rocksdb/db/version_edit_handler.cc

981 lines
34 KiB
C++
Raw Permalink 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.
#include "db/version_edit_handler.h"
#include <cinttypes>
#include <sstream>
#include "db/blob/blob_file_cache.h"
#include "db/blob/blob_file_reader.h"
#include "logging/logging.h"
#include "monitoring/persistent_stats_history.h"
namespace ROCKSDB_NAMESPACE {
void VersionEditHandlerBase::Iterate(log::Reader& reader,
Status* log_read_status) {
Slice record;
std::string scratch;
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
assert(log_read_status);
assert(log_read_status->ok());
size_t recovered_edits = 0;
Status s = Initialize();
while (reader.LastRecordEnd() < max_manifest_read_size_ && s.ok() &&
reader.ReadRecord(&record, &scratch) && log_read_status->ok()) {
VersionEdit edit;
s = edit.DecodeFrom(record);
if (!s.ok()) {
break;
}
s = read_buffer_.AddEdit(&edit);
if (!s.ok()) {
break;
}
ColumnFamilyData* cfd = nullptr;
if (edit.is_in_atomic_group_) {
if (read_buffer_.IsFull()) {
for (auto& e : read_buffer_.replay_buffer()) {
s = ApplyVersionEdit(e, &cfd);
if (!s.ok()) {
break;
}
++recovered_edits;
}
if (!s.ok()) {
break;
}
read_buffer_.Clear();
}
} else {
s = ApplyVersionEdit(edit, &cfd);
if (s.ok()) {
++recovered_edits;
}
}
}
Fail recovery when MANIFEST record checksum mismatch (#6996) Summary: https://github.com/facebook/rocksdb/issues/5411 refactored `VersionSet::Recover` but introduced a bug, explained as follows. Before, once a checksum mismatch happens, `reporter` will set `s` to be non-ok. Therefore, Recover will stop processing the MANIFEST any further. ``` // Correct // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); while (reader.ReadRecord() && s.ok()) { ... } ``` The bug is that, the local variable `s` in `ReadAndRecover` won't be updated by `reporter` while reading the MANIFEST. It is possible that the reader sees a checksum mismatch in a record, but `ReadRecord` retries internally read and finds the next valid record. The mismatched record will be ignored and no error is reported. ``` // Incorrect // Inside Recover LogReporter reporter; reporter.status = &s; log::Reader reader(..., reporter); s = ReadAndRecover(reader, ...); // Inside ReadAndRecover Status s; // Shadows the s in Recover. while (reader.ReadRecord() && s.ok()) { ... } ``` `LogReporter` can use a separate `log_read_status` to track the errors while reading the MANIFEST. RocksDB can process more MANIFEST entries only if `log_read_status.ok()`. Test plan (devserver): make check Pull Request resolved: https://github.com/facebook/rocksdb/pull/6996 Reviewed By: ajkr Differential Revision: D22105746 Pulled By: riversand963 fbshipit-source-id: b22f717a423457a41ca152a242abbb64cf91fc38
2020-06-18 19:07:42 +02:00
if (!log_read_status->ok()) {
s = *log_read_status;
}
CheckIterationResult(reader, &s);
if (!s.ok()) {
if (s.IsCorruption()) {
// when we find a Corruption error, something is
// wrong with the underlying file. in this case we
// want to report the filename, so in here we append
// the filename to the Corruption message
assert(reader.file());
// build a new error message
std::stringstream message;
// append previous dynamic state message
const char* state = s.getState();
if (state != nullptr) {
message << state;
message << ' ';
}
// append the filename to the corruption message
message << "in file " << reader.file()->file_name();
// overwrite the status with the extended status
s = Status(s.code(), s.subcode(), s.severity(), message.str());
}
status_ = s;
}
TEST_SYNC_POINT_CALLBACK("VersionEditHandlerBase::Iterate:Finish",
&recovered_edits);
}
Status ListColumnFamiliesHandler::ApplyVersionEdit(
VersionEdit& edit, ColumnFamilyData** /*unused*/) {
Status s;
if (edit.is_column_family_add_) {
if (column_family_names_.find(edit.column_family_) !=
column_family_names_.end()) {
s = Status::Corruption("Manifest adding the same column family twice");
} else {
column_family_names_.insert(
{edit.column_family_, edit.column_family_name_});
}
} else if (edit.is_column_family_drop_) {
if (column_family_names_.find(edit.column_family_) ==
column_family_names_.end()) {
s = Status::Corruption("Manifest - dropping non-existing column family");
} else {
column_family_names_.erase(edit.column_family_);
}
}
return s;
}
Status FileChecksumRetriever::ApplyVersionEdit(VersionEdit& edit,
ColumnFamilyData** /*unused*/) {
for (const auto& deleted_file : edit.GetDeletedFiles()) {
Status s = file_checksum_list_.RemoveOneFileChecksum(deleted_file.second);
if (!s.ok()) {
return s;
}
}
for (const auto& new_file : edit.GetNewFiles()) {
Status s = file_checksum_list_.InsertOneFileChecksum(
new_file.second.fd.GetNumber(), new_file.second.file_checksum,
new_file.second.file_checksum_func_name);
if (!s.ok()) {
return s;
}
}
for (const auto& new_blob_file : edit.GetBlobFileAdditions()) {
std::string checksum_value = new_blob_file.GetChecksumValue();
std::string checksum_method = new_blob_file.GetChecksumMethod();
assert(checksum_value.empty() == checksum_method.empty());
if (checksum_method.empty()) {
checksum_value = kUnknownFileChecksum;
checksum_method = kUnknownFileChecksumFuncName;
}
Status s = file_checksum_list_.InsertOneFileChecksum(
new_blob_file.GetBlobFileNumber(), checksum_value, checksum_method);
if (!s.ok()) {
return s;
}
}
return Status::OK();
}
VersionEditHandler::VersionEditHandler(
bool read_only, std::vector<ColumnFamilyDescriptor> column_families,
VersionSet* version_set, bool track_missing_files,
bool no_error_if_files_missing, const std::shared_ptr<IOTracer>& io_tracer,
bool skip_load_table_files)
: VersionEditHandlerBase(),
read_only_(read_only),
column_families_(std::move(column_families)),
version_set_(version_set),
track_missing_files_(track_missing_files),
no_error_if_files_missing_(no_error_if_files_missing),
io_tracer_(io_tracer),
skip_load_table_files_(skip_load_table_files),
initialized_(false) {
assert(version_set_ != nullptr);
}
Status VersionEditHandler::Initialize() {
Status s;
if (!initialized_) {
for (const auto& cf_desc : column_families_) {
name_to_options_.emplace(cf_desc.name, cf_desc.options);
}
auto default_cf_iter = name_to_options_.find(kDefaultColumnFamilyName);
if (default_cf_iter == name_to_options_.end()) {
s = Status::InvalidArgument("Default column family not specified");
}
if (s.ok()) {
VersionEdit default_cf_edit;
default_cf_edit.AddColumnFamily(kDefaultColumnFamilyName);
default_cf_edit.SetColumnFamily(0);
ColumnFamilyData* cfd =
CreateCfAndInit(default_cf_iter->second, default_cf_edit);
assert(cfd != nullptr);
#ifdef NDEBUG
(void)cfd;
#endif
initialized_ = true;
}
}
return s;
}
Status VersionEditHandler::ApplyVersionEdit(VersionEdit& edit,
ColumnFamilyData** cfd) {
Status s;
if (edit.is_column_family_add_) {
s = OnColumnFamilyAdd(edit, cfd);
} else if (edit.is_column_family_drop_) {
s = OnColumnFamilyDrop(edit, cfd);
} else if (edit.IsWalAddition()) {
s = OnWalAddition(edit);
} else if (edit.IsWalDeletion()) {
s = OnWalDeletion(edit);
} else {
s = OnNonCfOperation(edit, cfd);
}
if (s.ok()) {
assert(cfd != nullptr);
s = ExtractInfoFromVersionEdit(*cfd, edit);
}
return s;
}
Status VersionEditHandler::OnColumnFamilyAdd(VersionEdit& edit,
ColumnFamilyData** cfd) {
bool cf_in_not_found = false;
bool cf_in_builders = false;
CheckColumnFamilyId(edit, &cf_in_not_found, &cf_in_builders);
assert(cfd != nullptr);
*cfd = nullptr;
Status s;
if (cf_in_builders || cf_in_not_found) {
s = Status::Corruption("MANIFEST adding the same column family twice: " +
edit.column_family_name_);
}
if (s.ok()) {
auto cf_options = name_to_options_.find(edit.column_family_name_);
// implicitly add persistent_stats column family without requiring user
// to specify
ColumnFamilyData* tmp_cfd = nullptr;
bool is_persistent_stats_column_family =
edit.column_family_name_.compare(kPersistentStatsColumnFamilyName) == 0;
if (cf_options == name_to_options_.end() &&
!is_persistent_stats_column_family) {
column_families_not_found_.emplace(edit.column_family_,
edit.column_family_name_);
} else {
if (is_persistent_stats_column_family) {
ColumnFamilyOptions cfo;
OptimizeForPersistentStats(&cfo);
tmp_cfd = CreateCfAndInit(cfo, edit);
} else {
tmp_cfd = CreateCfAndInit(cf_options->second, edit);
}
*cfd = tmp_cfd;
}
}
return s;
}
Status VersionEditHandler::OnColumnFamilyDrop(VersionEdit& edit,
ColumnFamilyData** cfd) {
bool cf_in_not_found = false;
bool cf_in_builders = false;
CheckColumnFamilyId(edit, &cf_in_not_found, &cf_in_builders);
assert(cfd != nullptr);
*cfd = nullptr;
ColumnFamilyData* tmp_cfd = nullptr;
Status s;
if (cf_in_builders) {
tmp_cfd = DestroyCfAndCleanup(edit);
} else if (cf_in_not_found) {
column_families_not_found_.erase(edit.column_family_);
} else {
s = Status::Corruption("MANIFEST - dropping non-existing column family");
}
*cfd = tmp_cfd;
return s;
}
Status VersionEditHandler::OnWalAddition(VersionEdit& edit) {
assert(edit.IsWalAddition());
return version_set_->wals_.AddWals(edit.GetWalAdditions());
}
Status VersionEditHandler::OnWalDeletion(VersionEdit& edit) {
assert(edit.IsWalDeletion());
return version_set_->wals_.DeleteWalsBefore(
edit.GetWalDeletion().GetLogNumber());
}
Status VersionEditHandler::OnNonCfOperation(VersionEdit& edit,
ColumnFamilyData** cfd) {
bool cf_in_not_found = false;
bool cf_in_builders = false;
CheckColumnFamilyId(edit, &cf_in_not_found, &cf_in_builders);
assert(cfd != nullptr);
*cfd = nullptr;
Status s;
if (!cf_in_not_found) {
if (!cf_in_builders) {
s = Status::Corruption(
"MANIFEST record referencing unknown column family");
}
ColumnFamilyData* tmp_cfd = nullptr;
if (s.ok()) {
auto builder_iter = builders_.find(edit.column_family_);
assert(builder_iter != builders_.end());
tmp_cfd = version_set_->GetColumnFamilySet()->GetColumnFamily(
edit.column_family_);
assert(tmp_cfd != nullptr);
s = MaybeCreateVersion(edit, tmp_cfd, /*force_create_version=*/false);
if (s.ok()) {
s = builder_iter->second->version_builder()->Apply(&edit);
}
}
*cfd = tmp_cfd;
}
return s;
}
// TODO maybe cache the computation result
bool VersionEditHandler::HasMissingFiles() const {
bool ret = false;
for (const auto& elem : cf_to_missing_files_) {
const auto& missing_files = elem.second;
if (!missing_files.empty()) {
ret = true;
break;
}
}
if (!ret) {
for (const auto& elem : cf_to_missing_blob_files_high_) {
if (elem.second != kInvalidBlobFileNumber) {
ret = true;
break;
}
}
}
return ret;
}
void VersionEditHandler::CheckColumnFamilyId(const VersionEdit& edit,
bool* cf_in_not_found,
bool* cf_in_builders) const {
assert(cf_in_not_found != nullptr);
assert(cf_in_builders != nullptr);
// Not found means that user didn't supply that column
// family option AND we encountered column family add
// record. Once we encounter column family drop record,
// we will delete the column family from
// column_families_not_found.
bool in_not_found = column_families_not_found_.find(edit.column_family_) !=
column_families_not_found_.end();
// in builders means that user supplied that column family
// option AND that we encountered column family add record
bool in_builders = builders_.find(edit.column_family_) != builders_.end();
// They cannot both be true
assert(!(in_not_found && in_builders));
*cf_in_not_found = in_not_found;
*cf_in_builders = in_builders;
}
void VersionEditHandler::CheckIterationResult(const log::Reader& reader,
Status* s) {
assert(s != nullptr);
if (!s->ok()) {
// Do nothing here.
} else if (!version_edit_params_.has_log_number_ ||
!version_edit_params_.has_next_file_number_ ||
!version_edit_params_.has_last_sequence_) {
std::string msg("no ");
if (!version_edit_params_.has_log_number_) {
msg.append("log_file_number, ");
}
if (!version_edit_params_.has_next_file_number_) {
msg.append("next_file_number, ");
}
if (!version_edit_params_.has_last_sequence_) {
msg.append("last_sequence, ");
}
msg = msg.substr(0, msg.size() - 2);
msg.append(" entry in MANIFEST");
*s = Status::Corruption(msg);
}
// There were some column families in the MANIFEST that weren't specified
// in the argument. This is OK in read_only mode
if (s->ok() && MustOpenAllColumnFamilies() &&
!column_families_not_found_.empty()) {
std::string msg;
for (const auto& cf : column_families_not_found_) {
msg.append(", ");
msg.append(cf.second);
}
msg = msg.substr(2);
*s = Status::InvalidArgument("Column families not opened: " + msg);
}
if (s->ok()) {
version_set_->GetColumnFamilySet()->UpdateMaxColumnFamily(
version_edit_params_.max_column_family_);
Fix a race condition in WAL tracking causing DB open failure (#9715) Summary: There is a race condition if WAL tracking in the MANIFEST is enabled in a database that disables 2PC. The race condition is between two background flush threads trying to install flush results to the MANIFEST. Consider an example database with two column families: "default" (cfd0) and "cf1" (cfd1). Initially, both column families have one mutable (active) memtable whose data backed by 6.log. 1. Trigger a manual flush for "cf1", creating a 7.log 2. Insert another key to "default", and trigger flush for "default", creating 8.log 3. BgFlushThread1 finishes writing 9.sst 4. BgFlushThread2 finishes writing 10.sst ``` Time BgFlushThread1 BgFlushThread2 | mutex_.Lock() | precompute min_wal_to_keep as 6 | mutex_.Unlock() | mutex_.Lock() | precompute min_wal_to_keep as 6 | join MANIFEST write queue and mutex_.Unlock() | write to MANIFEST | mutex_.Lock() | cfd1->log_number = 7 | Signal bg_flush_2 and mutex_.Unlock() | wake up and mutex_.Lock() | cfd0->log_number = 8 | FindObsoleteFiles() with job_context->log_number == 7 | mutex_.Unlock() | PurgeObsoleteFiles() deletes 6.log V ``` As shown in the above, BgFlushThread2 thinks that the min wal to keep is 6.log because "cf1" has unflushed data in 6.log (cf1.log_number=6). Similarly, BgThread1 thinks that min wal to keep is also 6.log because "default" has unflushed data (default.log_number=6). No WAL deletion will be written to MANIFEST because 6 is equal to `versions_->wals_.min_wal_number_to_keep`, due to https://github.com/facebook/rocksdb/blob/7.1.fb/db/memtable_list.cc#L513:L514. The bg flush thread that finishes last will perform file purging. `job_context.log_number` will be evaluated as 7, i.e. the min wal that contains unflushed data, causing 6.log to be deleted. However, MANIFEST thinks 6.log should still exist. If you close the db at this point, you won't be able to re-open it if `track_and_verify_wal_in_manifest` is true. We must handle the case of multiple bg flush threads, and it is difficult for one bg flush thread to know the correct min wal number until the other bg flush threads have finished committing to the manifest and updated the `cfd::log_number`. To fix this issue, we rename an existing variable `min_log_number_to_keep_2pc` to `min_log_number_to_keep`, and use it to track WAL file deletion in non-2pc mode as well. This variable is updated only 1) during recovery with mutex held, or 2) in the MANIFEST write thread. `min_log_number_to_keep` means RocksDB will delete WALs below it, although there may be WALs above it which are also obsolete. Formally, we will have [min_wal_to_keep, max_obsolete_wal]. During recovery, we make sure that only WALs above max_obsolete_wal are checked and added back to `alive_log_files_`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9715 Test Plan: ``` make check ``` Also ran stress test below (with asan) to make sure it completes successfully. ``` TEST_TMPDIR=/dev/shm/rocksdb OPT=-g ASAN_OPTIONS=disable_coredump=0 \ CRASH_TEST_EXT_ARGS=--compression_type=zstd SKIP_FORMAT_BUCK_CHECKS=1 \ make J=52 -j52 blackbox_asan_crash_test ``` Reviewed By: ltamasi Differential Revision: D34984412 Pulled By: riversand963 fbshipit-source-id: c7b21a8d84751bb55ea79c9f387103d21b231005
2022-03-24 03:41:31 +01:00
version_set_->MarkMinLogNumberToKeep(
version_edit_params_.min_log_number_to_keep_);
version_set_->MarkFileNumberUsed(version_edit_params_.prev_log_number_);
version_set_->MarkFileNumberUsed(version_edit_params_.log_number_);
for (auto* cfd : *(version_set_->GetColumnFamilySet())) {
if (cfd->IsDropped()) {
continue;
}
auto builder_iter = builders_.find(cfd->GetID());
assert(builder_iter != builders_.end());
auto* builder = builder_iter->second->version_builder();
if (!builder->CheckConsistencyForNumLevels()) {
*s = Status::InvalidArgument(
"db has more levels than options.num_levels");
break;
}
}
}
if (s->ok()) {
for (auto* cfd : *(version_set_->GetColumnFamilySet())) {
if (cfd->IsDropped()) {
continue;
}
if (read_only_) {
cfd->table_cache()->SetTablesAreImmortal();
}
*s = LoadTables(cfd, /*prefetch_index_and_filter_in_cache=*/false,
/*is_initial_load=*/true);
if (!s->ok()) {
// If s is IOError::PathNotFound, then we mark the db as corrupted.
if (s->IsPathNotFound()) {
*s = Status::Corruption("Corruption: " + s->ToString());
}
break;
}
}
}
if (s->ok()) {
for (auto* cfd : *(version_set_->column_family_set_)) {
if (cfd->IsDropped()) {
continue;
}
assert(cfd->initialized());
VersionEdit edit;
*s = MaybeCreateVersion(edit, cfd, /*force_create_version=*/true);
if (!s->ok()) {
break;
}
}
}
if (s->ok()) {
version_set_->manifest_file_size_ = reader.GetReadOffset();
assert(version_set_->manifest_file_size_ > 0);
version_set_->next_file_number_.store(
version_edit_params_.next_file_number_ + 1);
SequenceNumber last_seq = version_edit_params_.last_sequence_;
assert(last_seq != kMaxSequenceNumber);
if (last_seq != kMaxSequenceNumber &&
last_seq > version_set_->last_allocated_sequence_.load()) {
version_set_->last_allocated_sequence_.store(last_seq);
}
if (last_seq != kMaxSequenceNumber &&
last_seq > version_set_->last_published_sequence_.load()) {
version_set_->last_published_sequence_.store(last_seq);
}
if (last_seq != kMaxSequenceNumber &&
last_seq > version_set_->last_sequence_.load()) {
version_set_->last_sequence_.store(last_seq);
}
Recover to exact latest seqno of data committed to MANIFEST (#9305) Summary: The LastSequence field in the MANIFEST file is the baseline seqno for a recovered DB. Recovering WAL entries might cause the recovered DB's seqno to advance above this baseline, but the recovered DB will never use a smaller seqno. Before this PR, we were writing the DB's seqno at the time of LogAndApply() as the LastSequence value. This works in the sense that it is a large enough baseline for the recovered DB that it'll never overwrite any records in existing SST files. At the same time, it's arbitrarily larger than what's needed. This behavior comes from LevelDB, where there was no tracking of largest seqno in an SST file. Now we know the largest seqno of newly written SST files, so we can write an exact value in LastSequence that actually reflects the largest seqno in any file referred to by the MANIFEST. This is primarily useful for correctness testing with unsynced data loss, where the recovered DB's seqno needs to indicate what records were recovered. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9305 Test Plan: - https://github.com/facebook/rocksdb/issues/9338 adds crash-recovery correctness testing coverage for WAL disabled use cases - https://github.com/facebook/rocksdb/issues/9357 will extend that testing to cover file ingestion - Added assertion at end of LogAndApply() for `VersionSet::descriptor_last_sequence_` consistency with files - Manually tested upgrade/downgrade compatibility with a custom crash test that randomly picks between a `db_stress` built with and without this PR (for old code it must run with `-disable_wal=0`) Reviewed By: riversand963 Differential Revision: D33182770 Pulled By: ajkr fbshipit-source-id: 0bfafaf685f347cc8cb0e1d62e0186340a738f7d
2022-01-06 01:00:41 +01:00
if (last_seq != kMaxSequenceNumber &&
last_seq > version_set_->descriptor_last_sequence_) {
// This is the maximum last sequence of all `VersionEdit`s iterated. It
// may be greater than the maximum `largest_seqno` of all files in case
// the newest data referred to by the MANIFEST has been dropped or had its
// sequence number zeroed through compaction.
version_set_->descriptor_last_sequence_ = last_seq;
}
version_set_->prev_log_number_ = version_edit_params_.prev_log_number_;
}
}
ColumnFamilyData* VersionEditHandler::CreateCfAndInit(
const ColumnFamilyOptions& cf_options, const VersionEdit& edit) {
ColumnFamilyData* cfd = version_set_->CreateColumnFamily(cf_options, &edit);
assert(cfd != nullptr);
cfd->set_initialized();
assert(builders_.find(edit.column_family_) == builders_.end());
builders_.emplace(edit.column_family_,
VersionBuilderUPtr(new BaseReferencedVersionBuilder(cfd)));
if (track_missing_files_) {
cf_to_missing_files_.emplace(edit.column_family_,
std::unordered_set<uint64_t>());
cf_to_missing_blob_files_high_.emplace(edit.column_family_,
kInvalidBlobFileNumber);
}
return cfd;
}
ColumnFamilyData* VersionEditHandler::DestroyCfAndCleanup(
const VersionEdit& edit) {
auto builder_iter = builders_.find(edit.column_family_);
assert(builder_iter != builders_.end());
builders_.erase(builder_iter);
if (track_missing_files_) {
auto missing_files_iter = cf_to_missing_files_.find(edit.column_family_);
assert(missing_files_iter != cf_to_missing_files_.end());
cf_to_missing_files_.erase(missing_files_iter);
auto missing_blob_files_high_iter =
cf_to_missing_blob_files_high_.find(edit.column_family_);
assert(missing_blob_files_high_iter !=
cf_to_missing_blob_files_high_.end());
cf_to_missing_blob_files_high_.erase(missing_blob_files_high_iter);
}
ColumnFamilyData* ret =
version_set_->GetColumnFamilySet()->GetColumnFamily(edit.column_family_);
assert(ret != nullptr);
ret->SetDropped();
ret->UnrefAndTryDelete();
ret = nullptr;
return ret;
}
Status VersionEditHandler::MaybeCreateVersion(const VersionEdit& /*edit*/,
ColumnFamilyData* cfd,
bool force_create_version) {
assert(cfd->initialized());
Status s;
if (force_create_version) {
auto builder_iter = builders_.find(cfd->GetID());
assert(builder_iter != builders_.end());
auto* builder = builder_iter->second->version_builder();
auto* v = new Version(cfd, version_set_, version_set_->file_options_,
*cfd->GetLatestMutableCFOptions(), io_tracer_,
version_set_->current_version_number_++);
s = builder->SaveTo(v->storage_info());
if (s.ok()) {
// Install new version
v->PrepareApply(
*cfd->GetLatestMutableCFOptions(),
!(version_set_->db_options_->skip_stats_update_on_db_open));
version_set_->AppendVersion(cfd, v);
} else {
delete v;
}
}
return s;
}
Status VersionEditHandler::LoadTables(ColumnFamilyData* cfd,
bool prefetch_index_and_filter_in_cache,
bool is_initial_load) {
Improve / clean up meta block code & integrity (#9163) Summary: * Checksums are now checked on meta blocks unless specifically suppressed or not applicable (e.g. plain table). (Was other way around.) This means a number of cases that were not checking checksums now are, including direct read TableProperties in Version::GetTableProperties (fixed in meta_blocks ReadTableProperties), reading any block from PersistentCache (fixed in BlockFetcher), read TableProperties in SstFileDumper (ldb/sst_dump/BackupEngine) before table reader open, maybe more. * For that to work, I moved the global_seqno+TableProperties checksum logic to the shared table/ code, because that is used by many utilies such as SstFileDumper. * Also for that to work, we have to know when we're dealing with a block that has a checksum (trailer), so added that capability to Footer based on magic number, and from there BlockFetcher. * Knowledge of trailer presence has also fixed a problem where other table formats were reading blocks including bytes for a non-existant trailer--and awkwardly kind-of not using them, e.g. no shared code checking checksums. (BlockFetcher compression type was populated incorrectly.) Now we only read what is needed. * Minimized code duplication and differing/incompatible/awkward abstractions in meta_blocks.{cc,h} (e.g. SeekTo in metaindex block without parsing block handle) * Moved some meta block handling code from table_properties*.* * Moved some code specific to block-based table from shared table/ code to BlockBasedTable class. The checksum stuff means we can't completely separate it, but things that don't need to be in shared table/ code should not be. * Use unique_ptr rather than raw ptr in more places. (Note: you can std::move from unique_ptr to shared_ptr.) Without enhancements to GetPropertiesOfAllTablesTest (see below), net reduction of roughly 100 lines of code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9163 Test Plan: existing tests and * Enhanced DBTablePropertiesTest.GetPropertiesOfAllTablesTest to verify that checksums are now checked on direct read of table properties by TableCache (new test would fail before this change) * Also enhanced DBTablePropertiesTest.GetPropertiesOfAllTablesTest to test putting table properties under old meta name * Also generally enhanced that same test to actually test what it was supposed to be testing already, by kicking things out of table cache when we don't want them there. Reviewed By: ajkr, mrambacher Differential Revision: D32514757 Pulled By: pdillinger fbshipit-source-id: 507964b9311d186ae8d1131182290cbd97a99fa9
2021-11-18 20:42:12 +01:00
bool skip_load_table_files = skip_load_table_files_;
TEST_SYNC_POINT_CALLBACK(
"VersionEditHandler::LoadTables:skip_load_table_files",
&skip_load_table_files);
if (skip_load_table_files) {
return Status::OK();
}
assert(cfd != nullptr);
assert(!cfd->IsDropped());
auto builder_iter = builders_.find(cfd->GetID());
assert(builder_iter != builders_.end());
assert(builder_iter->second != nullptr);
VersionBuilder* builder = builder_iter->second->version_builder();
assert(builder);
Status s = builder->LoadTableHandlers(
cfd->internal_stats(),
version_set_->db_options_->max_file_opening_threads,
prefetch_index_and_filter_in_cache, is_initial_load,
Fast path for detecting unchanged prefix_extractor (#9407) Summary: Fixes a major performance regression in 6.26, where extra CPU is spent in SliceTransform::AsString when reads involve a prefix_extractor (Get, MultiGet, Seek). Common case performance is now better than 6.25. This change creates a "fast path" for verifying that the current prefix extractor is unchanged and compatible with what was used to generate a table file. This fast path detects the common case by pointer comparison on the current prefix_extractor and a "known good" prefix extractor (if applicable) that is saved at the time the table reader is opened. The "known good" prefix extractor is saved as another shared_ptr copy (in an existing field, however) to ensure the pointer is not recycled. When the prefix_extractor has changed to a different instance but same compatible configuration (rare, odd), performance is still a regression compared to 6.25, but this is likely acceptable because of the oddity of such a case. The performance of incompatible prefix_extractor is essentially unchanged. Also fixed a minor case (ForwardIterator) where a prefix_extractor could be used via a raw pointer after being freed as a shared_ptr, if replaced via SetOptions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9407 Test Plan: ## Performance Populate DB with `TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=12` Running head-to-head comparisons simultaneously with `TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -use_existing_db -readonly -benchmarks=seekrandom -num=10000000 -duration=20 -disable_wal=1 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=12` Below each is compared by ops/sec vs. baseline which is version 6.25 (multiple baseline runs because of variable machine load) v6.26: 4833 vs. 6698 (<- major regression!) v6.27: 4737 vs. 6397 (still) New: 6704 vs. 6461 (better than baseline in common case) Disabled fastpath: 4843 vs. 6389 (e.g. if prefix extractor instance changes but is still compatible) Changed prefix size (no usable filter) in new: 787 vs. 5927 Changed prefix size (no usable filter) in new & baseline: 773 vs. 784 Reviewed By: mrambacher Differential Revision: D33677812 Pulled By: pdillinger fbshipit-source-id: 571d9711c461fb97f957378a061b7e7dbc4d6a76
2022-01-21 20:36:36 +01:00
cfd->GetLatestMutableCFOptions()->prefix_extractor,
MaxFileSizeForL0MetaPin(*cfd->GetLatestMutableCFOptions()));
if ((s.IsPathNotFound() || s.IsCorruption()) && no_error_if_files_missing_) {
s = Status::OK();
}
if (!s.ok() && !version_set_->db_options_->paranoid_checks) {
s = Status::OK();
}
return s;
}
Status VersionEditHandler::ExtractInfoFromVersionEdit(ColumnFamilyData* cfd,
const VersionEdit& edit) {
Status s;
if (edit.has_db_id_) {
version_set_->db_id_ = edit.GetDbId();
version_edit_params_.SetDBId(edit.db_id_);
}
if (cfd != nullptr) {
if (edit.has_log_number_) {
if (cfd->GetLogNumber() > edit.log_number_) {
ROCKS_LOG_WARN(
version_set_->db_options()->info_log,
"MANIFEST corruption detected, but ignored - Log numbers in "
"records NOT monotonically increasing");
} else {
cfd->SetLogNumber(edit.log_number_);
version_edit_params_.SetLogNumber(edit.log_number_);
}
}
if (edit.has_comparator_ &&
edit.comparator_ != cfd->user_comparator()->Name()) {
if (!cf_to_cmp_names_) {
s = Status::InvalidArgument(
cfd->user_comparator()->Name(),
"does not match existing comparator " + edit.comparator_);
} else {
cf_to_cmp_names_->emplace(cfd->GetID(), edit.comparator_);
}
}
if (edit.HasFullHistoryTsLow()) {
const std::string& new_ts = edit.GetFullHistoryTsLow();
cfd->SetFullHistoryTsLow(new_ts);
}
}
if (s.ok()) {
if (edit.has_prev_log_number_) {
version_edit_params_.SetPrevLogNumber(edit.prev_log_number_);
}
if (edit.has_next_file_number_) {
version_edit_params_.SetNextFile(edit.next_file_number_);
}
if (edit.has_max_column_family_) {
version_edit_params_.SetMaxColumnFamily(edit.max_column_family_);
}
if (edit.has_min_log_number_to_keep_) {
version_edit_params_.min_log_number_to_keep_ =
std::max(version_edit_params_.min_log_number_to_keep_,
edit.min_log_number_to_keep_);
}
if (edit.has_last_sequence_) {
Recover to exact latest seqno of data committed to MANIFEST (#9305) Summary: The LastSequence field in the MANIFEST file is the baseline seqno for a recovered DB. Recovering WAL entries might cause the recovered DB's seqno to advance above this baseline, but the recovered DB will never use a smaller seqno. Before this PR, we were writing the DB's seqno at the time of LogAndApply() as the LastSequence value. This works in the sense that it is a large enough baseline for the recovered DB that it'll never overwrite any records in existing SST files. At the same time, it's arbitrarily larger than what's needed. This behavior comes from LevelDB, where there was no tracking of largest seqno in an SST file. Now we know the largest seqno of newly written SST files, so we can write an exact value in LastSequence that actually reflects the largest seqno in any file referred to by the MANIFEST. This is primarily useful for correctness testing with unsynced data loss, where the recovered DB's seqno needs to indicate what records were recovered. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9305 Test Plan: - https://github.com/facebook/rocksdb/issues/9338 adds crash-recovery correctness testing coverage for WAL disabled use cases - https://github.com/facebook/rocksdb/issues/9357 will extend that testing to cover file ingestion - Added assertion at end of LogAndApply() for `VersionSet::descriptor_last_sequence_` consistency with files - Manually tested upgrade/downgrade compatibility with a custom crash test that randomly picks between a `db_stress` built with and without this PR (for old code it must run with `-disable_wal=0`) Reviewed By: riversand963 Differential Revision: D33182770 Pulled By: ajkr fbshipit-source-id: 0bfafaf685f347cc8cb0e1d62e0186340a738f7d
2022-01-06 01:00:41 +01:00
// `VersionEdit::last_sequence_`s are assumed to be non-decreasing. This
// is legacy behavior that cannot change without breaking downgrade
// compatibility.
assert(!version_edit_params_.has_last_sequence_ ||
version_edit_params_.last_sequence_ <= edit.last_sequence_);
version_edit_params_.SetLastSequence(edit.last_sequence_);
}
if (!version_edit_params_.has_prev_log_number_) {
version_edit_params_.SetPrevLogNumber(0);
}
}
return s;
}
VersionEditHandlerPointInTime::VersionEditHandlerPointInTime(
bool read_only, std::vector<ColumnFamilyDescriptor> column_families,
VersionSet* version_set, const std::shared_ptr<IOTracer>& io_tracer)
: VersionEditHandler(read_only, column_families, version_set,
/*track_missing_files=*/true,
/*no_error_if_files_missing=*/true, io_tracer) {}
VersionEditHandlerPointInTime::~VersionEditHandlerPointInTime() {
for (const auto& elem : versions_) {
delete elem.second;
}
versions_.clear();
}
void VersionEditHandlerPointInTime::CheckIterationResult(
const log::Reader& reader, Status* s) {
VersionEditHandler::CheckIterationResult(reader, s);
assert(s != nullptr);
if (s->ok()) {
for (auto* cfd : *(version_set_->column_family_set_)) {
if (cfd->IsDropped()) {
continue;
}
assert(cfd->initialized());
auto v_iter = versions_.find(cfd->GetID());
if (v_iter != versions_.end()) {
assert(v_iter->second != nullptr);
version_set_->AppendVersion(cfd, v_iter->second);
versions_.erase(v_iter);
}
}
} else {
for (const auto& elem : versions_) {
delete elem.second;
}
versions_.clear();
}
}
ColumnFamilyData* VersionEditHandlerPointInTime::DestroyCfAndCleanup(
const VersionEdit& edit) {
ColumnFamilyData* cfd = VersionEditHandler::DestroyCfAndCleanup(edit);
auto v_iter = versions_.find(edit.column_family_);
if (v_iter != versions_.end()) {
delete v_iter->second;
versions_.erase(v_iter);
}
return cfd;
}
Status VersionEditHandlerPointInTime::MaybeCreateVersion(
const VersionEdit& edit, ColumnFamilyData* cfd, bool force_create_version) {
assert(cfd != nullptr);
if (!force_create_version) {
assert(edit.column_family_ == cfd->GetID());
}
auto missing_files_iter = cf_to_missing_files_.find(cfd->GetID());
assert(missing_files_iter != cf_to_missing_files_.end());
std::unordered_set<uint64_t>& missing_files = missing_files_iter->second;
auto missing_blob_files_high_iter =
cf_to_missing_blob_files_high_.find(cfd->GetID());
assert(missing_blob_files_high_iter != cf_to_missing_blob_files_high_.end());
const uint64_t prev_missing_blob_file_high =
missing_blob_files_high_iter->second;
VersionBuilder* builder = nullptr;
if (prev_missing_blob_file_high != kInvalidBlobFileNumber) {
auto builder_iter = builders_.find(cfd->GetID());
assert(builder_iter != builders_.end());
builder = builder_iter->second->version_builder();
assert(builder != nullptr);
}
// At this point, we have not yet applied the new version edits read from the
// MANIFEST. We check whether we have any missing table and blob files.
const bool prev_has_missing_files =
!missing_files.empty() ||
(prev_missing_blob_file_high != kInvalidBlobFileNumber &&
prev_missing_blob_file_high >= builder->GetMinOldestBlobFileNumber());
for (const auto& file : edit.GetDeletedFiles()) {
uint64_t file_num = file.second;
auto fiter = missing_files.find(file_num);
if (fiter != missing_files.end()) {
missing_files.erase(fiter);
}
}
assert(!cfd->ioptions()->cf_paths.empty());
Status s;
for (const auto& elem : edit.GetNewFiles()) {
const FileMetaData& meta = elem.second;
const FileDescriptor& fd = meta.fd;
uint64_t file_num = fd.GetNumber();
const std::string fpath =
MakeTableFileName(cfd->ioptions()->cf_paths[0].path, file_num);
s = VerifyFile(fpath, meta);
if (s.IsPathNotFound() || s.IsNotFound() || s.IsCorruption()) {
missing_files.insert(file_num);
s = Status::OK();
} else if (!s.ok()) {
break;
}
}
uint64_t missing_blob_file_num = prev_missing_blob_file_high;
for (const auto& elem : edit.GetBlobFileAdditions()) {
uint64_t file_num = elem.GetBlobFileNumber();
s = VerifyBlobFile(cfd, file_num, elem);
if (s.IsPathNotFound() || s.IsNotFound() || s.IsCorruption()) {
missing_blob_file_num = std::max(missing_blob_file_num, file_num);
s = Status::OK();
} else if (!s.ok()) {
break;
}
}
bool has_missing_blob_files = false;
if (missing_blob_file_num != kInvalidBlobFileNumber &&
missing_blob_file_num >= prev_missing_blob_file_high) {
missing_blob_files_high_iter->second = missing_blob_file_num;
has_missing_blob_files = true;
} else if (missing_blob_file_num < prev_missing_blob_file_high) {
assert(false);
}
// We still have not applied the new version edit, but have tried to add new
// table and blob files after verifying their presence and consistency.
// Therefore, we know whether we will see new missing table and blob files
// later after actually applying the version edit. We perform the check here
// and record the result.
const bool has_missing_files =
!missing_files.empty() || has_missing_blob_files;
bool missing_info = !version_edit_params_.has_log_number_ ||
!version_edit_params_.has_next_file_number_ ||
!version_edit_params_.has_last_sequence_;
// Create version before apply edit. The version will represent the state
// before applying the version edit.
// A new version will created if:
// 1) no error has occurred so far, and
// 2) log_number_, next_file_number_ and last_sequence_ are known, and
// 3) any of the following:
// a) no missing file before, but will have missing file(s) after applying
// this version edit.
// b) no missing file after applying the version edit, and the caller
// explicitly request that a new version be created.
if (s.ok() && !missing_info &&
((has_missing_files && !prev_has_missing_files) ||
(!has_missing_files && force_create_version))) {
if (!builder) {
auto builder_iter = builders_.find(cfd->GetID());
assert(builder_iter != builders_.end());
builder = builder_iter->second->version_builder();
assert(builder);
}
auto* version = new Version(cfd, version_set_, version_set_->file_options_,
*cfd->GetLatestMutableCFOptions(), io_tracer_,
version_set_->current_version_number_++);
s = builder->SaveTo(version->storage_info());
if (s.ok()) {
version->PrepareApply(
*cfd->GetLatestMutableCFOptions(),
!version_set_->db_options_->skip_stats_update_on_db_open);
auto v_iter = versions_.find(cfd->GetID());
if (v_iter != versions_.end()) {
delete v_iter->second;
v_iter->second = version;
} else {
versions_.emplace(cfd->GetID(), version);
}
} else {
delete version;
}
}
return s;
}
Status VersionEditHandlerPointInTime::VerifyFile(const std::string& fpath,
const FileMetaData& fmeta) {
return version_set_->VerifyFileMetadata(fpath, fmeta);
}
Status VersionEditHandlerPointInTime::VerifyBlobFile(
ColumnFamilyData* cfd, uint64_t blob_file_num,
const BlobFileAddition& blob_addition) {
BlobFileCache* blob_file_cache = cfd->blob_file_cache();
assert(blob_file_cache);
CacheHandleGuard<BlobFileReader> blob_file_reader;
Status s =
blob_file_cache->GetBlobFileReader(blob_file_num, &blob_file_reader);
if (!s.ok()) {
return s;
}
// TODO: verify checksum
(void)blob_addition;
return s;
}
Status ManifestTailer::Initialize() {
if (Mode::kRecovery == mode_) {
return VersionEditHandler::Initialize();
}
assert(Mode::kCatchUp == mode_);
Status s;
if (!initialized_) {
ColumnFamilySet* cfd_set = version_set_->GetColumnFamilySet();
assert(cfd_set);
ColumnFamilyData* default_cfd = cfd_set->GetDefault();
assert(default_cfd);
auto builder_iter = builders_.find(default_cfd->GetID());
assert(builder_iter != builders_.end());
Version* dummy_version = default_cfd->dummy_versions();
assert(dummy_version);
Version* base_version = dummy_version->Next();
assert(base_version);
base_version->Ref();
VersionBuilderUPtr new_builder(
new BaseReferencedVersionBuilder(default_cfd, base_version));
builder_iter->second = std::move(new_builder);
initialized_ = true;
}
return s;
}
Status ManifestTailer::ApplyVersionEdit(VersionEdit& edit,
ColumnFamilyData** cfd) {
Status s = VersionEditHandler::ApplyVersionEdit(edit, cfd);
if (s.ok()) {
assert(cfd);
if (*cfd) {
cfds_changed_.insert(*cfd);
}
}
return s;
}
Status ManifestTailer::OnColumnFamilyAdd(VersionEdit& edit,
ColumnFamilyData** cfd) {
if (Mode::kRecovery == mode_) {
return VersionEditHandler::OnColumnFamilyAdd(edit, cfd);
}
assert(Mode::kCatchUp == mode_);
ColumnFamilySet* cfd_set = version_set_->GetColumnFamilySet();
assert(cfd_set);
ColumnFamilyData* tmp_cfd = cfd_set->GetColumnFamily(edit.GetColumnFamily());
assert(cfd);
*cfd = tmp_cfd;
if (!tmp_cfd) {
// For now, ignore new column families created after Recover() succeeds.
return Status::OK();
}
auto builder_iter = builders_.find(edit.GetColumnFamily());
assert(builder_iter != builders_.end());
Version* dummy_version = tmp_cfd->dummy_versions();
assert(dummy_version);
Version* base_version = dummy_version->Next();
assert(base_version);
base_version->Ref();
VersionBuilderUPtr new_builder(
new BaseReferencedVersionBuilder(tmp_cfd, base_version));
builder_iter->second = std::move(new_builder);
#ifndef NDEBUG
auto version_iter = versions_.find(edit.GetColumnFamily());
assert(version_iter == versions_.end());
#endif // !NDEBUG
return Status::OK();
}
void ManifestTailer::CheckIterationResult(const log::Reader& reader,
Status* s) {
VersionEditHandlerPointInTime::CheckIterationResult(reader, s);
assert(s);
if (s->ok()) {
if (Mode::kRecovery == mode_) {
mode_ = Mode::kCatchUp;
} else {
assert(Mode::kCatchUp == mode_);
}
}
}
Status ManifestTailer::VerifyFile(const std::string& fpath,
const FileMetaData& fmeta) {
Status s = VersionEditHandlerPointInTime::VerifyFile(fpath, fmeta);
// TODO: Open file or create hard link to prevent the file from being
// deleted.
return s;
}
void DumpManifestHandler::CheckIterationResult(const log::Reader& reader,
Status* s) {
VersionEditHandler::CheckIterationResult(reader, s);
if (!s->ok()) {
fprintf(stdout, "%s\n", s->ToString().c_str());
return;
}
assert(cf_to_cmp_names_);
for (auto* cfd : *(version_set_->column_family_set_)) {
fprintf(stdout,
"--------------- Column family \"%s\" (ID %" PRIu32
") --------------\n",
cfd->GetName().c_str(), cfd->GetID());
fprintf(stdout, "log number: %" PRIu64 "\n", cfd->GetLogNumber());
auto it = cf_to_cmp_names_->find(cfd->GetID());
if (it != cf_to_cmp_names_->end()) {
fprintf(stdout,
"comparator: <%s>, but the comparator object is not available.\n",
it->second.c_str());
} else {
fprintf(stdout, "comparator: %s\n", cfd->user_comparator()->Name());
}
assert(cfd->current());
// Print out DebugStrings. Can include non-terminating null characters.
fwrite(cfd->current()->DebugString(hex_).data(), sizeof(char),
cfd->current()->DebugString(hex_).size(), stdout);
}
fprintf(stdout,
"next_file_number %" PRIu64 " last_sequence %" PRIu64
" prev_log_number %" PRIu64 " max_column_family %" PRIu32
Fix a race condition in WAL tracking causing DB open failure (#9715) Summary: There is a race condition if WAL tracking in the MANIFEST is enabled in a database that disables 2PC. The race condition is between two background flush threads trying to install flush results to the MANIFEST. Consider an example database with two column families: "default" (cfd0) and "cf1" (cfd1). Initially, both column families have one mutable (active) memtable whose data backed by 6.log. 1. Trigger a manual flush for "cf1", creating a 7.log 2. Insert another key to "default", and trigger flush for "default", creating 8.log 3. BgFlushThread1 finishes writing 9.sst 4. BgFlushThread2 finishes writing 10.sst ``` Time BgFlushThread1 BgFlushThread2 | mutex_.Lock() | precompute min_wal_to_keep as 6 | mutex_.Unlock() | mutex_.Lock() | precompute min_wal_to_keep as 6 | join MANIFEST write queue and mutex_.Unlock() | write to MANIFEST | mutex_.Lock() | cfd1->log_number = 7 | Signal bg_flush_2 and mutex_.Unlock() | wake up and mutex_.Lock() | cfd0->log_number = 8 | FindObsoleteFiles() with job_context->log_number == 7 | mutex_.Unlock() | PurgeObsoleteFiles() deletes 6.log V ``` As shown in the above, BgFlushThread2 thinks that the min wal to keep is 6.log because "cf1" has unflushed data in 6.log (cf1.log_number=6). Similarly, BgThread1 thinks that min wal to keep is also 6.log because "default" has unflushed data (default.log_number=6). No WAL deletion will be written to MANIFEST because 6 is equal to `versions_->wals_.min_wal_number_to_keep`, due to https://github.com/facebook/rocksdb/blob/7.1.fb/db/memtable_list.cc#L513:L514. The bg flush thread that finishes last will perform file purging. `job_context.log_number` will be evaluated as 7, i.e. the min wal that contains unflushed data, causing 6.log to be deleted. However, MANIFEST thinks 6.log should still exist. If you close the db at this point, you won't be able to re-open it if `track_and_verify_wal_in_manifest` is true. We must handle the case of multiple bg flush threads, and it is difficult for one bg flush thread to know the correct min wal number until the other bg flush threads have finished committing to the manifest and updated the `cfd::log_number`. To fix this issue, we rename an existing variable `min_log_number_to_keep_2pc` to `min_log_number_to_keep`, and use it to track WAL file deletion in non-2pc mode as well. This variable is updated only 1) during recovery with mutex held, or 2) in the MANIFEST write thread. `min_log_number_to_keep` means RocksDB will delete WALs below it, although there may be WALs above it which are also obsolete. Formally, we will have [min_wal_to_keep, max_obsolete_wal]. During recovery, we make sure that only WALs above max_obsolete_wal are checked and added back to `alive_log_files_`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9715 Test Plan: ``` make check ``` Also ran stress test below (with asan) to make sure it completes successfully. ``` TEST_TMPDIR=/dev/shm/rocksdb OPT=-g ASAN_OPTIONS=disable_coredump=0 \ CRASH_TEST_EXT_ARGS=--compression_type=zstd SKIP_FORMAT_BUCK_CHECKS=1 \ make J=52 -j52 blackbox_asan_crash_test ``` Reviewed By: ltamasi Differential Revision: D34984412 Pulled By: riversand963 fbshipit-source-id: c7b21a8d84751bb55ea79c9f387103d21b231005
2022-03-24 03:41:31 +01:00
" min_log_number_to_keep %" PRIu64 "\n",
version_set_->current_next_file_number(),
version_set_->LastSequence(), version_set_->prev_log_number(),
version_set_->column_family_set_->GetMaxColumnFamily(),
Fix a race condition in WAL tracking causing DB open failure (#9715) Summary: There is a race condition if WAL tracking in the MANIFEST is enabled in a database that disables 2PC. The race condition is between two background flush threads trying to install flush results to the MANIFEST. Consider an example database with two column families: "default" (cfd0) and "cf1" (cfd1). Initially, both column families have one mutable (active) memtable whose data backed by 6.log. 1. Trigger a manual flush for "cf1", creating a 7.log 2. Insert another key to "default", and trigger flush for "default", creating 8.log 3. BgFlushThread1 finishes writing 9.sst 4. BgFlushThread2 finishes writing 10.sst ``` Time BgFlushThread1 BgFlushThread2 | mutex_.Lock() | precompute min_wal_to_keep as 6 | mutex_.Unlock() | mutex_.Lock() | precompute min_wal_to_keep as 6 | join MANIFEST write queue and mutex_.Unlock() | write to MANIFEST | mutex_.Lock() | cfd1->log_number = 7 | Signal bg_flush_2 and mutex_.Unlock() | wake up and mutex_.Lock() | cfd0->log_number = 8 | FindObsoleteFiles() with job_context->log_number == 7 | mutex_.Unlock() | PurgeObsoleteFiles() deletes 6.log V ``` As shown in the above, BgFlushThread2 thinks that the min wal to keep is 6.log because "cf1" has unflushed data in 6.log (cf1.log_number=6). Similarly, BgThread1 thinks that min wal to keep is also 6.log because "default" has unflushed data (default.log_number=6). No WAL deletion will be written to MANIFEST because 6 is equal to `versions_->wals_.min_wal_number_to_keep`, due to https://github.com/facebook/rocksdb/blob/7.1.fb/db/memtable_list.cc#L513:L514. The bg flush thread that finishes last will perform file purging. `job_context.log_number` will be evaluated as 7, i.e. the min wal that contains unflushed data, causing 6.log to be deleted. However, MANIFEST thinks 6.log should still exist. If you close the db at this point, you won't be able to re-open it if `track_and_verify_wal_in_manifest` is true. We must handle the case of multiple bg flush threads, and it is difficult for one bg flush thread to know the correct min wal number until the other bg flush threads have finished committing to the manifest and updated the `cfd::log_number`. To fix this issue, we rename an existing variable `min_log_number_to_keep_2pc` to `min_log_number_to_keep`, and use it to track WAL file deletion in non-2pc mode as well. This variable is updated only 1) during recovery with mutex held, or 2) in the MANIFEST write thread. `min_log_number_to_keep` means RocksDB will delete WALs below it, although there may be WALs above it which are also obsolete. Formally, we will have [min_wal_to_keep, max_obsolete_wal]. During recovery, we make sure that only WALs above max_obsolete_wal are checked and added back to `alive_log_files_`. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9715 Test Plan: ``` make check ``` Also ran stress test below (with asan) to make sure it completes successfully. ``` TEST_TMPDIR=/dev/shm/rocksdb OPT=-g ASAN_OPTIONS=disable_coredump=0 \ CRASH_TEST_EXT_ARGS=--compression_type=zstd SKIP_FORMAT_BUCK_CHECKS=1 \ make J=52 -j52 blackbox_asan_crash_test ``` Reviewed By: ltamasi Differential Revision: D34984412 Pulled By: riversand963 fbshipit-source-id: c7b21a8d84751bb55ea79c9f387103d21b231005
2022-03-24 03:41:31 +01:00
version_set_->min_log_number_to_keep());
}
} // namespace ROCKSDB_NAMESPACE