unlock binary log Backup tool will then use this information to copy InnoDB, RocksDB and MySQL binary logs up to specified positions to end up with consistent DB state after restore. Currently, RocksDB allows to obtain the list of WAL files. Only missing bit is the method to lock the writes to WAL files. LockWAL method must flush the WAL in order for the reported size to be accurate (GetSortedWALFiles is using file system stat call to return the file size), also, since backup tool is going to copy the WAL, it is better to be flushed. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5146 Differential Revision: D14815447 Pulled By: maysamyabandeh fbshipit-source-id: eec9535a6025229ed471119f19fe7b3d8ae888a3
2019-04-06 15:36:42 +02:00
bool TEST_WALBufferIsEmpty(bool lock = true);
virtual Status SyncWAL() override;
Expose DB methods to lock and unlock the WAL (#5146) Summary: Expose DB methods to lock and unlock the WAL. These methods are intended to use by MyRocks in order to obtain WAL coordinates in consistent way. Usage scenario is following: MySQL has performance_schema.log_status which provides information that enables a backup tool to copy the required log files without locking for the duration of copy. To populate this table MySQL does following: 1. Lock the binary log. Transactions are not allowed to commit now 2. Save the binary log coordinates 3. Walk through the storage engines and lock writes on each engine. For InnoDB, redo log is locked. For MyRocks, WAL should be locked. 4. Ask storage engines for their coordinates. InnoDB reports its current LSN and checkpoint LSN. MyRocks should report active WAL files names and sizes. 5. Release storage engine's locks 6. Unlock binary log Backup tool will then use this information to copy InnoDB, RocksDB and MySQL binary logs up to specified positions to end up with consistent DB state after restore. Currently, RocksDB allows to obtain the list of WAL files. Only missing bit is the method to lock the writes to WAL files. LockWAL method must flush the WAL in order for the reported size to be accurate (GetSortedWALFiles is using file system stat call to return the file size), also, since backup tool is going to copy the WAL, it is better to be flushed. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5146 Differential Revision: D14815447 Pulled By: maysamyabandeh fbshipit-source-id: eec9535a6025229ed471119f19fe7b3d8ae888a3
2019-04-06 15:36:42 +02:00
virtual Status LockWAL() override;
virtual Status UnlockWAL() override;
virtual SequenceNumber GetLatestSequenceNumber() const override;
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
2017-11-02 02:43:29 +01:00
virtual bool SetPreserveDeletesSequenceNumber(SequenceNumber seqnum) override;
virtual Status GetDbIdentity(std::string& identity) const override;
virtual Status GetDbIdentityFromIdentityFile(std::string* identity) const;
virtual Status GetDbSessionId(std::string& session_id) const override;
ColumnFamilyHandle* DefaultColumnFamily() const override;
ColumnFamilyHandle* PersistentStatsColumnFamily() const;
virtual Status Close() override;
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
2020-06-25 04:05:47 +02:00
virtual Status DisableFileDeletions() override;
virtual Status EnableFileDeletions(bool force) override;
virtual bool IsFileDeletionsEnabled() const;
Status GetStatsHistory(
uint64_t start_time, uint64_t end_time,
std::unique_ptr<StatsHistoryIterator>* stats_iterator) override;
#ifndef ROCKSDB_LITE
using DB::ResetStats;
virtual Status ResetStats() override;
// All the returned filenames start with "/"
virtual Status GetLiveFiles(std::vector<std::string>&,
uint64_t* manifest_file_size,
bool flush_memtable = true) override;
virtual Status GetSortedWalFiles(VectorLogPtr& files) override;
virtual Status GetCurrentWalFile(
std::unique_ptr<LogFile>* current_log_file) override;
virtual Status GetCreationTimeOfOldestFile(
uint64_t* creation_time) override;
virtual Status GetUpdatesSince(
SequenceNumber seq_number, std::unique_ptr<TransactionLogIterator>* iter,
const TransactionLogIterator::ReadOptions& read_options =
TransactionLogIterator::ReadOptions()) override;
virtual Status DeleteFile(std::string name) override;
Status DeleteFilesInRanges(ColumnFamilyHandle* column_family,
const RangePtr* ranges, size_t n,
bool include_end = true);
virtual void GetLiveFilesMetaData(
std::vector<LiveFileMetaData>* metadata) override;
virtual Status GetLiveFilesChecksumInfo(
FileChecksumList* checksum_list) override;
// Obtains the meta data of the specified column family of the DB.
// Status::NotFound() will be returned if the current DB does not have
// any column family match the specified name.
// TODO(yhchiang): output parameter is placed in the end in this codebase.
virtual void GetColumnFamilyMetaData(ColumnFamilyHandle* column_family,
ColumnFamilyMetaData* metadata) override;
Status SuggestCompactRange(ColumnFamilyHandle* column_family,
const Slice* begin, const Slice* end) override;
Status PromoteL0(ColumnFamilyHandle* column_family,
int target_level) override;
using DB::IngestExternalFile;
virtual Status IngestExternalFile(
ColumnFamilyHandle* column_family,
const std::vector<std::string>& external_files,
const IngestExternalFileOptions& ingestion_options) override;
using DB::IngestExternalFiles;
virtual Status IngestExternalFiles(
const std::vector<IngestExternalFileArg>& args) override;
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
2019-07-17 21:22:21 +02:00
using DB::CreateColumnFamilyWithImport;
virtual Status CreateColumnFamilyWithImport(
const ColumnFamilyOptions& options, const std::string& column_family_name,
const ImportColumnFamilyOptions& import_options,
const ExportImportFilesMetaData& metadata,
ColumnFamilyHandle** handle) override;
using DB::VerifyChecksum;
virtual Status VerifyChecksum(const ReadOptions& /*read_options*/) override;
using DB::StartTrace;
virtual Status StartTrace(
const TraceOptions& options,
std::unique_ptr<TraceWriter>&& trace_writer) override;
using DB::EndTrace;
virtual Status EndTrace() override;
using DB::StartBlockCacheTrace;
Status StartBlockCacheTrace(
const TraceOptions& options,
std::unique_ptr<TraceWriter>&& trace_writer) override;
using DB::EndBlockCacheTrace;
Status EndBlockCacheTrace() override;
using DB::StartIOTrace;
Status StartIOTrace(Env* env, const TraceOptions& options,
std::unique_ptr<TraceWriter>&& trace_writer) override;
using DB::EndIOTrace;
Status EndIOTrace() override;
using DB::GetPropertiesOfAllTables;
virtual Status GetPropertiesOfAllTables(
ColumnFamilyHandle* column_family,
TablePropertiesCollection* props) override;
virtual Status GetPropertiesOfTablesInRange(
ColumnFamilyHandle* column_family, const Range* range, std::size_t n,
TablePropertiesCollection* props) override;
#endif // ROCKSDB_LITE
// ---- End of implementations of the DB interface ----
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
2019-08-06 23:22:34 +02:00
struct GetImplOptions {
ColumnFamilyHandle* column_family = nullptr;
PinnableSlice* value = nullptr;
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
2020-03-03 00:58:32 +01:00
std::string* timestamp = nullptr;
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
2019-08-06 23:22:34 +02:00
bool* value_found = nullptr;
ReadCallback* callback = nullptr;
bool* is_blob_index = nullptr;
// If true return value associated with key via value pointer else return
// all merge operands for key via merge_operands pointer
bool get_value = true;
// Pointer to an array of size
// get_merge_operands_options.expected_max_number_of_operands allocated by
// user
PinnableSlice* merge_operands = nullptr;
GetMergeOperandsOptions* get_merge_operands_options = nullptr;
int* number_of_operands = nullptr;
};
// Function that Get and KeyMayExist call with no_io true or false
// Note: 'value_found' from KeyMayExist propagates here
New API to get all merge operands for a Key (#5604) Summary: This is a new API added to db.h to allow for fetching all merge operands associated with a Key. The main motivation for this API is to support use cases where doing a full online merge is not necessary as it is performance sensitive. Example use-cases: 1. Update subset of columns and read subset of columns - Imagine a SQL Table, a row is encoded as a K/V pair (as it is done in MyRocks). If there are many columns and users only updated one of them, we can use merge operator to reduce write amplification. While users only read one or two columns in the read query, this feature can avoid a full merging of the whole row, and save some CPU. 2. Updating very few attributes in a value which is a JSON-like document - Updating one attribute can be done efficiently using merge operator, while reading back one attribute can be done more efficiently if we don't need to do a full merge. ---------------------------------------------------------------------------------------------------- API : Status GetMergeOperands( const ReadOptions& options, ColumnFamilyHandle* column_family, const Slice& key, PinnableSlice* merge_operands, GetMergeOperandsOptions* get_merge_operands_options, int* number_of_operands) Example usage : int size = 100; int number_of_operands = 0; std::vector<PinnableSlice> values(size); GetMergeOperandsOptions merge_operands_info; db_->GetMergeOperands(ReadOptions(), db_->DefaultColumnFamily(), "k1", values.data(), merge_operands_info, &number_of_operands); Description : Returns all the merge operands corresponding to the key. If the number of merge operands in DB is greater than merge_operands_options.expected_max_number_of_operands no merge operands are returned and status is Incomplete. Merge operands returned are in the order of insertion. merge_operands-> Points to an array of at-least merge_operands_options.expected_max_number_of_operands and the caller is responsible for allocating it. If the status returned is Incomplete then number_of_operands will contain the total number of merge operands found in DB for key. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5604 Test Plan: Added unit test and perf test in db_bench that can be run using the command: ./db_bench -benchmarks=getmergeoperands --merge_operator=sortlist Differential Revision: D16657366 Pulled By: vjnadimpalli fbshipit-source-id: 0faadd752351745224ee12d4ae9ef3cb529951bf
2019-08-06 23:22:34 +02:00
// This function is also called by GetMergeOperands
// If get_impl_options.get_value = true get value associated with
// get_impl_options.key via get_impl_options.value
// If get_impl_options.get_value = false get merge operands associated with
// get_impl_options.key via get_impl_options.merge_operands
Status GetImpl(const ReadOptions& options, const Slice& key,
return timestamp from get (#6409) Summary: Added new Get() methods that return timestamp. Dummy implementation is given so that classes derived from DB don't need to be touched to provide their implementation. MultiGet is not included. ReadRandom perf test (10 minutes) on the same development machine ram drive with the same DB data shows no regression (within marge of error). The test is adapted from https://github.com/facebook/rocksdb/wiki/RocksDB-In-Memory-Workload-Performance-Benchmarks. base line (commit 72ee067b9): 101.712 micros/op 314602 ops/sec; 36.0 MB/s (5658999 of 5658999 found) This PR: 100.288 micros/op 319071 ops/sec; 36.5 MB/s (5674999 of 5674999 found) ./db_bench --db=r:\rocksdb.github --num_levels=6 --key_size=20 --prefix_size=20 --keys_per_prefix=0 --value_size=100 --cache_size=2147483648 --cache_numshardbits=6 --compression_type=none --compression_ratio=1 --min_level_to_compress=-1 --disable_seek_compaction=1 --hard_rate_limit=2 --write_buffer_size=134217728 --max_write_buffer_number=2 --level0_file_num_compaction_trigger=8 --target_file_size_base=134217728 --max_bytes_for_level_base=1073741824 --disable_wal=0 --wal_dir=r:\rocksdb.github\WAL_LOG --sync=0 --verify_checksum=1 --delete_obsolete_files_period_micros=314572800 --max_background_compactions=4 --max_background_flushes=0 --level0_slowdown_writes_trigger=16 --level0_stop_writes_trigger=24 --statistics=0 --stats_per_interval=0 --stats_interval=1048576 --histogram=0 --use_plain_table=1 --open_files=-1 --mmap_read=1 --mmap_write=0 --memtablerep=prefix_hash --bloom_bits=10 --bloom_locality=1 --duration=600 --benchmarks=readrandom --use_existing_db=1 --num=25000000 --threads=32 Pull Request resolved: https://github.com/facebook/rocksdb/pull/6409 Differential Revision: D20200086 Pulled By: riversand963 fbshipit-source-id: 490edd74d924f62bd8ae9c29c2a6bbbb8410ca50
2020-03-03 00:58:32 +01:00
GetImplOptions& get_impl_options);
// If `snapshot` == kMaxSequenceNumber, set a recent one inside the file.
ArenaWrappedDBIter* NewIteratorImpl(const ReadOptions& options,
ColumnFamilyData* cfd,
SequenceNumber snapshot,
ReadCallback* read_callback,
bool allow_blob = false,
bool allow_refresh = true);
virtual SequenceNumber GetLastPublishedSequence() const {
if (last_seq_same_as_publish_seq_) {
return versions_->LastSequence();
} else {
return versions_->LastPublishedSequence();
}
}
// REQUIRES: joined the main write queue if two_write_queues is disabled, and
// the second write queue otherwise.
virtual void SetLastPublishedSequence(SequenceNumber seq);
// Returns LastSequence in last_seq_same_as_publish_seq_
// mode and LastAllocatedSequence otherwise. This is useful when visiblility
// depends also on data written to the WAL but not to the memtable.
SequenceNumber TEST_GetLastVisibleSequence() const;
#ifndef ROCKSDB_LITE
// Similar to Write() but will call the callback once on the single write
// thread to determine whether it is safe to perform the write.
virtual Status WriteWithCallback(const WriteOptions& write_options,
WriteBatch* my_batch,
WriteCallback* callback);
// Returns the sequence number that is guaranteed to be smaller than or equal
// to the sequence number of any key that could be inserted into the current
// memtables. It can then be assumed that any write with a larger(or equal)
// sequence number will be present in this memtable or a later memtable.
//
// If the earliest sequence number could not be determined,
// kMaxSequenceNumber will be returned.
//
// If include_history=true, will also search Memtables in MemTableList
// History.
SequenceNumber GetEarliestMemTableSequenceNumber(SuperVersion* sv,
bool include_history);
// For a given key, check to see if there are any records for this key
// in the memtables, including memtable history. If cache_only is false,
// SST files will also be checked.
//
// If a key is found, *found_record_for_key will be set to true and
// *seq will be set to the stored sequence number for the latest
// operation on this key or kMaxSequenceNumber if unknown.
// If no key is found, *found_record_for_key will be set to false.
//
// Note: If cache_only=false, it is possible for *seq to be set to 0 if
// the sequence number has been cleared from the record. If the caller is
// holding an active db snapshot, we know the missing sequence must be less
// than the snapshot's sequence number (sequence numbers are only cleared
// when there are no earlier active snapshots).
//
// If NotFound is returned and found_record_for_key is set to false, then no
// record for this key was found. If the caller is holding an active db
// snapshot, we know that no key could have existing after this snapshot
// (since we do not compact keys that have an earlier snapshot).
//
// Only records newer than or at `lower_bound_seq` are guaranteed to be
// returned. Memtables and files may not be checked if it only contains data
// older than `lower_bound_seq`.
//
// Returns OK or NotFound on success,
// other status on unexpected error.
// TODO(andrewkr): this API need to be aware of range deletion operations
Status GetLatestSequenceForKey(SuperVersion* sv, const Slice& key,
bool cache_only,
SequenceNumber lower_bound_seq,
SequenceNumber* seq,
bool* found_record_for_key,
bool* is_blob_index = nullptr);
Status TraceIteratorSeek(const uint32_t& cf_id, const Slice& key);
Status TraceIteratorSeekForPrev(const uint32_t& cf_id, const Slice& key);
#endif // ROCKSDB_LITE
// Similar to GetSnapshot(), but also lets the db know that this snapshot
// will be used for transaction write-conflict checking. The DB can then
// make sure not to compact any keys that would prevent a write-conflict from
// being detected.
const Snapshot* GetSnapshotForWriteConflictBoundary();
// checks if all live files exist on file system and that their file sizes
// match to our in-memory records
virtual Status CheckConsistency();
// max_file_num_to_ignore allows bottom level compaction to filter out newly
// compacted SST files. Setting max_file_num_to_ignore to kMaxUint64 will
// disable the filtering
Status 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);
// Return an internal iterator over the current state of the database.
// The keys of this iterator are internal keys (see format.h).
// The returned iterator should be deleted when no longer needed.
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
// If allow_unprepared_value is true, the returned iterator may defer reading
// the value and so will require PrepareValue() to be called before value();
// allow_unprepared_value = false is convenient when this optimization is not
// useful, e.g. when reading the whole column family.
// @param read_options Must outlive the returned iterator.
InternalIterator* NewInternalIterator(
const ReadOptions& read_options, Arena* arena,
RangeDelAggregator* range_del_agg, SequenceNumber sequence,
Properly report IO errors when IndexType::kBinarySearchWithFirstKey is used (#6621) Summary: Context: Index type `kBinarySearchWithFirstKey` added the ability for sst file iterator to sometimes report a key from index without reading the corresponding data block. This is useful when sst blocks are cut at some meaningful boundaries (e.g. one block per key prefix), and many seeks land between blocks (e.g. for each prefix, the ranges of keys in different sst files are nearly disjoint, so a typical seek needs to read a data block from only one file even if all files have the prefix). But this added a new error condition, which rocksdb code was really not equipped to deal with: `InternalIterator::value()` may fail with an IO error or Status::Incomplete, but it's just a method returning a Slice, with no way to report error instead. Before this PR, this type of error wasn't handled at all (an empty slice was returned), and kBinarySearchWithFirstKey implementation was considered a prototype. Now that we (LogDevice) have experimented with kBinarySearchWithFirstKey for a while and confirmed that it's really useful, this PR is adding the missing error handling. It's a pretty inconvenient situation implementation-wise. The error needs to be reported from InternalIterator when trying to access value. But there are ~700 call sites of `InternalIterator::value()`, most of which either can't hit the error condition (because the iterator is reading from memtable or from index or something) or wouldn't benefit from the deferred loading of the value (e.g. compaction iterator that reads all values anyway). Adding error handling to all these call sites would needlessly bloat the code. So instead I made the deferred value loading optional: only the call sites that may use deferred loading have to call the new method `PrepareValue()` before calling `value()`. The feature is enabled with a new bool argument `allow_unprepared_value` to a bunch of methods that create iterators (it wouldn't make sense to put it in ReadOptions because it's completely internal to iterators, with virtually no user-visible effect). Lmk if you have better ideas. Note that the deferred value loading only happens for *internal* iterators. The user-visible iterator (DBIter) always prepares the value before returning from Seek/Next/etc. We could go further and add an API to defer that value loading too, but that's most likely not useful for LogDevice, so it doesn't seem worth the complexity for now. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6621 Test Plan: make -j5 check . Will also deploy to some logdevice test clusters and look at stats. Reviewed By: siying Differential Revision: D20786930 Pulled By: al13n321 fbshipit-source-id: 6da77d918bad3780522e918f17f4d5513d3e99ee
2020-04-16 02:37:23 +02:00
ColumnFamilyHandle* column_family = nullptr,
bool allow_unprepared_value = false);
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
LogsWithPrepTracker* logs_with_prep_tracker() {
return &logs_with_prep_tracker_;
}
struct BGJobLimits {
int max_flushes;
int max_compactions;
};
// Returns maximum background flushes and compactions allowed to be scheduled
BGJobLimits GetBGJobLimits() const;
// Need a static version that can be called during SanitizeOptions().
static BGJobLimits GetBGJobLimits(int max_background_flushes,
int max_background_compactions,
int max_background_jobs,
bool parallelize_compactions);
// move logs pending closing from job_context to the DB queue and
// schedule a purge
void ScheduleBgLogWriterClose(JobContext* job_context);
uint64_t MinLogNumberToKeep();
// Returns the lower bound file number for SSTs that won't be deleted, even if
// they're obsolete. This lower bound is used internally to prevent newly
// created flush/compaction output files from being deleted before they're
// installed. This technique avoids the need for tracking the exact numbers of
// files pending creation, although it prevents more files than necessary from
// being deleted.
uint64_t MinObsoleteSstNumberToKeep();
// Returns the list of live files in 'live' and the list
// of all files in the filesystem in 'candidate_files'.
// If force == false and the last call was less than
// db_options_.delete_obsolete_files_period_micros microseconds ago,
// it will not fill up the job_context
void FindObsoleteFiles(JobContext* job_context, bool force,
bool no_full_scan = false);
// Diffs the files listed in filenames and those that do not
// belong to live files are possibly removed. Also, removes all the
// files in sst_delete_files and log_delete_files.
// It is not necessary to hold the mutex when invoking this method.
// If FindObsoleteFiles() was run, we need to also run
// PurgeObsoleteFiles(), even if disable_delete_obsolete_files_ is true
void PurgeObsoleteFiles(JobContext& background_contet,
bool schedule_only = false);
// Schedule a background job to actually delete obsolete files.
void SchedulePurge();
const SnapshotList& snapshots() const { return snapshots_; }
// load list of snapshots to `snap_vector` that is no newer than `max_seq`
// in ascending order.
// `oldest_write_conflict_snapshot` is filled with the oldest snapshot
// which satisfies SnapshotImpl.is_write_conflict_boundary_ = true.
void LoadSnapshots(std::vector<SequenceNumber>* snap_vector,
SequenceNumber* oldest_write_conflict_snapshot,
const SequenceNumber& max_seq) const {
InstrumentedMutexLock l(mutex());
snapshots().GetAll(snap_vector, oldest_write_conflict_snapshot, max_seq);
}
const ImmutableDBOptions& immutable_db_options() const {
return immutable_db_options_;
}
// Cancel all background jobs, including flush, compaction, background
// purging, stats dumping threads, etc. If `wait` = true, wait for the
// running jobs to abort or finish before returning. Otherwise, only
// sends the signals.
void CancelAllBackgroundWork(bool wait);
// Find Super version and reference it. Based on options, it might return
// the thread local cached one.
// Call ReturnAndCleanupSuperVersion() when it is no longer needed.
SuperVersion* GetAndRefSuperVersion(ColumnFamilyData* cfd);
// Similar to the previous function but looks up based on a column family id.
// nullptr will be returned if this column family no longer exists.
// REQUIRED: this function should only be called on the write thread or if the
// mutex is held.
SuperVersion* GetAndRefSuperVersion(uint32_t column_family_id);
// Un-reference the super version and clean it up if it is the last reference.
void CleanupSuperVersion(SuperVersion* sv);
// Un-reference the super version and return it to thread local cache if
// needed. If it is the last reference of the super version. Clean it up
// after un-referencing it.
void ReturnAndCleanupSuperVersion(ColumnFamilyData* cfd, SuperVersion* sv);
// Similar to the previous function but looks up based on a column family id.
// nullptr will be returned if this column family no longer exists.
// REQUIRED: this function should only be called on the write thread.
void ReturnAndCleanupSuperVersion(uint32_t colun_family_id, SuperVersion* sv);
// REQUIRED: this function should only be called on the write thread or if the
// mutex is held. Return value only valid until next call to this function or
// mutex is released.
ColumnFamilyHandle* GetColumnFamilyHandle(uint32_t column_family_id);
// Same as above, should called without mutex held and not on write thread.
std::unique_ptr<ColumnFamilyHandle> GetColumnFamilyHandleUnlocked(
uint32_t column_family_id);
// Returns the number of currently running flushes.
// REQUIREMENT: mutex_ must be held when calling this function.
int num_running_flushes() {
mutex_.AssertHeld();
return num_running_flushes_;
}
// Returns the number of currently running compactions.
// REQUIREMENT: mutex_ must be held when calling this function.
int num_running_compactions() {
mutex_.AssertHeld();
return num_running_compactions_;
}
const WriteController& write_controller() { return write_controller_; }
// @param read_options Must outlive the returned iterator.
InternalIterator* NewInternalIterator(const ReadOptions& read_options,
ColumnFamilyData* cfd,
SuperVersion* super_version,
Arena* arena,
RangeDelAggregator* range_del_agg,
SequenceNumber sequence,
bool allow_unprepared_value);
// hollow transactions shell used for recovery.
// these will then be passed to TransactionDB so that
// locks can be reacquired before writing can resume.
struct RecoveredTransaction {
std::string name_;
bool unprepared_;
struct BatchInfo {
uint64_t log_number_;
// TODO(lth): For unprepared, the memory usage here can be big for
// unprepared transactions. This is only useful for rollbacks, and we
// can in theory just keep keyset for that.
WriteBatch* batch_;
// Number of sub-batches. A new sub-batch is created if txn attempts to
// insert a duplicate key,seq to memtable. This is currently used in
// WritePreparedTxn/WriteUnpreparedTxn.
size_t batch_cnt_;
};
// This maps the seq of the first key in the batch to BatchInfo, which
// contains WriteBatch and other information relevant to the batch.
//
// For WriteUnprepared, batches_ can have size greater than 1, but for
// other write policies, it must be of size 1.
std::map<SequenceNumber, BatchInfo> batches_;
explicit RecoveredTransaction(const uint64_t log, const std::string& name,
WriteBatch* batch, SequenceNumber seq,
size_t batch_cnt, bool unprepared)
: name_(name), unprepared_(unprepared) {
batches_[seq] = {log, batch, batch_cnt};
}
~RecoveredTransaction() {
for (auto& it : batches_) {
delete it.second.batch_;
}
}
void AddBatch(SequenceNumber seq, uint64_t log_number, WriteBatch* batch,
size_t batch_cnt, bool unprepared) {
assert(batches_.count(seq) == 0);
batches_[seq] = {log_number, batch, batch_cnt};
// Prior state must be unprepared, since the prepare batch must be the
// last batch.
assert(unprepared_);
unprepared_ = unprepared;
}
};
bool allow_2pc() const { return immutable_db_options_.allow_2pc; }
std::unordered_map<std::string, RecoveredTransaction*>
recovered_transactions() {
return recovered_transactions_;
}
RecoveredTransaction* GetRecoveredTransaction(const std::string& name) {
auto it = recovered_transactions_.find(name);
if (it == recovered_transactions_.end()) {
return nullptr;
} else {
return it->second;
}
}
void InsertRecoveredTransaction(const uint64_t log, const std::string& name,
WriteBatch* batch, SequenceNumber seq,
size_t batch_cnt, bool unprepared_batch) {
// For WriteUnpreparedTxn, InsertRecoveredTransaction is called multiple
// times for every unprepared batch encountered during recovery.
//
// If the transaction is prepared, then the last call to
// InsertRecoveredTransaction will have unprepared_batch = false.
auto rtxn = recovered_transactions_.find(name);
if (rtxn == recovered_transactions_.end()) {
recovered_transactions_[name] = new RecoveredTransaction(
log, name, batch, seq, batch_cnt, unprepared_batch);
} else {
rtxn->second->AddBatch(seq, log, batch, batch_cnt, unprepared_batch);
}
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
logs_with_prep_tracker_.MarkLogAsContainingPrepSection(log);
}
void DeleteRecoveredTransaction(const std::string& name) {
auto it = recovered_transactions_.find(name);
assert(it != recovered_transactions_.end());
auto* trx = it->second;
recovered_transactions_.erase(it);
for (const auto& info : trx->batches_) {
logs_with_prep_tracker_.MarkLogAsHavingPrepSectionFlushed(
info.second.log_number_);
}
delete trx;
}
void DeleteAllRecoveredTransactions() {
for (auto it = recovered_transactions_.begin();
it != recovered_transactions_.end(); ++it) {
delete it->second;
}
recovered_transactions_.clear();
}
void AddToLogsToFreeQueue(log::Writer* log_writer) {
logs_to_free_queue_.push_back(log_writer);
}
void AddSuperVersionsToFreeQueue(SuperVersion* sv) {
superversions_to_free_queue_.push_back(sv);
}
void SetSnapshotChecker(SnapshotChecker* snapshot_checker);
// Fill JobContext with snapshot information needed by flush and compaction.
void GetSnapshotContext(JobContext* job_context,
std::vector<SequenceNumber>* snapshot_seqs,
SequenceNumber* earliest_write_conflict_snapshot,
SnapshotChecker** snapshot_checker);
// Not thread-safe.
void SetRecoverableStatePreReleaseCallback(PreReleaseCallback* callback);
InstrumentedMutex* mutex() const { return &mutex_; }
// Initialize a brand new DB. The DB directory is expected to be empty before
// calling it. Push new manifest file name into `new_filenames`.
Status NewDB(std::vector<std::string>* new_filenames);
// This is to be used only by internal rocksdb classes.
static Status Open(const DBOptions& db_options, const std::string& name,
const std::vector<ColumnFamilyDescriptor>& column_families,
std::vector<ColumnFamilyHandle*>* handles, DB** dbptr,
const bool seq_per_batch, const bool batch_per_txn);
static IOStatus CreateAndNewDirectory(
FileSystem* fs, const std::string& dirname,
std::unique_ptr<FSDirectory>* directory);
// find stats map from stats_history_ with smallest timestamp in
// the range of [start_time, end_time)
bool FindStatsByTime(uint64_t start_time, uint64_t end_time,
uint64_t* new_time,
std::map<std::string, uint64_t>* stats_map);
// Print information of all tombstones of all iterators to the std::string
// This is only used by ldb. The output might be capped. Tombstones
// printed out are not guaranteed to be in any order.
Status TablesRangeTombstoneSummary(ColumnFamilyHandle* column_family,
int max_entries_to_print,
std::string* out_str);
#ifndef NDEBUG
// Compact any files in the named level that overlap [*begin, *end]
Status TEST_CompactRange(int level, const Slice* begin, const Slice* end,
ColumnFamilyHandle* column_family = nullptr,
bool disallow_trivial_move = false);
void TEST_SwitchWAL();
bool TEST_UnableToReleaseOldestLog() { return unable_to_release_oldest_log_; }
bool TEST_IsLogGettingFlushed() {
return alive_log_files_.begin()->getting_flushed;
}
Status TEST_SwitchMemtable(ColumnFamilyData* cfd = nullptr);
// Force current memtable contents to be flushed.
Status TEST_FlushMemTable(bool wait = true, bool allow_write_stall = false,
ColumnFamilyHandle* cfh = nullptr);
Status TEST_FlushMemTable(ColumnFamilyData* cfd,
const FlushOptions& flush_opts);
// Flush (multiple) ColumnFamilyData without using ColumnFamilyHandle. This
// is because in certain cases, we can flush column families, wait for the
// flush to complete, but delete the column family handle before the wait
// finishes. For example in CompactRange.
Status TEST_AtomicFlushMemTables(const autovector<ColumnFamilyData*>& cfds,
const FlushOptions& flush_opts);
// Wait for memtable compaction
Status TEST_WaitForFlushMemTable(ColumnFamilyHandle* column_family = nullptr);
// Wait for any compaction
// We add a bool parameter to wait for unscheduledCompactions_ == 0, but this
// is only for the special test of CancelledCompactions
Status TEST_WaitForCompact(bool waitUnscheduled = false);
// Return the maximum overlapping data (in bytes) at next level for any
// file at a level >= 1.
int64_t TEST_MaxNextLevelOverlappingBytes(
ColumnFamilyHandle* column_family = nullptr);
// Return the current manifest file no.
uint64_t TEST_Current_Manifest_FileNo();
// Returns the number that'll be assigned to the next file that's created.
uint64_t TEST_Current_Next_FileNo();
// get total level0 file size. Only for testing.
uint64_t TEST_GetLevel0TotalSize();
void TEST_GetFilesMetaData(ColumnFamilyHandle* column_family,
std::vector<std::vector<FileMetaData>>* metadata);
void TEST_LockMutex();
void TEST_UnlockMutex();
// REQUIRES: mutex locked
void* TEST_BeginWrite();
// REQUIRES: mutex locked
// pass the pointer that you got from TEST_BeginWrite()
void TEST_EndWrite(void* w);
uint64_t TEST_MaxTotalInMemoryState() const {
return max_total_in_memory_state_;
}
size_t TEST_LogsToFreeSize();
uint64_t TEST_LogfileNumber();
uint64_t TEST_total_log_size() const { return total_log_size_; }
// Returns column family name to ImmutableCFOptions map.
Status TEST_GetAllImmutableCFOptions(
std::unordered_map<std::string, const ImmutableCFOptions*>* iopts_map);
// Return the lastest MutableCFOptions of a column family
Status TEST_GetLatestMutableCFOptions(ColumnFamilyHandle* column_family,
MutableCFOptions* mutable_cf_options);
Cache* TEST_table_cache() { return table_cache_.get(); }
WriteController& TEST_write_controler() { return write_controller_; }
uint64_t TEST_FindMinLogContainingOutstandingPrep();
uint64_t TEST_FindMinPrepLogReferencedByMemTable();
size_t TEST_PreparedSectionCompletedSize();
size_t TEST_LogsWithPrepSize();
int TEST_BGCompactionsAllowed() const;
int TEST_BGFlushesAllowed() const;
size_t TEST_GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;
void TEST_WaitForStatsDumpRun(std::function<void()> callback) const;
size_t TEST_EstimateInMemoryStatsHistorySize() const;
VersionSet* TEST_GetVersionSet() const { return versions_.get(); }
const std::unordered_set<uint64_t>& TEST_GetFilesGrabbedForPurge() const {
return files_grabbed_for_purge_;
}
#ifndef ROCKSDB_LITE
PeriodicWorkTestScheduler* TEST_GetPeriodicWorkScheduler() const;
#endif // !ROCKSDB_LITE
#endif // NDEBUG
// persist stats to column family "_persistent_stats"
void PersistStats();
// dump rocksdb.stats to LOG
void DumpStats();
// flush LOG out of application buffer
void FlushInfoLog();
protected:
const std::string dbname_;
std::string db_id_;
// db_session_id_ is an identifier that gets reset
// every time the DB is opened
std::string db_session_id_;
std::unique_ptr<VersionSet> versions_;
// Flag to check whether we allocated and own the info log file
bool own_info_log_;
const DBOptions initial_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
2019-12-13 23:47:08 +01:00
Env* const env_;
std::shared_ptr<IOTracer> io_tracer_;
const ImmutableDBOptions immutable_db_options_;
FileSystemPtr fs_;
MutableDBOptions mutable_db_options_;
Statistics* stats_;
std::unordered_map<std::string, RecoveredTransaction*>
recovered_transactions_;
std::unique_ptr<Tracer> tracer_;
InstrumentedMutex trace_mutex_;
BlockCacheTracer block_cache_tracer_;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
// State below is protected by mutex_
// With two_write_queues enabled, some of the variables that accessed during
// WriteToWAL need different synchronization: log_empty_, alive_log_files_,
// logs_, logfile_number_. Refer to the definition of each variable below for
// more description.
mutable InstrumentedMutex mutex_;
ColumnFamilyHandleImpl* default_cf_handle_;
InternalStats* default_cf_internal_stats_;
// only used for dynamically adjusting max_total_wal_size. it is a sum of
// [write_buffer_size * max_write_buffer_number] over all column families
uint64_t max_total_in_memory_state_;
// If true, we have only one (default) column family. We use this to optimize
// some code-paths
bool single_column_family_mode_;
// The options to access storage files
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
2019-12-13 23:47:08 +01:00
const FileOptions file_options_;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
// Additonal options for compaction and flush
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
2019-12-13 23:47:08 +01:00
FileOptions file_options_for_compaction_;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
std::unique_ptr<ColumnFamilyMemTablesImpl> column_family_memtables_;
// Increase the sequence number after writing each batch, whether memtable is
// disabled for that or not. Otherwise the sequence number is increased after
// writing each key into memtable. This implies that when disable_memtable is
// set, the seq is not increased at all.
//
// Default: false
const bool seq_per_batch_;
// This determines during recovery whether we expect one writebatch per
// recovered transaction, or potentially multiple writebatches per
// transaction. For WriteUnprepared, this is set to false, since multiple
// batches can exist per transaction.
//
// Default: true
const bool batch_per_txn_;
// Except in DB::Open(), WriteOptionsFile can only be called when:
// Persist options to options file.
// If need_mutex_lock = false, the method will lock DB mutex.
// If need_enter_write_thread = false, the method will enter write thread.
Status WriteOptionsFile(bool need_mutex_lock, bool need_enter_write_thread);
// The following two functions can only be called when:
// 1. WriteThread::Writer::EnterUnbatched() is used.
// 2. db_mutex is NOT held
Status RenameTempFileToOptionsFile(const std::string& file_name);
Status DeleteObsoleteOptionsFiles();
void NotifyOnFlushBegin(ColumnFamilyData* cfd, FileMetaData* file_meta,
const MutableCFOptions& mutable_cf_options,
int job_id);
void NotifyOnFlushCompleted(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
std::list<std::unique_ptr<FlushJobInfo>>* flush_jobs_info);
void NotifyOnCompactionBegin(ColumnFamilyData* cfd, Compaction* c,
const Status& st,
const CompactionJobStats& job_stats, int job_id);
void NotifyOnCompactionCompleted(ColumnFamilyData* cfd, Compaction* c,
const Status& st,
const CompactionJobStats& job_stats,
int job_id);
void NotifyOnMemTableSealed(ColumnFamilyData* cfd,
const MemTableInfo& mem_table_info);
#ifndef ROCKSDB_LITE
void NotifyOnExternalFileIngested(
ColumnFamilyData* cfd, const ExternalSstFileIngestionJob& ingestion_job);
#endif // !ROCKSDB_LITE
void NewThreadStatusCfInfo(ColumnFamilyData* cfd) const;
void EraseThreadStatusCfInfo(ColumnFamilyData* cfd) const;
void EraseThreadStatusDbInfo() const;
// If disable_memtable is set the application logic must guarantee that the
// batch will still be skipped from memtable during the recovery. An excption
// to this is seq_per_batch_ mode, in which since each batch already takes one
// seq, it is ok for the batch to write to memtable during recovery as long as
// it only takes one sequence number: i.e., no duplicate keys.
// In WriteCommitted it is guarnateed since disable_memtable is used for
// prepare batch which will be written to memtable later during the commit,
// and in WritePrepared it is guaranteed since it will be used only for WAL
// markers which will never be written to memtable. If the commit marker is
// accompanied with CommitTimeWriteBatch that is not written to memtable as
// long as it has no duplicate keys, it does not violate the one-seq-per-batch
// policy.
// batch_cnt is expected to be non-zero in seq_per_batch mode and
// indicates the number of sub-patches. A sub-patch is a subset of the write
// batch that does not have duplicate keys.
Status WriteImpl(const WriteOptions& options, WriteBatch* updates,
WriteCallback* callback = nullptr,
uint64_t* log_used = nullptr, uint64_t log_ref = 0,
bool disable_memtable = false, uint64_t* seq_used = nullptr,
size_t batch_cnt = 0,
PreReleaseCallback* pre_release_callback = nullptr);
Status PipelinedWriteImpl(const WriteOptions& options, WriteBatch* updates,
WriteCallback* callback = nullptr,
uint64_t* log_used = nullptr, uint64_t log_ref = 0,
bool disable_memtable = false,
uint64_t* seq_used = nullptr);
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
// Write only to memtables without joining any write queue
Status UnorderedWriteMemtable(const WriteOptions& write_options,
WriteBatch* my_batch, WriteCallback* callback,
uint64_t log_ref, SequenceNumber seq,
const size_t sub_batch_cnt);
// Whether the batch requires to be assigned with an order
enum AssignOrder : bool { kDontAssignOrder, kDoAssignOrder };
// Whether it requires publishing last sequence or not
enum PublishLastSeq : bool { kDontPublishLastSeq, kDoPublishLastSeq };
// Join the write_thread to write the batch only to the WAL. It is the
// responsibility of the caller to also write the write batch to the memtable
// if it required.
//
// sub_batch_cnt is expected to be non-zero when assign_order = kDoAssignOrder
// indicating the number of sub-batches in my_batch. A sub-patch is a subset
// of the write batch that does not have duplicate keys. When seq_per_batch is
// not set, each key is a separate sub_batch. Otherwise each duplicate key
// marks start of a new sub-batch.
Status WriteImplWALOnly(
WriteThread* write_thread, const WriteOptions& options,
WriteBatch* updates, WriteCallback* callback, uint64_t* log_used,
const uint64_t log_ref, uint64_t* seq_used, const size_t sub_batch_cnt,
PreReleaseCallback* pre_release_callback, const AssignOrder assign_order,
const PublishLastSeq publish_last_seq, const bool disable_memtable);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
// write cached_recoverable_state_ to memtable if it is not empty
// The writer must be the leader in write_thread_ and holding mutex_
Status WriteRecoverableState();
// Actual implementation of Close()
Status CloseImpl();
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
// Recover the descriptor from persistent storage. May do a significant
// amount of work to recover recently logged updates. Any changes to
// be made to the descriptor are added to *edit.
// recovered_seq is set to less than kMaxSequenceNumber if the log's tail is
// skipped.
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
virtual Status Recover(
const std::vector<ColumnFamilyDescriptor>& column_families,
bool read_only = false, bool error_if_wal_file_exists = false,
bool error_if_data_exists_in_wals = false,
uint64_t* recovered_seq = nullptr);
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
virtual bool OwnTablesAndLogs() const { return true; }
// REQUIRES: db mutex held when calling this function, but the db mutex can
// be released and re-acquired. Db mutex will be held when the function
// returns.
// After best-efforts recovery, there may be SST files in db/cf paths that are
// not referenced in the MANIFEST. We delete these SST files. In the
// meantime, we find out the largest file number present in the paths, and
// bump up the version set's next_file_number_ to be 1 + largest_file_number.
Status FinishBestEffortsRecovery();
// SetDbSessionId() should be called in the constuctor DBImpl()
// to ensure that db_session_id_ gets updated every time the DB is opened
void SetDbSessionId();
private:
friend class DB;
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
2018-09-15 22:36:19 +02:00
friend class ErrorHandler;
friend class InternalStats;
friend class PessimisticTransaction;
friend class TransactionBaseImpl;
friend class WriteCommittedTxn;
friend class WritePreparedTxn;
friend class WritePreparedTxnDB;
friend class WriteBatchWithIndex;
friend class WriteUnpreparedTxnDB;
friend class WriteUnpreparedTxn;
#ifndef ROCKSDB_LITE
friend class ForwardIterator;
#endif
friend struct SuperVersion;
friend class CompactedDBImpl;
friend class DBTest_ConcurrentFlushWAL_Test;
friend class DBTest_MixedSlowdownOptionsStop_Test;
friend class DBCompactionTest_CompactBottomLevelFilesWithDeletions_Test;
friend class DBCompactionTest_CompactionDuringShutdown_Test;
friend class StatsHistoryTest_PersistentStatsCreateColumnFamilies_Test;
#ifndef NDEBUG
friend class DBTest2_ReadCallbackTest_Test;
friend class WriteCallbackPTest_WriteWithCallbackTest_Test;
friend class XFTransactionWriteHandler;
friend class DBBlobIndexTest;
friend class WriteUnpreparedTransactionTest_RecoveryTest_Test;
#endif
struct CompactionState;
struct PrepickedCompaction;
struct PurgeFileInfo;
struct WriteContext {
SuperVersionContext superversion_context;
autovector<MemTable*> memtables_to_free_;
explicit WriteContext(bool create_superversion = false)
: superversion_context(create_superversion) {}
~WriteContext() {
superversion_context.Clean();
for (auto& m : memtables_to_free_) {
delete m;
}
}
};
struct LogFileNumberSize {
explicit LogFileNumberSize(uint64_t _number) : number(_number) {}
void AddSize(uint64_t new_size) { size += new_size; }
uint64_t number;
uint64_t size = 0;
bool getting_flushed = false;
};
struct LogWriterNumber {
// pass ownership of _writer
LogWriterNumber(uint64_t _number, log::Writer* _writer)
: number(_number), writer(_writer) {}
log::Writer* ReleaseWriter() {
auto* w = writer;
writer = nullptr;
return w;
}
Status ClearWriter() {
Status s = writer->WriteBuffer();
delete writer;
writer = nullptr;
return s;
}
uint64_t number;
// Visual Studio doesn't support deque's member to be noncopyable because
// of a std::unique_ptr as a member.
log::Writer* writer; // own
// true for some prefix of logs_
bool getting_synced = false;
};
// PurgeFileInfo is a structure to hold information of files to be deleted in
// purge_files_
struct PurgeFileInfo {
std::string fname;
std::string dir_to_sync;
FileType type;
uint64_t number;
int job_id;
PurgeFileInfo(std::string fn, std::string d, FileType t, uint64_t num,
int jid)
: fname(fn), dir_to_sync(d), type(t), number(num), job_id(jid) {}
};
// Argument required by background flush thread.
struct BGFlushArg {
BGFlushArg()
: cfd_(nullptr), max_memtable_id_(0), superversion_context_(nullptr) {}
BGFlushArg(ColumnFamilyData* cfd, uint64_t max_memtable_id,
SuperVersionContext* superversion_context)
: cfd_(cfd),
max_memtable_id_(max_memtable_id),
superversion_context_(superversion_context) {}
// Column family to flush.
ColumnFamilyData* cfd_;
// Maximum ID of memtable to flush. In this column family, memtables with
// IDs smaller than this value must be flushed before this flush completes.
uint64_t max_memtable_id_;
// Pointer to a SuperVersionContext object. After flush completes, RocksDB
// installs a new superversion for the column family. This operation
// requires a SuperVersionContext object (currently embedded in JobContext).
SuperVersionContext* superversion_context_;
};
// Argument passed to flush thread.
struct FlushThreadArg {
DBImpl* db_;
Env::Priority thread_pri_;
};
// Information for a manual compaction
struct ManualCompactionState {
ColumnFamilyData* cfd;
int input_level;
int output_level;
uint32_t output_path_id;
Status status;
bool done;
bool in_progress; // compaction request being processed?
bool incomplete; // only part of requested range compacted
bool exclusive; // current behavior of only one manual
bool disallow_trivial_move; // Force actual compaction to run
const InternalKey* begin; // nullptr means beginning of key range
const InternalKey* end; // nullptr means end of key range
InternalKey* manual_end; // how far we are compacting
InternalKey tmp_storage; // Used to keep track of compaction progress
InternalKey tmp_storage1; // Used to keep track of compaction progress
};
struct PrepickedCompaction {
// background compaction takes ownership of `compaction`.
Compaction* compaction;
// caller retains ownership of `manual_compaction_state` as it is reused
// across background compactions.
ManualCompactionState* manual_compaction_state; // nullptr if non-manual
// task limiter token is requested during compaction picking.
std::unique_ptr<TaskLimiterToken> task_token;
};
struct CompactionArg {
// caller retains ownership of `db`.
DBImpl* db;
// background compaction takes ownership of `prepicked_compaction`.
PrepickedCompaction* prepicked_compaction;
};
// Initialize the built-in column family for persistent stats. Depending on
// whether on-disk persistent stats have been enabled before, it may either
// create a new column family and column family handle or just a column family
// handle.
// Required: DB mutex held
Status InitPersistStatsColumnFamily();
// Persistent Stats column family has two format version key which are used
// for compatibility check. Write format version if it's created for the
// first time, read format version and check compatibility if recovering
// from disk. This function requires DB mutex held at entrance but may
// release and re-acquire DB mutex in the process.
// Required: DB mutex held
Status PersistentStatsProcessFormatVersion();
Status ResumeImpl(DBRecoverContext context);
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
2018-09-15 22:36:19 +02:00
void MaybeIgnoreError(Status* s) const;
const Status CreateArchivalDirectory();
Status CreateColumnFamilyImpl(const ColumnFamilyOptions& cf_options,
const std::string& cf_name,
ColumnFamilyHandle** handle);
Status DropColumnFamilyImpl(ColumnFamilyHandle* column_family);
// Delete any unneeded files and stale in-memory entries.
void DeleteObsoleteFiles();
// Delete obsolete files and log status and information of file deletion
void DeleteObsoleteFileImpl(int job_id, const std::string& fname,
const std::string& path_to_sync, FileType type,
uint64_t number);
// Background process needs to call
// auto x = CaptureCurrentFileNumberInPendingOutputs()
// auto file_num = versions_->NewFileNumber();
// <do something>
// ReleaseFileNumberFromPendingOutputs(x)
// This will protect any file with number `file_num` or greater from being
// deleted while <do something> is running.
// -----------
// This function will capture current file number and append it to
// pending_outputs_. This will prevent any background process to delete any
// file created after this point.
std::list<uint64_t>::iterator CaptureCurrentFileNumberInPendingOutputs();
// This function should be called with the result of
// CaptureCurrentFileNumberInPendingOutputs(). It then marks that any file
// created between the calls CaptureCurrentFileNumberInPendingOutputs() and
// ReleaseFileNumberFromPendingOutputs() can now be deleted (if it's not live
// and blocked by any other pending_outputs_ calls)
void ReleaseFileNumberFromPendingOutputs(
std::unique_ptr<std::list<uint64_t>::iterator>& v);
IOStatus SyncClosedLogs(JobContext* job_context);
// Flush the in-memory write buffer to storage. Switches to a new
// log-file/memtable and writes a new descriptor iff successful. Then
// installs a new super version for the column family.
Status FlushMemTableToOutputFile(
ColumnFamilyData* cfd, const MutableCFOptions& mutable_cf_options,
bool* madeProgress, 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);
// Flush the memtables of (multiple) column families to multiple files on
// persistent storage.
Status FlushMemTablesToOutputFiles(
const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
JobContext* job_context, LogBuffer* log_buffer, Env::Priority thread_pri);
Status AtomicFlushMemTablesToOutputFiles(
const autovector<BGFlushArg>& bg_flush_args, bool* made_progress,
JobContext* job_context, LogBuffer* log_buffer, Env::Priority thread_pri);
// REQUIRES: log_numbers are sorted in ascending order
// corrupted_log_found is set to true if we recover from a corrupted log file.
Status RecoverLogFiles(const std::vector<uint64_t>& log_numbers,
SequenceNumber* next_sequence, bool read_only,
bool* corrupted_log_found);
// The following two methods are used to flush a memtable to
// storage. The first one is used at database RecoveryTime (when the
// database is opened) and is heavyweight because it holds the mutex
// for the entire period. The second method WriteLevel0Table supports
// concurrent flush memtables to storage.
Status WriteLevel0TableForRecovery(int job_id, ColumnFamilyData* cfd,
MemTable* mem, VersionEdit* edit);
// Restore alive_log_files_ and total_log_size_ after recovery.
// It needs to run only when there's no flush during recovery
// (e.g. avoid_flush_during_recovery=true). May also trigger flush
// in case total_log_size > max_total_wal_size.
Status RestoreAliveLogFiles(const std::vector<uint64_t>& log_numbers);
// num_bytes: for slowdown case, delay time is calculated based on
// `num_bytes` going through.
Status DelayWrite(uint64_t num_bytes, const WriteOptions& write_options);
Status ThrottleLowPriWritesIfNeeded(const WriteOptions& write_options,
WriteBatch* my_batch);
// REQUIRES: mutex locked and in write thread.
Status ScheduleFlushes(WriteContext* context);
void MaybeFlushStatsCF(autovector<ColumnFamilyData*>* cfds);
Refactor trimming logic for immutable memtables (#5022) Summary: MyRocks currently sets `max_write_buffer_number_to_maintain` in order to maintain enough history for transaction conflict checking. The effectiveness of this approach depends on the size of memtables. When memtables are small, it may not keep enough history; when memtables are large, this may consume too much memory. We are proposing a new way to configure memtable list history: by limiting the memory usage of immutable memtables. The new option is `max_write_buffer_size_to_maintain` and it will take precedence over the old `max_write_buffer_number_to_maintain` if they are both set to non-zero values. The new option accounts for the total memory usage of flushed immutable memtables and mutable memtable. When the total usage exceeds the limit, RocksDB may start dropping immutable memtables (which is also called trimming history), starting from the oldest one. The semantics of the old option actually works both as an upper bound and lower bound. History trimming will start if number of immutable memtables exceeds the limit, but it will never go below (limit-1) due to history trimming. In order the mimic the behavior with the new option, history trimming will stop if dropping the next immutable memtable causes the total memory usage go below the size limit. For example, assuming the size limit is set to 64MB, and there are 3 immutable memtables with sizes of 20, 30, 30. Although the total memory usage is 80MB > 64MB, dropping the oldest memtable will reduce the memory usage to 60MB < 64MB, so in this case no memtable will be dropped. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5022 Differential Revision: D14394062 Pulled By: miasantreble fbshipit-source-id: 60457a509c6af89d0993f988c9b5c2aa9e45f5c5
2019-08-23 22:54:09 +02:00
Status TrimMemtableHistory(WriteContext* context);
Status SwitchMemtable(ColumnFamilyData* cfd, WriteContext* context);
void SelectColumnFamiliesForAtomicFlush(autovector<ColumnFamilyData*>* cfds);
// Force current memtable contents to be flushed.
Status FlushMemTable(ColumnFamilyData* cfd, const FlushOptions& options,
FlushReason flush_reason, bool writes_stopped = false);
Status AtomicFlushMemTables(
const autovector<ColumnFamilyData*>& column_family_datas,
const FlushOptions& options, FlushReason flush_reason,
bool writes_stopped = false);
// Wait until flushing this column family won't stall writes
Status WaitUntilFlushWouldNotStallWrites(ColumnFamilyData* cfd,
bool* flush_needed);
// Wait for memtable flushed.
// If flush_memtable_id is non-null, wait until the memtable with the ID
// gets flush. Otherwise, wait until the column family don't have any
// memtable pending flush.
// resuming_from_bg_err indicates whether the caller is attempting to resume
// from background error.
Status WaitForFlushMemTable(ColumnFamilyData* cfd,
const uint64_t* flush_memtable_id = nullptr,
bool resuming_from_bg_err = false) {
return WaitForFlushMemTables({cfd}, {flush_memtable_id},
resuming_from_bg_err);
}
// Wait for memtables to be flushed for multiple column families.
Status WaitForFlushMemTables(
const autovector<ColumnFamilyData*>& cfds,
const autovector<const uint64_t*>& flush_memtable_ids,
bool resuming_from_bg_err);
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
inline void WaitForPendingWrites() {
mutex_.AssertHeld();
TEST_SYNC_POINT("DBImpl::WaitForPendingWrites:BeforeBlock");
// In case of pipelined write is enabled, wait for all pending memtable
// writers.
if (immutable_db_options_.enable_pipelined_write) {
// Memtable writers may call DB::Get in case max_successive_merges > 0,
// which may lock mutex. Unlocking mutex here to avoid deadlock.
mutex_.Unlock();
write_thread_.WaitForMemTableWriters();
mutex_.Lock();
}
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
if (!immutable_db_options_.unordered_write) {
// Then the writes are finished before the next write group starts
return;
}
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
// Wait for the ones who already wrote to the WAL to finish their
// memtable write.
if (pending_memtable_writes_.load() != 0) {
std::unique_lock<std::mutex> guard(switch_mutex_);
switch_cv_.wait(guard,
[&] { return pending_memtable_writes_.load() == 0; });
}
}
// REQUIRES: mutex locked and in write thread.
void AssignAtomicFlushSeq(const autovector<ColumnFamilyData*>& cfds);
// REQUIRES: mutex locked and in write thread.
Status SwitchWAL(WriteContext* write_context);
// REQUIRES: mutex locked and in write thread.
Status HandleWriteBufferFull(WriteContext* write_context);
// REQUIRES: mutex locked
Status PreprocessWrite(const WriteOptions& write_options, bool* need_log_sync,
WriteContext* write_context);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
WriteBatch* MergeBatch(const WriteThread::WriteGroup& write_group,
WriteBatch* tmp_batch, size_t* write_with_wal,
WriteBatch** to_be_cached_state);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
IOStatus WriteToWAL(const WriteBatch& merged_batch, log::Writer* log_writer,
uint64_t* log_used, uint64_t* log_size);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
IOStatus WriteToWAL(const WriteThread::WriteGroup& write_group,
log::Writer* log_writer, uint64_t* log_used,
bool need_log_sync, bool need_log_dir_sync,
SequenceNumber sequence);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
IOStatus ConcurrentWriteToWAL(const WriteThread::WriteGroup& write_group,
uint64_t* log_used,
SequenceNumber* last_sequence, size_t seq_inc);
// Used by WriteImpl to update bg_error_ if paranoid check is enabled.
// Caller must hold mutex_.
void WriteStatusCheckOnLocked(const Status& status);
// Used by WriteImpl to update bg_error_ if paranoid check is enabled.
void WriteStatusCheck(const Status& status);
// Used by WriteImpl to update bg_error_ when IO error happens, e.g., write
// WAL, sync WAL fails, if paranoid check is enabled.
void IOStatusCheck(const IOStatus& status);
// Used by WriteImpl to update bg_error_ in case of memtable insert error.
void MemTableInsertStatusCheck(const Status& memtable_insert_status);
#ifndef ROCKSDB_LITE
Status 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);
// Wait for current IngestExternalFile() calls to finish.
// REQUIRES: mutex_ held
void WaitForIngestFile();
#else
// IngestExternalFile is not supported in ROCKSDB_LITE so this function
// will be no-op
void WaitForIngestFile() {}
#endif // ROCKSDB_LITE
ColumnFamilyData* GetColumnFamilyDataByName(const std::string& cf_name);
void MaybeScheduleFlushOrCompaction();
// A flush request specifies the column families to flush as well as the
// largest memtable id to persist for each column family. Once all the
// memtables whose IDs are smaller than or equal to this per-column-family
// specified value, this flush request is considered to have completed its
// work of flushing this column family. After completing the work for all
// column families in this request, this flush is considered complete.
typedef std::vector<std::pair<ColumnFamilyData*, uint64_t>> FlushRequest;
void GenerateFlushRequest(const autovector<ColumnFamilyData*>& cfds,
FlushRequest* req);
void SchedulePendingFlush(const FlushRequest& req, FlushReason flush_reason);
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
void SchedulePendingCompaction(ColumnFamilyData* cfd);
void SchedulePendingPurge(std::string fname, std::string dir_to_sync,
FileType type, uint64_t number, int job_id);
static void BGWorkCompaction(void* 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
2017-08-04 00:36:28 +02:00
// Runs a pre-chosen universal compaction involving bottom level in a
// separate, bottom-pri thread pool.
static void BGWorkBottomCompaction(void* arg);
static void BGWorkFlush(void* arg);
static void BGWorkPurge(void* arg);
static void UnscheduleCompactionCallback(void* arg);
static void UnscheduleFlushCallback(void* 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
2017-08-04 00:36:28 +02:00
void BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
Env::Priority thread_pri);
void BackgroundCallFlush(Env::Priority thread_pri);
void BackgroundCallPurge();
Status BackgroundCompaction(bool* madeProgress, 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
2017-08-04 00:36:28 +02:00
LogBuffer* log_buffer,
PrepickedCompaction* prepicked_compaction,
Env::Priority thread_pri);
Status BackgroundFlush(bool* madeProgress, JobContext* job_context,
LogBuffer* log_buffer, FlushReason* reason,
Env::Priority thread_pri);
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
2018-09-15 22:36:19 +02:00
bool EnoughRoomForCompaction(ColumnFamilyData* cfd,
const std::vector<CompactionInputFiles>& inputs,
bool* sfm_bookkeeping, LogBuffer* log_buffer);
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
2018-12-13 22:16:04 +01:00
// Request compaction tasks token from compaction thread limiter.
// It always succeeds if force = true or limiter is disable.
bool RequestCompactionToken(ColumnFamilyData* cfd, bool force,
std::unique_ptr<TaskLimiterToken>* token,
LogBuffer* log_buffer);
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
2018-10-09 07:52:58 +02:00
// Schedule background tasks
void StartPeriodicWorkScheduler();
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
2018-10-09 07:52:58 +02:00
void PrintStatistics();
size_t EstimateInMemoryStatsHistorySize() const;
// Return the minimum empty level that could hold the total data in the
// input level. Return the input level, if such level could not be found.
int FindMinimumEmptyLevelFitting(ColumnFamilyData* cfd,
const MutableCFOptions& mutable_cf_options,
int level);
// Move the files in the input level to the target level.
// If target_level < 0, automatically calculate the minimum level that could
// hold the data set.
Status ReFitLevel(ColumnFamilyData* cfd, int level, int target_level = -1);
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
// helper functions for adding and removing from flush & compaction queues
void AddToCompactionQueue(ColumnFamilyData* cfd);
ColumnFamilyData* PopFirstFromCompactionQueue();
FlushRequest PopFirstFromFlushQueue();
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
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
2018-12-13 22:16:04 +01:00
// Pick the first unthrottled compaction with task token from queue.
ColumnFamilyData* PickCompactionFromQueue(
std::unique_ptr<TaskLimiterToken>* token, LogBuffer* log_buffer);
// helper function to call after some of the logs_ were synced
void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);
SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary,
bool lock = true);
uint64_t GetMaxTotalWalSize() const;
FSDirectory* GetDataDir(ColumnFamilyData* cfd, size_t path_id) const;
Status CloseHelper();
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
2018-09-15 22:36:19 +02:00
void WaitForBackgroundWork();
// Background threads call this function, which is just a wrapper around
// the InstallSuperVersion() function. Background threads carry
// sv_context which can have new_superversion already
// allocated.
// All ColumnFamily state changes go through this function. Here we analyze
// the new state and we schedule background work if we detect that the new
// state needs flush or compaction.
void InstallSuperVersionAndScheduleWork(
ColumnFamilyData* cfd, SuperVersionContext* sv_context,
const MutableCFOptions& mutable_cf_options);
bool GetIntPropertyInternal(ColumnFamilyData* cfd,
const DBPropertyInfo& property_info,
bool is_locked, uint64_t* value);
bool GetPropertyHandleOptionsStatistics(std::string* value);
bool HasPendingManualCompaction();
bool HasExclusiveManualCompaction();
void AddManualCompaction(ManualCompactionState* m);
void RemoveManualCompaction(ManualCompactionState* m);
bool ShouldntRunManualCompaction(ManualCompactionState* m);
bool HaveManualCompaction(ColumnFamilyData* cfd);
bool MCOverlap(ManualCompactionState* m, ManualCompactionState* m1);
#ifndef ROCKSDB_LITE
void 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;
// Reserve the next 'num' file numbers for to-be-ingested external SST files,
// and return the current file_number in 'next_file_number'.
// Write a version edit to the MANIFEST.
Status ReserveFileNumbersBeforeIngestion(
ColumnFamilyData* cfd, uint64_t num,
std::unique_ptr<std::list<uint64_t>::iterator>& pending_output_elem,
uint64_t* next_file_number);
#endif //! ROCKSDB_LITE
bool ShouldPurge(uint64_t file_number) const;
void MarkAsGrabbedForPurge(uint64_t file_number);
size_t GetWalPreallocateBlockSize(uint64_t write_buffer_size) const;
Env::WriteLifeTimeHint CalculateWALWriteHint() { return Env::WLTH_SHORT; }
IOStatus CreateWAL(uint64_t log_file_num, uint64_t recycle_log_number,
size_t preallocate_block_size, log::Writer** new_log);
// Validate self-consistency of DB options
static Status ValidateOptions(const DBOptions& db_options);
// Validate self-consistency of DB options and its consistency with cf options
static Status ValidateOptions(
const DBOptions& db_options,
const std::vector<ColumnFamilyDescriptor>& column_families);
// Utility function to do some debug validation and sort the given vector
// of MultiGet keys
void PrepareMultiGetKeys(
const size_t num_keys, bool sorted,
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* key_ptrs);
// A structure to hold the information required to process MultiGet of keys
// belonging to one column family. For a multi column family MultiGet, there
// will be a container of these objects.
struct MultiGetColumnFamilyData {
ColumnFamilyHandle* cf;
ColumnFamilyData* cfd;
// For the batched MultiGet which relies on sorted keys, start specifies
// the index of first key belonging to this column family in the sorted
// list.
size_t start;
// For the batched MultiGet case, num_keys specifies the number of keys
// belonging to this column family in the sorted list
size_t num_keys;
// SuperVersion for the column family obtained in a manner that ensures a
// consistent view across all column families in the DB
SuperVersion* super_version;
MultiGetColumnFamilyData(ColumnFamilyHandle* column_family,
SuperVersion* sv)
: cf(column_family),
cfd(static_cast<ColumnFamilyHandleImpl*>(cf)->cfd()),
start(0),
num_keys(0),
super_version(sv) {}
MultiGetColumnFamilyData(ColumnFamilyHandle* column_family, size_t first,
size_t count, SuperVersion* sv)
: cf(column_family),
cfd(static_cast<ColumnFamilyHandleImpl*>(cf)->cfd()),
start(first),
num_keys(count),
super_version(sv) {}
MultiGetColumnFamilyData() = default;
};
// A common function to obtain a consistent snapshot, which can be implicit
// if the user doesn't specify a snapshot in read_options, across
// multiple column families for MultiGet. It will attempt to get an implicit
// snapshot without acquiring the db_mutes, but will give up after a few
// tries and acquire the mutex if a memtable flush happens. The template
// allows both the batched and non-batched MultiGet to call this with
// either an std::unordered_map or autovector of column families.
//
// If callback is non-null, the callback is refreshed with the snapshot
// sequence number
//
// A return value of true indicates that the SuperVersions were obtained
// from the ColumnFamilyData, whereas false indicates they are thread
// local
template <class T>
bool MultiCFSnapshot(
const ReadOptions& read_options, ReadCallback* callback,
std::function<MultiGetColumnFamilyData*(typename T::iterator&)>&
iter_deref_func,
T* cf_list, SequenceNumber* snapshot);
// The actual implementation of the batching MultiGet. The caller is expected
// to have acquired the SuperVersion and pass in a snapshot sequence number
// in order to construct the LookupKeys. The start_key and num_keys specify
// the range of keys in the sorted_keys vector for a single column family.
Status MultiGetImpl(
const ReadOptions& read_options, size_t start_key, size_t num_keys,
autovector<KeyContext*, MultiGetContext::MAX_BATCH_SIZE>* sorted_keys,
SuperVersion* sv, SequenceNumber snap_seqnum, ReadCallback* callback,
bool* is_blob_index);
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
2020-06-25 04:05:47 +02:00
Status DisableFileDeletionsWithLock();
// table_cache_ provides its own synchronization
std::shared_ptr<Cache> table_cache_;
// Lock over the persistent DB state. Non-nullptr iff successfully acquired.
FileLock* db_lock_;
// In addition to mutex_, log_write_mutex_ protected writes to stats_history_
InstrumentedMutex stats_history_mutex_;
// In addition to mutex_, log_write_mutex_ protected writes to logs_ and
// logfile_number_. With two_write_queues it also protects alive_log_files_,
// and log_empty_. Refer to the definition of each variable below for more
// details.
// Note: to avoid dealock, if needed to acquire both log_write_mutex_ and
// mutex_, the order should be first mutex_ and then log_write_mutex_.
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
InstrumentedMutex log_write_mutex_;
std::atomic<bool> shutting_down_;
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
2020-08-14 20:28:12 +02:00
// If zero, manual compactions are allowed to proceed. If non-zero, manual
// compactions may still be running, but will quickly fail with
// `Status::Incomplete`. The value indicates how many threads have paused
// manual compactions. It is accessed in read mode outside the DB mutex in
// compaction code paths.
std::atomic<int> manual_compaction_paused_;
// This condition variable is signaled on these conditions:
// * whenever bg_compaction_scheduled_ goes down to 0
// * if AnyManualCompaction, whenever a compaction finishes, even if it hasn't
// made any progress
// * whenever a compaction made any progress
// * whenever bg_flush_scheduled_ or bg_purge_scheduled_ value decreases
// (i.e. whenever a flush is done, even if it didn't make any progress)
// * whenever there is an error in background purge, flush or compaction
// * whenever num_running_ingest_file_ goes to 0.
// * whenever pending_purge_obsolete_files_ goes to 0.
// * whenever disable_delete_obsolete_files_ goes to 0.
// * whenever SetOptions successfully updates options.
// * whenever a column family is dropped.
InstrumentedCondVar bg_cv_;
// Writes are protected by locking both mutex_ and log_write_mutex_, and reads
// must be under either mutex_ or log_write_mutex_. Since after ::Open,
// logfile_number_ is currently updated only in write_thread_, it can be read
// from the same write_thread_ without any locks.
uint64_t logfile_number_;
std::deque<uint64_t>
log_recycle_files_; // a list of log files that we can recycle
bool log_dir_synced_;
// Without two_write_queues, read and writes to log_empty_ are protected by
// mutex_. Since it is currently updated/read only in write_thread_, it can be
// accessed from the same write_thread_ without any locks. With
// two_write_queues writes, where it can be updated in different threads,
// read and writes are protected by log_write_mutex_ instead. This is to avoid
// expesnive mutex_ lock during WAL write, which update log_empty_.
bool log_empty_;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
ColumnFamilyHandleImpl* persist_stats_cf_handle_;
bool persistent_stats_cfd_exists_ = true;
// Without two_write_queues, read and writes to alive_log_files_ are
// protected by mutex_. However since back() is never popped, and push_back()
// is done only from write_thread_, the same thread can access the item
// reffered by back() without mutex_. With two_write_queues_, writes
// are protected by locking both mutex_ and log_write_mutex_, and reads must
// be under either mutex_ or log_write_mutex_.
std::deque<LogFileNumberSize> alive_log_files_;
// Log files that aren't fully synced, and the current log file.
// Synchronization:
// - push_back() is done from write_thread_ with locked mutex_ and
// log_write_mutex_
// - pop_front() is done from any thread with locked mutex_ and
// log_write_mutex_
// - reads are done with either locked mutex_ or log_write_mutex_
// - back() and items with getting_synced=true are not popped,
// - The same thread that sets getting_synced=true will reset it.
// - it follows that the object referred by back() can be safely read from
// the write_thread_ without using mutex
// - it follows that the items with getting_synced=true can be safely read
// from the same thread that has set getting_synced=true
std::deque<LogWriterNumber> logs_;
// Signaled when getting_synced becomes false for some of the logs_.
InstrumentedCondVar log_sync_cv_;
// This is the app-level state that is written to the WAL but will be used
// only during recovery. Using this feature enables not writing the state to
// memtable on normal writes and hence improving the throughput. Each new
// write of the state will replace the previous state entirely even if the
// keys in the two consecuitive states do not overlap.
// It is protected by log_write_mutex_ when two_write_queues_ is enabled.
// Otherwise only the heaad of write_thread_ can access it.
WriteBatch cached_recoverable_state_;
std::atomic<bool> cached_recoverable_state_empty_ = {true};
std::atomic<uint64_t> total_log_size_;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
// If this is non-empty, we need to delete these log files in background
// threads. Protected by db mutex.
autovector<log::Writer*> logs_to_free_;
bool is_snapshot_supported_;
std::map<uint64_t, std::map<std::string, uint64_t>> stats_history_;
std::map<std::string, uint64_t> stats_slice_;
bool stats_slice_initialized_ = false;
Directories directories_;
WriteBufferManager* write_buffer_manager_;
WriteThread write_thread_;
WriteBatch tmp_batch_;
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
// The write thread when the writers have no memtable write. This will be used
// in 2PC to batch the prepares separately from the serial commit.
WriteThread nonmem_write_thread_;
WriteController write_controller_;
// Size of the last batch group. In slowdown mode, next write needs to
// sleep if it uses up the quota.
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
// Note: This is to protect memtable and compaction. If the batch only writes
// to the WAL its size need not to be included in this.
uint64_t last_batch_group_size_;
FlushScheduler flush_scheduler_;
Refactor trimming logic for immutable memtables (#5022) Summary: MyRocks currently sets `max_write_buffer_number_to_maintain` in order to maintain enough history for transaction conflict checking. The effectiveness of this approach depends on the size of memtables. When memtables are small, it may not keep enough history; when memtables are large, this may consume too much memory. We are proposing a new way to configure memtable list history: by limiting the memory usage of immutable memtables. The new option is `max_write_buffer_size_to_maintain` and it will take precedence over the old `max_write_buffer_number_to_maintain` if they are both set to non-zero values. The new option accounts for the total memory usage of flushed immutable memtables and mutable memtable. When the total usage exceeds the limit, RocksDB may start dropping immutable memtables (which is also called trimming history), starting from the oldest one. The semantics of the old option actually works both as an upper bound and lower bound. History trimming will start if number of immutable memtables exceeds the limit, but it will never go below (limit-1) due to history trimming. In order the mimic the behavior with the new option, history trimming will stop if dropping the next immutable memtable causes the total memory usage go below the size limit. For example, assuming the size limit is set to 64MB, and there are 3 immutable memtables with sizes of 20, 30, 30. Although the total memory usage is 80MB > 64MB, dropping the oldest memtable will reduce the memory usage to 60MB < 64MB, so in this case no memtable will be dropped. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5022 Differential Revision: D14394062 Pulled By: miasantreble fbshipit-source-id: 60457a509c6af89d0993f988c9b5c2aa9e45f5c5
2019-08-23 22:54:09 +02:00
TrimHistoryScheduler trim_history_scheduler_;
SnapshotList snapshots_;
// For each background job, pending_outputs_ keeps the current file number at
// the time that background job started.
// FindObsoleteFiles()/PurgeObsoleteFiles() never deletes any file that has
// number bigger than any of the file number in pending_outputs_. Since file
// numbers grow monotonically, this also means that pending_outputs_ is always
// sorted. After a background job is done executing, its file number is
// deleted from pending_outputs_, which allows PurgeObsoleteFiles() to clean
// it up.
// State is protected with db mutex.
std::list<uint64_t> pending_outputs_;
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
// flush_queue_ and compaction_queue_ hold column families that we need to
// flush and compact, respectively.
// A column family is inserted into flush_queue_ when it satisfies condition
// cfd->imm()->IsFlushPending()
// A column family is inserted into compaction_queue_ when it satisfied
// condition cfd->NeedsCompaction()
// Column families in this list are all Ref()-erenced
// TODO(icanadi) Provide some kind of ReferencedColumnFamily class that will
// do RAII on ColumnFamilyData
// Column families are in this queue when they need to be flushed or
// compacted. Consumers of these queues are flush and compaction threads. When
// column family is put on this queue, we increase unscheduled_flushes_ and
// unscheduled_compactions_. When these variables are bigger than zero, that
// means we need to schedule background threads for flush and compaction.
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
// Once the background threads are scheduled, we decrease unscheduled_flushes_
// and unscheduled_compactions_. That way we keep track of number of
// compaction and flush threads we need to schedule. This scheduling is done
// in MaybeScheduleFlushOrCompaction()
// invariant(column family present in flush_queue_ <==>
// ColumnFamilyData::pending_flush_ == true)
std::deque<FlushRequest> flush_queue_;
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
// invariant(column family present in compaction_queue_ <==>
// ColumnFamilyData::pending_compaction_ == true)
std::deque<ColumnFamilyData*> compaction_queue_;
// A map to store file numbers and filenames of the files to be purged
std::unordered_map<uint64_t, PurgeFileInfo> purge_files_;
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
2018-03-28 19:23:31 +02:00
// A vector to store the file numbers that have been assigned to certain
// JobContext. Current implementation tracks table and blob files only.
std::unordered_set<uint64_t> files_grabbed_for_purge_;
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
2018-03-28 19:23:31 +02:00
// A queue to store log writers to close
std::deque<log::Writer*> logs_to_free_queue_;
std::deque<SuperVersion*> superversions_to_free_queue_;
Rewritten system for scheduling background work Summary: When scaling to higher number of column families, the worst bottleneck was MaybeScheduleFlushOrCompaction(), which did a for loop over all column families while holding a mutex. This patch addresses the issue. The approach is similar to our earlier efforts: instead of a pull-model, where we do something for every column family, we can do a push-based model -- when we detect that column family is ready to be flushed/compacted, we add it to the flush_queue_/compaction_queue_. That way we don't need to loop over every column family in MaybeScheduleFlushOrCompaction. Here are the performance results: Command: ./db_bench --write_buffer_size=268435456 --db_write_buffer_size=268435456 --db=/fast-rocksdb-tmp/rocks_lots_of_cf --use_existing_db=0 --open_files=55000 --statistics=1 --histogram=1 --disable_data_sync=1 --max_write_buffer_number=2 --sync=0 --benchmarks=fillrandom --threads=16 --num_column_families=5000 --disable_wal=1 --max_background_flushes=16 --max_background_compactions=16 --level0_file_num_compaction_trigger=2 --level0_slowdown_writes_trigger=2 --level0_stop_writes_trigger=3 --hard_rate_limit=1 --num=33333333 --writes=33333333 Before the patch: fillrandom : 26.950 micros/op 37105 ops/sec; 4.1 MB/s After the patch: fillrandom : 17.404 micros/op 57456 ops/sec; 6.4 MB/s Next bottleneck is VersionSet::AddLiveFiles, which is painfully slow when we have a lot of files. This is coming in the next patch, but when I removed that code, here's what I got: fillrandom : 7.590 micros/op 131758 ops/sec; 14.6 MB/s Test Plan: make check two stress tests: Big number of compactions and flushes: ./db_stress --threads=30 --ops_per_thread=20000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=15 --max_background_compactions=10 --max_background_flushes=10 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 max_background_flushes=0, to verify that this case also works correctly ./db_stress --threads=30 --ops_per_thread=2000000 --max_key=10000 --column_families=20 --clear_column_family_one_in=10000000 --verify_before_write=0 --reopen=3 --max_background_compactions=3 --max_background_flushes=0 --db=/fast-rocksdb-tmp/db_stress --prefixpercent=0 --iterpercent=0 --writepercent=75 --db_write_buffer_size=2000000 Reviewers: ljin, rven, yhchiang, sdong Reviewed By: sdong Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D30123
2014-12-19 20:38:12 +01:00
int unscheduled_flushes_;
int unscheduled_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
2017-08-04 00:36:28 +02:00
// count how many background compactions are running or have been scheduled in
// the BOTTOM pool
int bg_bottom_compaction_scheduled_;
Fix a deadlock in CompactRange() Summary: The way DBImpl::TEST_CompactRange() throttles down the number of bg compactions can cause it to deadlock when CompactRange() is called concurrently from multiple threads. Imagine a following scenario with only two threads (max_background_compactions is 10 and bg_compaction_scheduled_ is initially 0): 1. Thread #1 increments bg_compaction_scheduled_ (to LargeNumber), sets bg_compaction_scheduled_ to 9 (newvalue), schedules the compaction (bg_compaction_scheduled_ is now 10) and waits for it to complete. 2. Thread #2 calls TEST_CompactRange(), increments bg_compaction_scheduled_ (now LargeNumber + 10) and waits on a cv for bg_compaction_scheduled_ to drop to LargeNumber. 3. BG thread completes the first manual compaction, decrements bg_compaction_scheduled_ and wakes up all threads waiting on bg_cv_. Thread #1 runs, increments bg_compaction_scheduled_ by LargeNumber again (now 2*LargeNumber + 9). Since that's more than LargeNumber + newvalue, thread #2 also goes to sleep (waiting on bg_cv_), without resetting bg_compaction_scheduled_. This diff attempts to address the problem by introducing a new counter bg_manual_only_ (when positive, MaybeScheduleFlushOrCompaction() will only schedule manual compactions). Test Plan: I could pretty much consistently reproduce the deadlock with a program that calls CompactRange(nullptr, nullptr) immediately after Write() from multiple threads. This no longer happens with this patch. Tests (make check) pass. Reviewers: dhruba, igor, sdong, haobo Reviewed By: igor CC: leveldb Differential Revision: https://reviews.facebook.net/D14799
2013-12-22 00:10:39 +01:00
// count how many background compactions are running or have been scheduled
int bg_compaction_scheduled_;
// stores the number of compactions are currently running
int num_running_compactions_;
// number of background memtable flush jobs, submitted to the HIGH pool
int bg_flush_scheduled_;
// stores the number of flushes are currently running
int num_running_flushes_;
// number of background obsolete file purge jobs, submitted to the HIGH pool
int bg_purge_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
2017-08-04 00:36:28 +02:00
std::deque<ManualCompactionState*> manual_compaction_dequeue_;
// shall we disable deletion of obsolete files
// if 0 the deletion is enabled.
// if non-zero, files will not be getting deleted
// This enables two different threads to call
// EnableFileDeletions() and DisableFileDeletions()
// without any synchronization
int disable_delete_obsolete_files_;
// Number of times FindObsoleteFiles has found deletable files and the
// corresponding call to PurgeObsoleteFiles has not yet finished.
int pending_purge_obsolete_files_;
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
// last time when DeleteObsoleteFiles with full scan was executed. Originally
// initialized with startup time.
uint64_t delete_obsolete_files_last_run_;
// last time stats were dumped to LOG
std::atomic<uint64_t> last_stats_dump_time_microsec_;
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
// The thread that wants to switch memtable, can wait on this cv until the
// pending writes to memtable finishes.
std::condition_variable switch_cv_;
// The mutex used by switch_cv_. mutex_ should be acquired beforehand.
std::mutex switch_mutex_;
// Number of threads intending to write to memtable
std::atomic<size_t> pending_memtable_writes_ = {};
// Each flush or compaction gets its own job id. this counter makes sure
// they're unique
std::atomic<int> next_job_id_;
// A flag indicating whether the current rocksdb database has any
// data that is not yet persisted into either WAL or SST file.
// Used when disableWAL is true.
std::atomic<bool> has_unpersisted_data_;
// if an attempt was made to flush all column families that
// the oldest log depends on but uncommitted data in the oldest
// log prevents the log from being released.
// We must attempt to free the dependent memtables again
// at a later time after the transaction in the oldest
// log is fully commited.
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
bool unable_to_release_oldest_log_;
static const int KEEP_LOG_FILE_NUM = 1000;
// MSVC version 1800 still does not have constexpr for ::max()
static const uint64_t kNoTimeOut = port::kMaxUint64;
std::string db_absolute_path_;
Export Import sst files (#5495) Summary: Refresh of the earlier change here - https://github.com/facebook/rocksdb/issues/5135 This is a review request for code change needed for - https://github.com/facebook/rocksdb/issues/3469 "Add support for taking snapshot of a column family and creating column family from a given CF snapshot" We have an implementation for this that we have been testing internally. We have two new APIs that together provide this functionality. (1) ExportColumnFamily() - This API is modelled after CreateCheckpoint() as below. // Exports all live SST files of a specified Column Family onto export_dir, // returning SST files information in metadata. // - SST files will be created as hard links when the directory specified // is in the same partition as the db directory, copied otherwise. // - export_dir should not already exist and will be created by this API. // - Always triggers a flush. virtual Status ExportColumnFamily(ColumnFamilyHandle* handle, const std::string& export_dir, ExportImportFilesMetaData** metadata); Internally, the API will DisableFileDeletions(), GetColumnFamilyMetaData(), Parse through metadata, creating links/copies of all the sst files, EnableFileDeletions() and complete the call by returning the list of file metadata. (2) CreateColumnFamilyWithImport() - This API is modeled after IngestExternalFile(), but invoked only during a CF creation as below. // CreateColumnFamilyWithImport() will create a new column family with // column_family_name and import external SST files specified in metadata into // this column family. // (1) External SST files can be created using SstFileWriter. // (2) External SST files can be exported from a particular column family in // an existing DB. // Option in import_options specifies whether the external files are copied or // moved (default is copy). When option specifies copy, managing files at // external_file_path is caller's responsibility. When option specifies a // move, the call ensures that the specified files at external_file_path are // deleted on successful return and files are not modified on any error // return. // On error return, column family handle returned will be nullptr. // ColumnFamily will be present on successful return and will not be present // on error return. ColumnFamily may be present on any crash during this call. virtual Status CreateColumnFamilyWithImport( const ColumnFamilyOptions& options, const std::string& column_family_name, const ImportColumnFamilyOptions& import_options, const ExportImportFilesMetaData& metadata, ColumnFamilyHandle** handle); Internally, this API creates a new CF, parses all the sst files and adds it to the specified column family, at the same level and with same sequence number as in the metadata. Also performs safety checks with respect to overlaps between the sst files being imported. If incoming sequence number is higher than current local sequence number, local sequence number is updated to reflect this. Note, as the sst files is are being moved across Column Families, Column Family name in sst file will no longer match the actual column family on destination DB. The API does not modify Column Family name or id in the sst files being imported. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5495 Differential Revision: D16018881 fbshipit-source-id: 9ae2251025d5916d35a9fc4ea4d6707f6be16ff9
2019-07-17 21:22:21 +02:00
// Number of running IngestExternalFile() or CreateColumnFamilyWithImport()
// calls.
// REQUIRES: mutex held
int num_running_ingest_file_;
#ifndef ROCKSDB_LITE
WalManager wal_manager_;
#endif // ROCKSDB_LITE
// Unified interface for logging events
EventLogger event_logger_;
// A value of > 0 temporarily disables scheduling of background work
int bg_work_paused_;
// A value of > 0 temporarily disables scheduling of background compaction
int bg_compaction_paused_;
// Guard against multiple concurrent refitting
bool refitting_level_;
// Indicate DB was opened successfully
bool opened_successfully_;
// The min threshold to triggere bottommost compaction for removing
// garbages, among all column families.
SequenceNumber bottommost_files_mark_threshold_ = kMaxSequenceNumber;
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
LogsWithPrepTracker logs_with_prep_tracker_;
// Callback for compaction to check if a key is visible to a snapshot.
// REQUIRES: mutex held
std::unique_ptr<SnapshotChecker> snapshot_checker_;
// Callback for when the cached_recoverable_state_ is written to memtable
// Only to be set during initialization
std::unique_ptr<PreReleaseCallback> recoverable_state_pre_release_callback_;
#ifndef ROCKSDB_LITE
// Scheduler to run DumpStats(), PersistStats(), and FlushInfoLog().
// Currently, it always use a global instance from
// PeriodicWorkScheduler::Default(). Only in unittest, it can be overrided by
// PeriodicWorkTestScheduler.
PeriodicWorkScheduler* periodic_work_scheduler_;
#endif
// When set, we use a separate queue for writes that don't write to memtable.
// In 2PC these are the writes at Prepare phase.
const bool two_write_queues_;
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 23:06:43 +02:00
const bool manual_wal_flush_;
// LastSequence also indicates last published sequence visibile to the
// readers. Otherwise LastPublishedSequence should be used.
const bool last_seq_same_as_publish_seq_;
// It indicates that a customized gc algorithm must be used for
// flush/compaction and if it is not provided vis SnapshotChecker, we should
// disable gc to be safe.
const bool use_custom_gc_;
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
2018-09-15 22:36:19 +02:00
// Flag to indicate that the DB instance shutdown has been initiated. This
// different from shutting_down_ atomic in that it is set at the beginning
// of shutdown sequence, specifically in order to prevent any background
// error recovery from going on in parallel. The latter, shutting_down_,
// is set a little later during the shutdown after scheduling memtable
// flushes
move dump stats to a separate thread (#4382) Summary: Currently statistics are supposed to be dumped to info log at intervals of `options.stats_dump_period_sec`. However the implementation choice was to bind it with compaction thread, meaning if the database has been serving very light traffic, the stats may not get dumped at all. We decided to separate stats dumping into a new timed thread using `TimerQueue`, which is already used in blob_db. This will allow us schedule new timed tasks with more deterministic behavior. Tested with db_bench using `--stats_dump_period_sec=20` in command line: > LOG:2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:05.643286 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:25.691325 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG:2018/09/17-14:08:45.740989 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- LOG content: > 2018/09/17-14:07:45.575025 7fe99fbfe700 [WARN] [db/db_impl.cc:605] ------- DUMPING STATS ------- 2018/09/17-14:07:45.575080 7fe99fbfe700 [WARN] [db/db_impl.cc:606] ** DB Stats ** Uptime(secs): 20.0 total, 20.0 interval Cumulative writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5.57 GB, 285.01 MB/s Cumulative WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 GB, 285.01 MB/s Cumulative stall: 00:00:0.012 H:M:S, 0.1 percent Interval writes: 4447K writes, 4447K keys, 4447K commit groups, 1.0 writes per commit group, ingest: 5700.71 MB, 285.01 MB/s Interval WAL: 4447K writes, 0 syncs, 4447638.00 writes per sync, written: 5.57 MB, 285.01 MB/s Interval stall: 00:00:0.012 H:M:S, 0.1 percent ** Compaction Stats [default] ** Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Pull Request resolved: https://github.com/facebook/rocksdb/pull/4382 Differential Revision: D9933051 Pulled By: miasantreble fbshipit-source-id: 6d12bb1e4977674eea4bf2d2ac6d486b814bb2fa
2018-10-09 07:52:58 +02:00
std::atomic<bool> shutdown_initiated_;
// Flag to indicate whether sst_file_manager object was allocated in
// DB::Open() or passed to us
bool own_sfm_;
Added support for differential snapshots Summary: The motivation for this PR is to add to RocksDB support for differential (incremental) snapshots, as snapshot of the DB changes between two points in time (one can think of it as diff between to sequence numbers, or the diff D which can be thought of as an SST file or just set of KVs that can be applied to sequence number S1 to get the database to the state at sequence number S2). This feature would be useful for various distributed storages layers built on top of RocksDB, as it should help reduce resources (time and network bandwidth) needed to recover and rebuilt DB instances as replicas in the context of distributed storages. From the API standpoint that would like client app requesting iterator between (start seqnum) and current DB state, and reading the "diff". This is a very draft PR for initial review in the discussion on the approach, i'm going to rework some parts and keep updating the PR. For now, what's done here according to initial discussions: Preserving deletes: - We want to be able to optionally preserve recent deletes for some defined period of time, so that if a delete came in recently and might need to be included in the next incremental snapshot it would't get dropped by a compaction. This is done by adding new param to Options (preserve deletes flag) and new variable to DB Impl where we keep track of the sequence number after which we don't want to drop tombstones, even if they are otherwise eligible for deletion. - I also added a new API call for clients to be able to advance this cutoff seqnum after which we drop deletes; i assume it's more flexible to let clients control this, since otherwise we'd need to keep some kind of timestamp < -- > seqnum mapping inside the DB, which sounds messy and painful to support. Clients could make use of it by periodically calling GetLatestSequenceNumber(), noting the timestamp, doing some calculation and figuring out by how much we need to advance the cutoff seqnum. - Compaction codepath in compaction_iterator.cc has been modified to avoid dropping tombstones with seqnum > cutoff seqnum. Iterator changes: - couple params added to ReadOptions, to optionally allow client to request internal keys instead of user keys (so that client can get the latest value of a key, be it delete marker or a put), as well as min timestamp and min seqnum. TableCache changes: - I modified table_cache code to be able to quickly exclude SST files from iterators heep if creation_time on the file is less then iter_start_ts as passed in ReadOptions. That would help a lot in some DB settings (like reading very recent data only or using FIFO compactions), but not so much for universal compaction with more or less long iterator time span. What's left: - Still looking at how to best plug that inside DBIter codepath. So far it seems that FindNextUserKeyInternal only parses values as UserKeys, and iter->key() call generally returns user key. Can we add new API to DBIter as internal_key(), and modify this internal method to optionally set saved_key_ to point to the full internal key? I don't need to store actual seqnum there, but I do need to store type. Closes https://github.com/facebook/rocksdb/pull/2999 Differential Revision: D6175602 Pulled By: mikhail-antonov fbshipit-source-id: c779a6696ee2d574d86c69cec866a3ae095aa900
2017-11-02 02:43:29 +01:00
// Clients must periodically call SetPreserveDeletesSequenceNumber()
// to advance this seqnum. Default value is 0 which means ALL deletes are
// preserved. Note that this has no effect if DBOptions.preserve_deletes
// is set to false.
std::atomic<SequenceNumber> preserve_deletes_seqnum_;
const bool preserve_deletes_;
// Flag to check whether Close() has been called on this DB
bool closed_;
ErrorHandler error_handler_;
// Conditional variable to coordinate installation of atomic flush results.
// With atomic flush, each bg thread installs the result of flushing multiple
// column families, and different threads can flush different column
// families. It's difficult to rely on one thread to perform batch
// installation for all threads. This is different from the non-atomic flush
// case.
// atomic_flush_install_cv_ makes sure that threads install atomic flush
// results sequentially. Flush results of memtables with lower IDs get
// installed to MANIFEST first.
InstrumentedCondVar atomic_flush_install_cv_;
bool wal_in_db_path_;
};
extern Options SanitizeOptions(const std::string& db, const Options& src);
extern DBOptions SanitizeOptions(const std::string& db, const DBOptions& src);
extern CompressionType GetCompressionFlush(
const ImmutableCFOptions& ioptions,
const MutableCFOptions& mutable_cf_options);
Skip deleted WALs during recovery Summary: This patch record min log number to keep to the manifest while flushing SST files to ignore them and any WAL older than them during recovery. This is to avoid scenarios when we have a gap between the WAL files are fed to the recovery procedure. The gap could happen by for example out-of-order WAL deletion. Such gap could cause problems in 2PC recovery where the prepared and commit entry are placed into two separate WAL and gap in the WALs could result into not processing the WAL with the commit entry and hence breaking the 2PC recovery logic. Before the commit, for 2PC case, we determined which log number to keep in FindObsoleteFiles(). We looked at the earliest logs with outstanding prepare entries, or prepare entries whose respective commit or abort are in memtable. With the commit, the same calculation is done while we apply the SST flush. Just before installing the flush file, we precompute the earliest log file to keep after the flush finishes using the same logic (but skipping the memtables just flushed), record this information to the manifest entry for this new flushed SST file. This pre-computed value is also remembered in memory, and will later be used to determine whether a log file can be deleted. This value is unlikely to change until next flush because the commit entry will stay in memtable. (In WritePrepared, we could have removed the older log files as soon as all prepared entries are committed. It's not yet done anyway. Even if we do it, the only thing we loss with this new approach is earlier log deletion between two flushes, which does not guarantee to happen anyway because the obsolete file clean-up function is only executed after flush or compaction) This min log number to keep is stored in the manifest using the safely-ignore customized field of AddFile entry, in order to guarantee that the DB generated using newer release can be opened by previous releases no older than 4.2. Closes https://github.com/facebook/rocksdb/pull/3765 Differential Revision: D7747618 Pulled By: siying fbshipit-source-id: d00c92105b4f83852e9754a1b70d6b64cb590729
2018-05-04 00:35:11 +02:00
// Return the earliest log file to keep after the memtable flush is
// finalized.
// `cfd_to_flush` is the column family whose memtable (specified in
// `memtables_to_flush`) will be flushed and thus will not depend on any WAL
// file.
// The function is only applicable to 2pc mode.
extern uint64_t PrecomputeMinLogNumberToKeep(
VersionSet* vset, const ColumnFamilyData& cfd_to_flush,
autovector<VersionEdit*> edit_list,
const autovector<MemTable*>& memtables_to_flush,
LogsWithPrepTracker* prep_tracker);
// `cfd_to_flush` is the column family whose memtable will be flushed and thus
// will not depend on any WAL file. nullptr means no memtable is being flushed.
// The function is only applicable to 2pc mode.
extern uint64_t FindMinPrepLogReferencedByMemTable(
VersionSet* vset, const ColumnFamilyData* cfd_to_flush,
const autovector<MemTable*>& memtables_to_flush);
// Fix user-supplied options to be reasonable
template <class T, class V>
static void ClipToRange(T* ptr, V minvalue, V maxvalue) {
if (static_cast<V>(*ptr) > maxvalue) *ptr = maxvalue;
if (static_cast<V>(*ptr) < minvalue) *ptr = minvalue;
}
} // namespace ROCKSDB_NAMESPACE