dolysis
/
rocksdb
mirror of https://github.com/facebook/rocksdb.git
3
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Snapshots with user-specified timestamps (#9879) 

Summary:
In RocksDB, keys are associated with (internal) sequence numbers which denote when the keys are written
to the database. Sequence numbers in different RocksDB instances are unrelated, thus not comparable.

It is nice if we can associate sequence numbers with their corresponding actual timestamps. One thing we can
do is to support user-defined timestamp, which allows the applications to specify the format of custom timestamps
and encode a timestamp with each key. More details can be found at https://github.com/facebook/rocksdb/wiki/User-defined-Timestamp-%28Experimental%29.

This PR provides a different but complementary approach. We can associate rocksdb snapshots (defined in
https://github.com/facebook/rocksdb/blob/7.2.fb/include/rocksdb/snapshot.h#L20) with **user-specified** timestamps.
Since a snapshot is essentially an object representing a sequence number, this PR establishes a bi-directional mapping between sequence numbers and timestamps.

In the past, snapshots are usually taken by readers. The current super-version is grabbed, and a `rocksdb::Snapshot`
object is created with the last published sequence number of the super-version. You can see that the reader actually
has no good idea of what timestamp to assign to this snapshot, because by the time the `GetSnapshot()` is called,
an arbitrarily long period of time may have already elapsed since the last write, which is when the last published
sequence number is written.

This observation motivates the creation of "timestamped" snapshots on the write path. Currently, this functionality is
exposed only to the layer of `TransactionDB`. Application can tell RocksDB to create a snapshot when a transaction
commits, effectively associating the last sequence number with a timestamp. It is also assumed that application will
ensure any two snapshots with timestamps should satisfy the following:
```
snapshot1.seq < snapshot2.seq iff. snapshot1.ts < snapshot2.ts
```

If the application can guarantee that when a reader takes a timestamped snapshot, there is no active writes going on
in the database, then we also allow the user to use a new API `TransactionDB::CreateTimestampedSnapshot()` to create
a snapshot with associated timestamp.

Code example
```cpp
// Create a timestamped snapshot when committing transaction.
txn->SetCommitTimestamp(100);
txn->SetSnapshotOnNextOperation();
txn->Commit();

// A wrapper API for convenience
Status Transaction::CommitAndTryCreateSnapshot(
    std::shared_ptr<TransactionNotifier> notifier,
    TxnTimestamp ts,
    std::shared_ptr<const Snapshot>* ret);

// Create a timestamped snapshot if caller guarantees no concurrent writes
std::pair<Status, std::shared_ptr<const Snapshot>> snapshot = txn_db->CreateTimestampedSnapshot(100);
```

The snapshots created in this way will be managed by RocksDB with ref-counting and potentially shared with
other readers. We provide the following APIs for readers to retrieve a snapshot given a timestamp.
```cpp
// Return the timestamped snapshot correponding to given timestamp. If ts is
// kMaxTxnTimestamp, then we return the latest timestamped snapshot if present.
// Othersise, we return the snapshot whose timestamp is equal to `ts`. If no
// such snapshot exists, then we return null.
std::shared_ptr<const Snapshot> TransactionDB::GetTimestampedSnapshot(TxnTimestamp ts) const;
// Return the latest timestamped snapshot if present.
std::shared_ptr<const Snapshot> TransactionDB::GetLatestTimestampedSnapshot() const;
```

We also provide two additional APIs for stats collection and reporting purposes.

```cpp
Status TransactionDB::GetAllTimestampedSnapshots(
    std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
// Return timestamped snapshots whose timestamps fall in [ts_lb, ts_ub) and store them in `snapshots`.
Status TransactionDB::GetTimestampedSnapshots(
    TxnTimestamp ts_lb,
    TxnTimestamp ts_ub,
    std::vector<std::shared_ptr<const Snapshot>>& snapshots) const;
```

To prevent the number of timestamped snapshots from growing infinitely, we provide the following API to release
timestamped snapshots whose timestamps are older than or equal to a given threshold.
```cpp
void TransactionDB::ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts);
```

Before shutdown, RocksDB will release all timestamped snapshots.

Comparison with user-defined timestamp and how they can be combined:
User-defined timestamp persists every key with a timestamp, while timestamped snapshots maintain a volatile
mapping between snapshots (sequence numbers) and timestamps.
Different internal keys with the same user key but different timestamps will be treated as different by compaction,
thus a newer version will not hide older versions (with smaller timestamps) unless they are eligible for garbage collection.
In contrast, taking a timestamped snapshot at a certain sequence number and timestamp prevents all the keys visible in
this snapshot from been dropped by compaction. Here, visible means (seq < snapshot and most recent).
The timestamped snapshot supports the semantics of reading at an exact point in time.

Timestamped snapshots can also be used with user-defined timestamp.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/9879

Test Plan:
```
make check
TEST_TMPDIR=/dev/shm make crash_test_with_txn
```

Reviewed By: siying

Differential Revision: D35783919

Pulled By: riversand963

fbshipit-source-id: 586ad905e169189e19d3bfc0cb0177a7239d1bd4
This commit is contained in:
Yanqin Jin 2022-06-10 16:07:03 -07:00 committed by Facebook GitHub Bot
parent f4052d13b7
commit 1777e5f7e9
35 changed files with 1204 additions and 90 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.gitignore vendored
View File

@ -36,8 +36,6 @@ manifest_dump
sst_dump
blob_dump
block_cache_trace_analyzer
db_readonly_with_timestamp_test
db_with_timestamp_basic_test
tools/block_cache_analyzer/*.pyc
column_aware_encoding_exp
util/build_version.cc
@ -53,7 +51,6 @@ rocksdb_dump
rocksdb_undump
db_test2
trace_analyzer
trace_analyzer_test
block_cache_trace_analyzer
io_tracer_parser
.DS_Store

1
CMakeLists.txt
View File

@ -1390,6 +1390,7 @@ if(WITH_TESTS)
utilities/transactions/write_prepared_transaction_test.cc
utilities/transactions/write_unprepared_transaction_test.cc
utilities/transactions/lock/range/range_locking_test.cc
utilities/transactions/timestamped_snapshot_test.cc
utilities/ttl/ttl_test.cc
utilities/util_merge_operators_test.cc
utilities/write_batch_with_index/write_batch_with_index_test.cc

1
HISTORY.md
View File

@ -28,6 +28,7 @@
* Add blob garbage collection parameters `blob_garbage_collection_policy` and `blob_garbage_collection_age_cutoff` to both force-enable and force-disable GC, as well as selectively override age cutoff when using CompactRange.
* Add an extra sanity check in `GetSortedWalFiles()` (also used by `GetLiveFilesStorageInfo()`, `BackupEngine`, and `Checkpoint`) to reduce risk of successfully created backup or checkpoint failing to open because of missing WAL file.
* Add a new column family option `blob_file_starting_level` to enable writing blob files during flushes and compactions starting from the specified LSM tree level.
* Add support for timestamped snapshots (#9879)
### Behavior changes
* DB::Open(), DB::OpenAsSecondary() will fail if a Logger cannot be created (#9984)

3
Makefile
View File

@ -1777,6 +1777,9 @@ write_prepared_transaction_test: $(OBJ_DIR)/utilities/transactions/write_prepare
write_unprepared_transaction_test: $(OBJ_DIR)/utilities/transactions/write_unprepared_transaction_test.o $(TEST_LIBRARY) $(LIBRARY)
$(AM_LINK)
timestamped_snapshot_test: $(OBJ_DIR)/utilities/transactions/timestamped_snapshot_test.o $(TEST_LIBRARY) $(LIBRARY)
$(AM_LINK)
sst_dump: $(OBJ_DIR)/tools/sst_dump.o $(TOOLS_LIBRARY) $(LIBRARY)
$(AM_LINK)

6
TARGETS
View File

@ -5768,6 +5768,12 @@ cpp_unittest_wrapper(name="timer_test",
extra_compiler_flags=[])
cpp_unittest_wrapper(name="timestamped_snapshot_test",
srcs=["utilities/transactions/timestamped_snapshot_test.cc"],
deps=[":rocksdb_test_lib"],
extra_compiler_flags=[])
cpp_unittest_wrapper(name="trace_analyzer_test",
srcs=["tools/trace_analyzer_test.cc"],
deps=[":rocksdb_test_lib"],

219
db/db_impl/db_impl.cc
View File

@ -511,6 +511,19 @@ void DBImpl::CancelAllBackgroundWork(bool wait) {
WaitForBackgroundWork();
}
Status DBImpl::MaybeReleaseTimestampedSnapshotsAndCheck() {
size_t num_snapshots = 0;
ReleaseTimestampedSnapshotsOlderThan(std::numeric_limits<uint64_t>::max(),
&num_snapshots);
// If there is unreleased snapshot, fail the close call
if (num_snapshots > 0) {
return Status::Aborted("Cannot close DB with unreleased snapshot.");
}
return Status::OK();
}
Status DBImpl::CloseHelper() {
// Guarantee that there is no background error recovery in progress before
// continuing with the shutdown
@ -732,11 +745,19 @@ Status DBImpl::CloseImpl() { return CloseHelper(); }
DBImpl::~DBImpl() {
InstrumentedMutexLock closing_lock_guard(&closing_mutex_);
if (!closed_) {
closed_ = true;
closing_status_ = CloseHelper();
closing_status_.PermitUncheckedError();
if (closed_) {
return;
}
closed_ = true;
{
const Status s = MaybeReleaseTimestampedSnapshotsAndCheck();
s.PermitUncheckedError();
}
closing_status_ = CloseImpl();
closing_status_.PermitUncheckedError();
}
void DBImpl::MaybeIgnoreError(Status* s) const {
@ -1797,11 +1818,7 @@ Status DBImpl::GetImpl(const ReadOptions& read_options, const Slice& key,
// data for the snapshot, so the reader would see neither data that was be
// visible to the snapshot before compaction nor the newer data inserted
// afterwards.
if (last_seq_same_as_publish_seq_) {
snapshot = versions_->LastSequence();
} else {
snapshot = versions_->LastPublishedSequence();
}
snapshot = GetLastPublishedSequence();
if (get_impl_options.callback) {
// The unprep_seqs are not published for write unprepared, so it could be
// that max_visible_seq is larger. Seek to the std::max of the two.
@ -2194,11 +2211,7 @@ bool DBImpl::MultiCFSnapshot(
// version because a flush happening in between may compact away data for
// the snapshot, but the snapshot is earlier than the data overwriting it,
// so users may see wrong results.
if (last_seq_same_as_publish_seq_) {
*snapshot = versions_->LastSequence();
} else {
*snapshot = versions_->LastPublishedSequence();
}
*snapshot = GetLastPublishedSequence();
}
} else {
// If we end up with the same issue of memtable geting sealed during 2
@ -2229,11 +2242,7 @@ bool DBImpl::MultiCFSnapshot(
// acquire the lock so we're sure to succeed
mutex_.Lock();
}
if (last_seq_same_as_publish_seq_) {
*snapshot = versions_->LastSequence();
} else {
*snapshot = versions_->LastPublishedSequence();
}
*snapshot = GetLastPublishedSequence();
} else {
*snapshot =
static_cast_with_check<const SnapshotImpl>(read_options.snapshot)
@ -3170,6 +3179,48 @@ const Snapshot* DBImpl::GetSnapshotForWriteConflictBoundary() {
}
#endif // ROCKSDB_LITE
std::pair<Status, std::shared_ptr<const Snapshot>>
DBImpl::CreateTimestampedSnapshot(SequenceNumber snapshot_seq, uint64_t ts) {
assert(ts != std::numeric_limits<uint64_t>::max());
auto ret = CreateTimestampedSnapshotImpl(snapshot_seq, ts, /*lock=*/true);
return ret;
}
std::shared_ptr<const SnapshotImpl> DBImpl::GetTimestampedSnapshot(
uint64_t ts) const {
InstrumentedMutexLock lock_guard(&mutex_);
return timestamped_snapshots_.GetSnapshot(ts);
}
void DBImpl::ReleaseTimestampedSnapshotsOlderThan(uint64_t ts,
size_t* remaining_total_ss) {
autovector<std::shared_ptr<const SnapshotImpl>> snapshots_to_release;
{
InstrumentedMutexLock lock_guard(&mutex_);
timestamped_snapshots_.ReleaseSnapshotsOlderThan(ts, snapshots_to_release);
}
snapshots_to_release.clear();
if (remaining_total_ss) {
InstrumentedMutexLock lock_guard(&mutex_);
*remaining_total_ss = static_cast<size_t>(snapshots_.count());
}
}
Status DBImpl::GetTimestampedSnapshots(
uint64_t ts_lb, uint64_t ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& timestamped_snapshots) const {
if (ts_lb >= ts_ub) {
return Status::InvalidArgument(
"timestamp lower bound must be smaller than upper bound");
}
timestamped_snapshots.clear();
InstrumentedMutexLock lock_guard(&mutex_);
timestamped_snapshots_.GetSnapshots(ts_lb, ts_ub, timestamped_snapshots);
return Status::OK();
}
SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
bool lock) {
int64_t unix_time = 0;
@ -3179,6 +3230,8 @@ SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
if (lock) {
mutex_.Lock();
} else {
mutex_.AssertHeld();
}
// returns null if the underlying memtable does not support snapshot.
if (!is_snapshot_supported_) {
@ -3188,9 +3241,7 @@ SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
delete s;
return nullptr;
}
auto snapshot_seq = last_seq_same_as_publish_seq_
? versions_->LastSequence()
: versions_->LastPublishedSequence();
auto snapshot_seq = GetLastPublishedSequence();
SnapshotImpl* snapshot =
snapshots_.New(s, snapshot_seq, unix_time, is_write_conflict_boundary);
if (lock) {
@ -3199,6 +3250,115 @@ SnapshotImpl* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary,
return snapshot;
}
std::pair<Status, std::shared_ptr<const SnapshotImpl>>
DBImpl::CreateTimestampedSnapshotImpl(SequenceNumber snapshot_seq, uint64_t ts,
bool lock) {
int64_t unix_time = 0;
immutable_db_options_.clock->GetCurrentTime(&unix_time)
.PermitUncheckedError(); // Ignore error
SnapshotImpl* s = new SnapshotImpl;
const bool need_update_seq = (snapshot_seq != kMaxSequenceNumber);
if (lock) {
mutex_.Lock();
} else {
mutex_.AssertHeld();
}
// returns null if the underlying memtable does not support snapshot.
if (!is_snapshot_supported_) {
if (lock) {
mutex_.Unlock();
}
delete s;
return std::make_pair(
Status::NotSupported("Memtable does not support snapshot"), nullptr);
}
// Caller is not write thread, thus didn't provide a valid snapshot_seq.
// Obtain seq from db.
if (!need_update_seq) {
snapshot_seq = GetLastPublishedSequence();
}
std::shared_ptr<const SnapshotImpl> latest =
timestamped_snapshots_.GetSnapshot(std::numeric_limits<uint64_t>::max());
// If there is already a latest timestamped snapshot, then we need to do some
// checks.
if (latest) {
uint64_t latest_snap_ts = latest->GetTimestamp();
SequenceNumber latest_snap_seq = latest->GetSequenceNumber();
assert(latest_snap_seq <= snapshot_seq);
bool needs_create_snap = true;
Status status;
std::shared_ptr<const SnapshotImpl> ret;
if (latest_snap_ts > ts) {
// A snapshot created later cannot have smaller timestamp than a previous
// timestamped snapshot.
needs_create_snap = false;
std::ostringstream oss;
oss << "snapshot exists with larger timestamp " << latest_snap_ts << " > "
<< ts;
status = Status::InvalidArgument(oss.str());
} else if (latest_snap_ts == ts) {
if (latest_snap_seq == snapshot_seq) {
// We are requesting the same sequence number and timestamp, thus can
// safely reuse (share) the current latest timestamped snapshot.
needs_create_snap = false;
ret = latest;
} else if (latest_snap_seq < snapshot_seq) {
// There may have been writes to the database since the latest
// timestamped snapshot, yet we are still requesting the same
// timestamp. In this case, we cannot create the new timestamped
// snapshot.
needs_create_snap = false;
std::ostringstream oss;
oss << "Allocated seq is " << snapshot_seq
<< ", while snapshot exists with smaller seq " << latest_snap_seq
<< " but same timestamp " << ts;
status = Status::InvalidArgument(oss.str());
}
}
if (!needs_create_snap) {
if (lock) {
mutex_.Unlock();
}
delete s;
return std::make_pair(status, ret);
} else {
status.PermitUncheckedError();
}
}
SnapshotImpl* snapshot =
snapshots_.New(s, snapshot_seq, unix_time,
/*is_write_conflict_boundary=*/true, ts);
std::shared_ptr<const SnapshotImpl> ret(
snapshot,
std::bind(&DBImpl::ReleaseSnapshot, this, std::placeholders::_1));
timestamped_snapshots_.AddSnapshot(ret);
// Caller is from write thread, and we need to update database's sequence
// number.
if (need_update_seq) {
assert(versions_);
if (last_seq_same_as_publish_seq_) {
versions_->SetLastSequence(snapshot_seq);
} else {
// TODO: support write-prepared/write-unprepared transactions with two
// write queues.
assert(false);
}
}
if (lock) {
mutex_.Unlock();
}
return std::make_pair(Status::OK(), ret);
}
namespace {
using CfdList = autovector<ColumnFamilyData*, 2>;
bool CfdListContains(const CfdList& list, ColumnFamilyData* cfd) {
@ -3224,11 +3384,7 @@ void DBImpl::ReleaseSnapshot(const Snapshot* s) {
snapshots_.Delete(casted_s);
uint64_t oldest_snapshot;
if (snapshots_.empty()) {
if (last_seq_same_as_publish_seq_) {
oldest_snapshot = versions_->LastSequence();
} else {
oldest_snapshot = versions_->LastPublishedSequence();
}
oldest_snapshot = GetLastPublishedSequence();
} else {
oldest_snapshot = snapshots_.oldest()->number_;
}
@ -4119,13 +4275,14 @@ Status DBImpl::Close() {
if (closed_) {
return closing_status_;
}
{
InstrumentedMutexLock l(&mutex_);
// If there is unreleased snapshot, fail the close call
if (!snapshots_.empty()) {
return Status::Aborted("Cannot close DB with unreleased snapshot.");
const Status s = MaybeReleaseTimestampedSnapshotsAndCheck();
if (!s.ok()) {
return s;
}
}
closing_status_ = CloseImpl();
closed_ = true;
return closing_status_;

35
db/db_impl/db_impl.h
View File

@ -32,6 +32,7 @@
#include "db/log_writer.h"
#include "db/logs_with_prep_tracker.h"
#include "db/memtable_list.h"
#include "db/post_memtable_callback.h"
#include "db/pre_release_callback.h"
#include "db/range_del_aggregator.h"
#include "db/read_callback.h"
@ -345,6 +346,19 @@ class DBImpl : public DB {
virtual const Snapshot* GetSnapshot() override;
virtual void ReleaseSnapshot(const Snapshot* snapshot) override;
// Create a timestamped snapshot. This snapshot can be shared by multiple
// readers. If any of them uses it for write conflict checking, then
// is_write_conflict_boundary is true. For simplicity, set it to true by
// default.
std::pair<Status, std::shared_ptr<const Snapshot>> CreateTimestampedSnapshot(
SequenceNumber snapshot_seq, uint64_t ts);
std::shared_ptr<const SnapshotImpl> GetTimestampedSnapshot(uint64_t ts) const;
void ReleaseTimestampedSnapshotsOlderThan(
uint64_t ts, size_t* remaining_total_ss = nullptr);
Status GetTimestampedSnapshots(uint64_t ts_lb, uint64_t ts_ub,
std::vector<std::shared_ptr<const Snapshot>>&
timestamped_snapshots) const;
using DB::GetProperty;
virtual bool GetProperty(ColumnFamilyHandle* column_family,
const Slice& property, std::string* value) override;
@ -1222,6 +1236,8 @@ class DBImpl : public DB {
static void TEST_ResetDbSessionIdGen();
static std::string GenerateDbSessionId(Env* env);
bool seq_per_batch() const { return seq_per_batch_; }
protected:
const std::string dbname_;
// TODO(peterd): unify with VersionSet::db_id_
@ -1404,7 +1420,8 @@ class DBImpl : public DB {
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);
PreReleaseCallback* pre_release_callback = nullptr,
PostMemTableCallback* post_memtable_callback = nullptr);
Status PipelinedWriteImpl(const WriteOptions& options, WriteBatch* updates,
WriteCallback* callback = nullptr,
@ -2039,10 +2056,24 @@ class DBImpl : public DB {
SnapshotImpl* GetSnapshotImpl(bool is_write_conflict_boundary,
bool lock = true);
// If snapshot_seq != kMaxSequenceNumber, then this function can only be
// called from the write thread that publishes sequence numbers to readers.
// For 1) write-committed, or 2) write-prepared + one-write-queue, this will
// be the write thread performing memtable writes. For write-prepared with
// two write queues, this will be the write thread writing commit marker to
// the WAL.
// If snapshot_seq == kMaxSequenceNumber, this function is called by a caller
// ensuring no writes to the database.
std::pair<Status, std::shared_ptr<const SnapshotImpl>>
CreateTimestampedSnapshotImpl(SequenceNumber snapshot_seq, uint64_t ts,
bool lock = true);
uint64_t GetMaxTotalWalSize() const;
FSDirectory* GetDataDir(ColumnFamilyData* cfd, size_t path_id) const;
Status MaybeReleaseTimestampedSnapshotsAndCheck();
Status CloseHelper();
void WaitForBackgroundWork();
@ -2308,6 +2339,8 @@ class DBImpl : public DB {
SnapshotList snapshots_;
TimestampedSnapshotList timestamped_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

35
db/db_impl/db_impl_write.cc
View File

@ -126,7 +126,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
uint64_t* log_used, uint64_t log_ref,
bool disable_memtable, uint64_t* seq_used,
size_t batch_cnt,
PreReleaseCallback* pre_release_callback) {
PreReleaseCallback* pre_release_callback,
PostMemTableCallback* post_memtable_callback) {
assert(!seq_per_batch_ || batch_cnt != 0);
if (my_batch == nullptr) {
return Status::InvalidArgument("Batch is nullptr!");
@ -185,6 +186,15 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
return Status::NotSupported(
"pipelined_writes is not compatible with unordered_write");
}
if (immutable_db_options_.enable_pipelined_write &&
post_memtable_callback != nullptr) {
return Status::NotSupported(
"pipelined write currently does not honor post_memtable_callback");
}
if (seq_per_batch_ && post_memtable_callback != nullptr) {
return Status::NotSupported(
"seq_per_batch currently does not honor post_memtable_callback");
}
// Otherwise IsLatestPersistentState optimization does not make sense
assert(!WriteBatchInternal::IsLatestPersistentState(my_batch) ||
disable_memtable);
@ -241,7 +251,8 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
PERF_TIMER_GUARD(write_pre_and_post_process_time);
WriteThread::Writer w(write_options, my_batch, callback, log_ref,
disable_memtable, batch_cnt, pre_release_callback);
disable_memtable, batch_cnt, pre_release_callback,
post_memtable_callback);
StopWatch write_sw(immutable_db_options_.clock, stats_, DB_WRITE);
write_thread_.JoinBatchGroup(&w);
@ -268,6 +279,16 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
// we're responsible for exit batch group
// TODO(myabandeh): propagate status to write_group
auto last_sequence = w.write_group->last_sequence;
for (auto* tmp_w : *(w.write_group)) {
assert(tmp_w);
if (tmp_w->post_memtable_callback) {
Status tmp_s =
(*tmp_w->post_memtable_callback)(last_sequence, disable_memtable);
// TODO: propagate the execution status of post_memtable_callback to
// caller.
assert(tmp_s.ok());
}
}
versions_->SetLastSequence(last_sequence);
MemTableInsertStatusCheck(w.status);
write_thread_.ExitAsBatchGroupFollower(&w);
@ -550,6 +571,16 @@ Status DBImpl::WriteImpl(const WriteOptions& write_options,
}
if (should_exit_batch_group) {
if (status.ok()) {
for (auto* tmp_w : write_group) {
assert(tmp_w);
if (tmp_w->post_memtable_callback) {
Status tmp_s =
(*tmp_w->post_memtable_callback)(last_sequence, disable_memtable);
// TODO: propagate the execution status of post_memtable_callback to
// caller.
assert(tmp_s.ok());
}
}
// Note: if we are to resume after non-OK statuses we need to revisit how
// we reacts to non-OK statuses here.
versions_->SetLastSequence(last_sequence);

6
db/db_test.cc
View File

@ -2875,6 +2875,12 @@ class ModelDB : public DB {
assert(false);
return 0;
}
uint64_t GetTimestamp() const override {
// no need to call this
assert(false);
return 0;
}
};
explicit ModelDB(const Options& options) : options_(options) {}

25
db/post_memtable_callback.h Normal file
View File

@ -0,0 +1,25 @@
// Copyright (c) Meta Platforms, Inc. and affiliates. 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).
#pragma once
#include "rocksdb/status.h"
#include "rocksdb/types.h"
namespace ROCKSDB_NAMESPACE {
// Callback invoked after finishing writing to the memtable but before
// publishing the sequence number to readers.
// Note that with write-prepared/write-unprepared transactions with
// two-write-queues, PreReleaseCallback is called before publishing the
// sequence numbers to readers.
class PostMemTableCallback {
public:
virtual ~PostMemTableCallback() {}
virtual Status operator()(SequenceNumber seq, bool disable_memtable) = 0;
};
} // namespace ROCKSDB_NAMESPACE

67
db/snapshot_impl.h
View File

@ -12,6 +12,7 @@
#include "db/dbformat.h"
#include "rocksdb/db.h"
#include "util/autovector.h"
namespace ROCKSDB_NAMESPACE {
@ -31,6 +32,8 @@ class SnapshotImpl : public Snapshot {
int64_t GetUnixTime() const override { return unix_time_; }
uint64_t GetTimestamp() const override { return timestamp_; }
private:
friend class SnapshotList;
@ -42,6 +45,8 @@ class SnapshotImpl : public Snapshot {
int64_t unix_time_;
uint64_t timestamp_;
// Will this snapshot be used by a Transaction to do write-conflict checking?
bool is_write_conflict_boundary_;
};
@ -55,6 +60,7 @@ class SnapshotList {
// Set all the variables to make UBSAN happy.
list_.list_ = nullptr;
list_.unix_time_ = 0;
list_.timestamp_ = 0;
list_.is_write_conflict_boundary_ = false;
count_ = 0;
}
@ -62,14 +68,19 @@ class SnapshotList {
// No copy-construct.
SnapshotList(const SnapshotList&) = delete;
bool empty() const { return list_.next_ == &list_; }
bool empty() const {
assert(list_.next_ != &list_ || 0 == count_);
return list_.next_ == &list_;
}
SnapshotImpl* oldest() const { assert(!empty()); return list_.next_; }
SnapshotImpl* newest() const { assert(!empty()); return list_.prev_; }
SnapshotImpl* New(SnapshotImpl* s, SequenceNumber seq, uint64_t unix_time,
bool is_write_conflict_boundary) {
bool is_write_conflict_boundary,
uint64_t ts = std::numeric_limits<uint64_t>::max()) {
s->number_ = seq;
s->unix_time_ = unix_time;
s->timestamp_ = ts;
s->is_write_conflict_boundary_ = is_write_conflict_boundary;
s->list_ = this;
s->next_ = &list_;
@ -167,4 +178,56 @@ class SnapshotList {
uint64_t count_;
};
// All operations on TimestampedSnapshotList must be protected by db mutex.
class TimestampedSnapshotList {
public:
explicit TimestampedSnapshotList() = default;
std::shared_ptr<const SnapshotImpl> GetSnapshot(uint64_t ts) const {
if (ts == std::numeric_limits<uint64_t>::max() && !snapshots_.empty()) {
auto it = snapshots_.rbegin();
assert(it != snapshots_.rend());
return it->second;
}
auto it = snapshots_.find(ts);
if (it == snapshots_.end()) {
return std::shared_ptr<const SnapshotImpl>();
}
return it->second;
}
void GetSnapshots(
uint64_t ts_lb, uint64_t ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& snapshots) const {
assert(ts_lb < ts_ub);
auto it_low = snapshots_.lower_bound(ts_lb);
auto it_high = snapshots_.lower_bound(ts_ub);
for (auto it = it_low; it != it_high; ++it) {
snapshots.emplace_back(it->second);
}
}
void AddSnapshot(const std::shared_ptr<const SnapshotImpl>& snapshot) {
assert(snapshot);
snapshots_.try_emplace(snapshot->GetTimestamp(), snapshot);
}
// snapshots_to_release: the container to where the timestamped snapshots will
// be moved so that it retains the last reference to the snapshots and the
// snapshots won't be actually released which requires db mutex. The
// snapshots will be released by caller of ReleaseSnapshotsOlderThan().
void ReleaseSnapshotsOlderThan(
uint64_t ts,
autovector<std::shared_ptr<const SnapshotImpl>>& snapshots_to_release) {
auto ub = snapshots_.lower_bound(ts);
for (auto it = snapshots_.begin(); it != ub; ++it) {
snapshots_to_release.emplace_back(it->second);
}
snapshots_.erase(snapshots_.begin(), ub);
}
private:
std::map<uint64_t, std::shared_ptr<const SnapshotImpl>> snapshots_;
};
} // namespace ROCKSDB_NAMESPACE

7
db/write_thread.h
View File

@ -15,6 +15,7 @@
#include <vector>
#include "db/dbformat.h"
#include "db/post_memtable_callback.h"
#include "db/pre_release_callback.h"
#include "db/write_callback.h"
#include "monitoring/instrumented_mutex.h"
@ -122,6 +123,7 @@ class WriteThread {
size_t batch_cnt; // if non-zero, number of sub-batches in the write batch
size_t protection_bytes_per_key;
PreReleaseCallback* pre_release_callback;
PostMemTableCallback* post_memtable_callback;
uint64_t log_used; // log number that this batch was inserted into
uint64_t log_ref; // log number that memtable insert should reference
WriteCallback* callback;
@ -147,6 +149,7 @@ class WriteThread {
batch_cnt(0),
protection_bytes_per_key(0),
pre_release_callback(nullptr),
post_memtable_callback(nullptr),
log_used(0),
log_ref(0),
callback(nullptr),
@ -160,7 +163,8 @@ class WriteThread {
Writer(const WriteOptions& write_options, WriteBatch* _batch,
WriteCallback* _callback, uint64_t _log_ref, bool _disable_memtable,
size_t _batch_cnt = 0,
PreReleaseCallback* _pre_release_callback = nullptr)
PreReleaseCallback* _pre_release_callback = nullptr,
PostMemTableCallback* _post_memtable_callback = nullptr)
: batch(_batch),
sync(write_options.sync),
no_slowdown(write_options.no_slowdown),
@ -170,6 +174,7 @@ class WriteThread {
batch_cnt(_batch_cnt),
protection_bytes_per_key(_batch->GetProtectionBytesPerKey()),
pre_release_callback(_pre_release_callback),
post_memtable_callback(_post_memtable_callback),
log_used(0),
log_ref(_log_ref),
callback(_callback),

28
db_stress_tool/db_stress_common.cc
View File

@ -148,6 +148,34 @@ void DbVerificationThread(void* v) {
}
}
void SnapshotGcThread(void* v) {
assert(FLAGS_create_timestamped_snapshot_one_in > 0);
auto* thread = reinterpret_cast<ThreadState*>(v);
assert(thread);
SharedState* shared = thread->shared;
assert(shared);
StressTest* stress_test = shared->GetStressTest();
assert(stress_test);
while (true) {
{
MutexLock l(shared->GetMutex());
if (shared->ShouldStopBgThread()) {
shared->IncBgThreadsFinished();
if (shared->BgThreadsFinished()) {
shared->GetCondVar()->SignalAll();
}
return;
}
}
uint64_t now = db_stress_env->NowNanos();
constexpr uint64_t time_diff = static_cast<uint64_t>(1000) * 1000 * 1000;
stress_test->ReleaseOldTimestampedSnapshots(now - time_diff);
db_stress_env->SleepForMicroseconds(1000 * 1000);
}
}
void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz) {
if (!FLAGS_verbose) {
return;

4
db_stress_tool/db_stress_common.h
View File

@ -295,6 +295,8 @@ DECLARE_bool(async_io);
DECLARE_string(wal_compression);
DECLARE_bool(verify_sst_unique_id_in_manifest);
DECLARE_int32(create_timestamped_snapshot_one_in);
constexpr long KB = 1024;
constexpr int kRandomValueMaxFactor = 3;
constexpr int kValueMaxLen = 100;
@ -583,6 +585,8 @@ extern void PoolSizeChangeThread(void* v);
extern void DbVerificationThread(void* v);
extern void SnapshotGcThread(void* v);
extern void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz);
extern int64_t GenerateOneKey(ThreadState* thread, uint64_t iteration);

12
db_stress_tool/db_stress_driver.cc
View File

@ -84,6 +84,10 @@ bool RunStressTest(StressTest* stress) {
shared.IncBgThreads();
}
if (FLAGS_create_timestamped_snapshot_one_in > 0) {
shared.IncBgThreads();
}
std::vector<ThreadState*> threads(n);
for (uint32_t i = 0; i < n; i++) {
threads[i] = new ThreadState(i, &shared);
@ -101,6 +105,11 @@ bool RunStressTest(StressTest* stress) {
&continuous_verification_thread);
}
ThreadState snapshots_gc_thread(0, &shared);
if (FLAGS_create_timestamped_snapshot_one_in > 0) {
db_stress_env->StartThread(SnapshotGcThread, &snapshots_gc_thread);
}
// Each thread goes through the following states:
// initializing -> wait for others to init -> read/populate/depopulate
// wait for others to operate -> verify -> done
@ -169,7 +178,8 @@ bool RunStressTest(StressTest* stress) {
stress->PrintStatistics();
if (FLAGS_compaction_thread_pool_adjust_interval > 0 ||
FLAGS_continuous_verification_interval > 0) {
FLAGS_continuous_verification_interval > 0 ||
FLAGS_create_timestamped_snapshot_one_in > 0) {
MutexLock l(shared.GetMutex());
shared.SetShouldStopBgThread();
while (!shared.BgThreadsFinished()) {

4
db_stress_tool/db_stress_gflags.cc
View File

@ -967,4 +967,8 @@ DEFINE_bool(
"DB-open try verifying the SST unique id between MANIFEST and SST "
"properties.");
DEFINE_int32(
create_timestamped_snapshot_one_in, 0,
"On non-zero, create timestamped snapshots upon transaction commits.");
#endif // GFLAGS

26
db_stress_tool/db_stress_test_base.cc
View File

@ -420,6 +420,20 @@ void StressTest::PrintStatistics() {
}
}
void StressTest::ReleaseOldTimestampedSnapshots(uint64_t ts) {
#ifndef ROCKSDB_LITE
if (!txn_db_) {
return;
}
assert(txn_db_);
txn_db_->ReleaseTimestampedSnapshotsOlderThan(ts);
#else
(void)ts;
fprintf(stderr, "timestamped snapshots not supported in LITE mode\n");
exit(1);
#endif // ROCKSDB_LITE
}
// Currently PreloadDb has to be single-threaded.
void StressTest::PreloadDbAndReopenAsReadOnly(int64_t number_of_keys,
SharedState* shared) {
@ -560,13 +574,21 @@ Status StressTest::NewTxn(WriteOptions& write_opts, Transaction** txn) {
return s;
}
Status StressTest::CommitTxn(Transaction* txn) {
Status StressTest::CommitTxn(Transaction* txn, ThreadState* thread) {
if (!FLAGS_use_txn) {
return Status::InvalidArgument("CommitTxn when FLAGS_use_txn is not set");
}
Status s = txn->Prepare();
std::shared_ptr<const Snapshot> timestamped_snapshot;
if (s.ok()) {
s = txn->Commit();
if (thread && FLAGS_create_timestamped_snapshot_one_in &&
thread->rand.OneIn(FLAGS_create_timestamped_snapshot_one_in)) {
uint64_t ts = db_stress_env->NowNanos();
s = txn->CommitAndTryCreateSnapshot(/*notifier=*/nullptr, ts,
&timestamped_snapshot);
} else {
s = txn->Commit();
}
}
delete txn;
return s;

4
db_stress_tool/db_stress_test_base.h
View File

@ -43,6 +43,8 @@ class StressTest {
void PrintStatistics();
void ReleaseOldTimestampedSnapshots(uint64_t ts);
protected:
Status AssertSame(DB* db, ColumnFamilyHandle* cf,
ThreadState::SnapshotState& snap_state);
@ -56,7 +58,7 @@ class StressTest {
#ifndef ROCKSDB_LITE
Status NewTxn(WriteOptions& write_opts, Transaction** txn);
Status CommitTxn(Transaction* txn);
Status CommitTxn(Transaction* txn, ThreadState* thread = nullptr);
Status RollbackTxn(Transaction* txn);
#endif

11
db_stress_tool/db_stress_tool.cc
View File

@ -276,6 +276,17 @@ int db_stress_tool(int argc, char** argv) {
CheckAndSetOptionsForMultiOpsTxnStressTest();
}
if (FLAGS_create_timestamped_snapshot_one_in > 0) {
if (!FLAGS_use_txn) {
fprintf(stderr, "timestamped snapshot supported only in TransactionDB\n");
exit(1);
} else if (FLAGS_txn_write_policy != 0) {
fprintf(stderr,
"timestamped snapshot supported only in write-committed\n");
exit(1);
}
}
#ifndef NDEBUG
KillPoint* kp = KillPoint::GetInstance();
kp->rocksdb_kill_odds = FLAGS_kill_random_test;

8
db_stress_tool/no_batched_ops_stress.cc
View File

@ -666,7 +666,7 @@ class NonBatchedOpsStressTest : public StressTest {
if (s.ok()) {
s = txn->Merge(cfh, key, v);
if (s.ok()) {
s = CommitTxn(txn);
s = CommitTxn(txn, thread);
}
}
#endif
@ -685,7 +685,7 @@ class NonBatchedOpsStressTest : public StressTest {
if (s.ok()) {
s = txn->Put(cfh, key, v);
if (s.ok()) {
s = CommitTxn(txn);
s = CommitTxn(txn, thread);
}
}
#endif
@ -746,7 +746,7 @@ class NonBatchedOpsStressTest : public StressTest {
if (s.ok()) {
s = txn->Delete(cfh, key);
if (s.ok()) {
s = CommitTxn(txn);
s = CommitTxn(txn, thread);
}
}
#endif
@ -783,7 +783,7 @@ class NonBatchedOpsStressTest : public StressTest {
if (s.ok()) {
s = txn->SingleDelete(cfh, key);
if (s.ok()) {
s = CommitTxn(txn);
s = CommitTxn(txn, thread);
}
}
#endif

2
include/rocksdb/snapshot.h
View File

@ -25,6 +25,8 @@ class Snapshot {
// 00:00:00 (UTC).
virtual int64_t GetUnixTime() const = 0;
virtual uint64_t GetTimestamp() const = 0;
protected:
virtual ~Snapshot();
};

34
include/rocksdb/utilities/transaction.h
View File

@ -115,6 +115,8 @@ class TransactionNotifier {
// Implement this method to receive notification when a snapshot is
// requested via SetSnapshotOnNextOperation.
// Do not take exclusive ownership of `newSnapshot` because it is shared with
// the underlying transaction.
virtual void SnapshotCreated(const Snapshot* newSnapshot) = 0;
};
@ -183,6 +185,10 @@ class Transaction {
// txn2->Put("A", ...);
// txn2->Commit();
// txn1->GetForUpdate(opts, "A", ...); // FAIL!
//
// WriteCommittedTxn only: a new snapshot will be taken upon next operation,
// and next operation can be a Commit.
// TODO(yanqin) remove the "write-committed only" limitation.
virtual void SetSnapshotOnNextOperation(
std::shared_ptr<TransactionNotifier> notifier = nullptr) = 0;
@ -193,6 +199,10 @@ class Transaction {
// is called, or the Transaction is deleted.
virtual const Snapshot* GetSnapshot() const = 0;
// Returns the Snapshot created by the last call to SetSnapshot().
// The returned snapshot can outlive the transaction.
virtual std::shared_ptr<const Snapshot> GetTimestampedSnapshot() const = 0;
// Clears the current snapshot (i.e. no snapshot will be 'set')
//
// This removes any snapshot that currently exists or is set to be created
@ -227,6 +237,28 @@ class Transaction {
// transaction before Commit.
virtual Status Commit() = 0;
// In addition to Commit(), also creates a snapshot of the db after all
// writes by this txn are visible to other readers.
// Caller is responsible for ensuring that
// snapshot1.seq < snapshot2.seq iff. snapshot1.ts < snapshot2.ts
// in which snapshot1 and snapshot2 are created by this API.
//
// Currently only supported by WriteCommittedTxn. Calling this method on
// other types of transactions will return non-ok Status resulting from
// Commit() or a `NotSupported` error.
// This method returns OK if and only if the transaction successfully
// commits. It is possible that transaction commits successfully but fails to
// create a timestamped snapshot. Therefore, the caller should check that the
// snapshot is created.
// notifier will be notified upon next snapshot creation. Nullable.
// ret non-null output argument storing a shared_ptr to the newly created
// snapshot.
Status CommitAndTryCreateSnapshot(
std::shared_ptr<TransactionNotifier> notifier =
std::shared_ptr<TransactionNotifier>(),
TxnTimestamp ts = kMaxTxnTimestamp,
std::shared_ptr<const Snapshot>* snapshot = nullptr);
// Discard all batched writes in this transaction.
virtual Status Rollback() = 0;
@ -620,6 +652,8 @@ class Transaction {
return Status::NotSupported("timestamp not supported");
}
virtual TxnTimestamp GetCommitTimestamp() const { return kMaxTxnTimestamp; }
protected:
explicit Transaction(const TransactionDB* /*db*/) {}
Transaction() : log_number_(0), txn_state_(STARTED) {}

36
include/rocksdb/utilities/transaction_db.h
View File

@ -458,6 +458,42 @@ class TransactionDB : public StackableDB {
virtual std::vector<DeadlockPath> GetDeadlockInfoBuffer() = 0;
virtual void SetDeadlockInfoBufferSize(uint32_t target_size) = 0;
// Create a snapshot and assign ts to it. Return the snapshot to caller. The
// snapshot-timestamp mapping is also tracked by the database.
// Caller must ensure there are no active writes when this API is called.
virtual std::pair<Status, std::shared_ptr<const Snapshot>>
CreateTimestampedSnapshot(TxnTimestamp ts) = 0;
// Return the latest timestamped snapshot if present.
std::shared_ptr<const Snapshot> GetLatestTimestampedSnapshot() const {
return GetTimestampedSnapshot(kMaxTxnTimestamp);
}
// Return the snapshot correponding to given timestamp. If ts is
// kMaxTxnTimestamp, then we return the latest timestamped snapshot if
// present. Othersise, we return the snapshot whose timestamp is equal to
// `ts`. If no such snapshot exists, then we return null.
virtual std::shared_ptr<const Snapshot> GetTimestampedSnapshot(
TxnTimestamp ts) const = 0;
// Release timestamped snapshots whose timestamps are less than or equal to
// ts.
virtual void ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts) = 0;
// Get all timestamped snapshots which will be stored in
// timestamped_snapshots.
Status GetAllTimestampedSnapshots(
std::vector<std::shared_ptr<const Snapshot>>& timestamped_snapshots)
const {
return GetTimestampedSnapshots(/*ts_lb=*/0, /*ts_ub=*/kMaxTxnTimestamp,
timestamped_snapshots);
}
// Get all timestamped snapshots whose timestamps fall within [ts_lb, ts_ub).
// timestamped_snapshots will be cleared and contain returned snapshots.
virtual Status GetTimestampedSnapshots(
TxnTimestamp ts_lb, TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& timestamped_snapshots)
const = 0;
protected:
// To Create an TransactionDB, call Open()
// The ownership of db is transferred to the base StackableDB

1
src.mk
View File

@ -594,6 +594,7 @@ TEST_MAIN_SOURCES = \
utilities/transactions/write_prepared_transaction_test.cc \
utilities/transactions/write_unprepared_transaction_test.cc \
utilities/transactions/write_committed_transaction_ts_test.cc \
utilities/transactions/timestamped_snapshot_test.cc \
utilities/ttl/ttl_test.cc \
utilities/util_merge_operators_test.cc \
utilities/write_batch_with_index/write_batch_with_index_test.cc \

6
tools/db_crashtest.py
View File

@ -318,6 +318,7 @@ txn_params = {
"checkpoint_one_in": 0,
# pipeline write is not currnetly compatible with WritePrepared txns
"enable_pipelined_write": 0,
"create_timestamped_snapshot_one_in": random.choice([0, 20]),
}
best_efforts_recovery_params = {
@ -533,6 +534,11 @@ def finalize_and_sanitize(src_params):
if dest_params["secondary_cache_uri"] != "":
# Currently the only cache type compatible with a secondary cache is LRUCache
dest_params["cache_type"] = "lru_cache"
# Remove the following once write-prepared/write-unprepared with/without
# unordered write supports timestamped snapshots
if dest_params.get("create_timestamped_snapshot_one_in", 0) > 0:
dest_params["txn_write_policy"] = 0
dest_params["unordered_write"] = 0
return dest_params

55
utilities/transactions/optimistic_transaction_test.cc
View File

@ -22,8 +22,6 @@
#include "util/crc32c.h"
#include "util/random.h"
using std::string;
namespace ROCKSDB_NAMESPACE {
class OptimisticTransactionTest
@ -31,7 +29,7 @@ class OptimisticTransactionTest
public testing::WithParamInterface<OccValidationPolicy> {
public:
OptimisticTransactionDB* txn_db;
string dbname;
std::string dbname;
Options options;
OptimisticTransactionTest() {
@ -78,7 +76,7 @@ private:
TEST_P(OptimisticTransactionTest, SuccessTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, Slice("foo"), Slice("bar")));
ASSERT_OK(txn_db->Put(write_options, Slice("foo2"), Slice("bar")));
@ -105,7 +103,7 @@ TEST_P(OptimisticTransactionTest, SuccessTest) {
TEST_P(OptimisticTransactionTest, WriteConflictTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "foo", "bar"));
ASSERT_OK(txn_db->Put(write_options, "foo2", "bar"));
@ -138,7 +136,7 @@ TEST_P(OptimisticTransactionTest, WriteConflictTest2) {
WriteOptions write_options;
ReadOptions read_options;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "foo", "bar"));
ASSERT_OK(txn_db->Put(write_options, "foo2", "bar"));
@ -172,7 +170,7 @@ TEST_P(OptimisticTransactionTest, ReadConflictTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "foo", "bar"));
ASSERT_OK(txn_db->Put(write_options, "foo2", "bar"));
@ -211,7 +209,7 @@ TEST_P(OptimisticTransactionTest, TxnOnlyTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
Transaction* txn = txn_db->BeginTransaction(write_options);
ASSERT_NE(txn, nullptr);
@ -226,7 +224,7 @@ TEST_P(OptimisticTransactionTest, TxnOnlyTest) {
TEST_P(OptimisticTransactionTest, FlushTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, Slice("foo"), Slice("bar")));
ASSERT_OK(txn_db->Put(write_options, Slice("foo2"), Slice("bar")));
@ -263,7 +261,7 @@ TEST_P(OptimisticTransactionTest, FlushTest) {
TEST_P(OptimisticTransactionTest, FlushTest2) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, Slice("foo"), Slice("bar")));
ASSERT_OK(txn_db->Put(write_options, Slice("foo2"), Slice("bar")));
@ -324,7 +322,7 @@ TEST_P(OptimisticTransactionTest, CheckKeySkipOldMemtable) {
ReadOptions read_options;
ReadOptions snapshot_read_options;
ReadOptions snapshot_read_options2;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, Slice("foo"), Slice("bar")));
ASSERT_OK(txn_db->Put(write_options, Slice("foo2"), Slice("bar")));
@ -426,7 +424,7 @@ TEST_P(OptimisticTransactionTest, CheckKeySkipOldMemtable) {
TEST_P(OptimisticTransactionTest, NoSnapshotTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "AAA", "bar"));
@ -453,7 +451,7 @@ TEST_P(OptimisticTransactionTest, NoSnapshotTest) {
TEST_P(OptimisticTransactionTest, MultipleSnapshotTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "AAA", "bar"));
ASSERT_OK(txn_db->Put(write_options, "BBB", "bar"));
@ -549,7 +547,7 @@ TEST_P(OptimisticTransactionTest, ColumnFamiliesTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
ColumnFamilyHandle *cfa, *cfb;
ColumnFamilyOptions cf_options;
@ -705,7 +703,7 @@ TEST_P(OptimisticTransactionTest, ColumnFamiliesTest) {
TEST_P(OptimisticTransactionTest, EmptyTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "aaa", "aaa"));
@ -739,7 +737,7 @@ TEST_P(OptimisticTransactionTest, PredicateManyPreceders) {
WriteOptions write_options;
ReadOptions read_options1, read_options2;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
txn_options.set_snapshot = true;
Transaction* txn1 = txn_db->BeginTransaction(write_options, txn_options);
@ -804,7 +802,7 @@ TEST_P(OptimisticTransactionTest, LostUpdate) {
WriteOptions write_options;
ReadOptions read_options, read_options1, read_options2;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
// Test 2 transactions writing to the same key in multiple orders and
// with/without snapshots
@ -892,7 +890,7 @@ TEST_P(OptimisticTransactionTest, LostUpdate) {
TEST_P(OptimisticTransactionTest, UntrackedWrites) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
Status s;
// Verify transaction rollback works for untracked keys.
@ -942,7 +940,7 @@ TEST_P(OptimisticTransactionTest, IteratorTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
// Write some keys to the db
ASSERT_OK(txn_db->Put(write_options, "A", "a"));
@ -1047,7 +1045,7 @@ TEST_P(OptimisticTransactionTest, SavepointTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
Transaction* txn = txn_db->BeginTransaction(write_options);
ASSERT_NE(txn, nullptr);
@ -1169,7 +1167,7 @@ TEST_P(OptimisticTransactionTest, UndoGetForUpdateTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
OptimisticTransactionOptions txn_options;
string value;
std::string value;
ASSERT_OK(txn_db->Put(write_options, "A", ""));
@ -1392,6 +1390,21 @@ TEST_P(OptimisticTransactionTest, SequenceNumberAfterRecoverTest) {
delete transaction;
}
TEST_P(OptimisticTransactionTest, TimestampedSnapshotMissingCommitTs) {
std::unique_ptr<Transaction> txn(txn_db->BeginTransaction(WriteOptions()));
ASSERT_OK(txn->Put("a", "v"));
Status s = txn->CommitAndTryCreateSnapshot();
ASSERT_TRUE(s.IsInvalidArgument());
}
TEST_P(OptimisticTransactionTest, TimestampedSnapshotSetCommitTs) {
std::unique_ptr<Transaction> txn(txn_db->BeginTransaction(WriteOptions()));
ASSERT_OK(txn->Put("a", "v"));
std::shared_ptr<const Snapshot> snapshot;
Status s = txn->CommitAndTryCreateSnapshot(nullptr, /*ts=*/100, &snapshot);
ASSERT_TRUE(s.IsNotSupported());
}
INSTANTIATE_TEST_CASE_P(
InstanceOccGroup, OptimisticTransactionTest,
testing::Values(OccValidationPolicy::kValidateSerial,

34
utilities/transactions/pessimistic_transaction.cc
View File

@ -689,10 +689,21 @@ Status WriteCommittedTxn::CommitWithoutPrepareInternal() {
}
uint64_t seq_used = kMaxSequenceNumber;
auto s =
db_impl_->WriteImpl(write_options_, wb,
/*callback*/ nullptr, /*log_used*/ nullptr,
/*log_ref*/ 0, /*disable_memtable*/ false, &seq_used);
SnapshotCreationCallback snapshot_creation_cb(db_impl_, commit_timestamp_,
snapshot_notifier_, snapshot_);
PostMemTableCallback* post_mem_cb = nullptr;
if (snapshot_needed_) {
if (commit_timestamp_ == kMaxTxnTimestamp) {
return Status::InvalidArgument("Must set transaction commit timestamp");
} else {
post_mem_cb = &snapshot_creation_cb;
}
}
auto s = db_impl_->WriteImpl(write_options_, wb,
/*callback*/ nullptr, /*log_used*/ nullptr,
/*log_ref*/ 0, /*disable_memtable*/ false,
&seq_used, /*batch_cnt=*/0,
/*pre_release_callback=*/nullptr, post_mem_cb);
assert(!s.ok() || seq_used != kMaxSequenceNumber);
if (s.ok()) {
SetId(seq_used);
@ -764,9 +775,22 @@ Status WriteCommittedTxn::CommitInternal() {
assert(s.ok());
uint64_t seq_used = kMaxSequenceNumber;
SnapshotCreationCallback snapshot_creation_cb(db_impl_, commit_timestamp_,
snapshot_notifier_, snapshot_);
PostMemTableCallback* post_mem_cb = nullptr;
if (snapshot_needed_) {
if (commit_timestamp_ == kMaxTxnTimestamp) {
s = Status::InvalidArgument("Must set transaction commit timestamp");
return s;
} else {
post_mem_cb = &snapshot_creation_cb;
}
}
s = db_impl_->WriteImpl(write_options_, working_batch, /*callback*/ nullptr,
/*log_used*/ nullptr, /*log_ref*/ log_number_,
/*disable_memtable*/ false, &seq_used);
/*disable_memtable*/ false, &seq_used,
/*batch_cnt=*/0, /*pre_release_callback=*/nullptr,
post_mem_cb);
assert(!s.ok() || seq_used != kMaxSequenceNumber);
if (s.ok()) {
SetId(seq_used);

1
utilities/transactions/pessimistic_transaction.h
View File

@ -277,6 +277,7 @@ class WriteCommittedTxn : public PessimisticTransaction {
Status SetReadTimestampForValidation(TxnTimestamp ts) override;
Status SetCommitTimestamp(TxnTimestamp ts) override;
TxnTimestamp GetCommitTimestamp() const override { return commit_timestamp_; }
private:
template <typename TValue>

63
utilities/transactions/pessimistic_transaction_db.cc
View File

@ -656,5 +656,68 @@ void PessimisticTransactionDB::UnregisterTransaction(Transaction* txn) {
transactions_.erase(it);
}
std::pair<Status, std::shared_ptr<const Snapshot>>
PessimisticTransactionDB::CreateTimestampedSnapshot(TxnTimestamp ts) {
if (kMaxTxnTimestamp == ts) {
return std::make_pair(Status::InvalidArgument("invalid ts"), nullptr);
}
assert(db_impl_);
return db_impl_->CreateTimestampedSnapshot(kMaxSequenceNumber, ts);
}
std::shared_ptr<const Snapshot>
PessimisticTransactionDB::GetTimestampedSnapshot(TxnTimestamp ts) const {
assert(db_impl_);
return db_impl_->GetTimestampedSnapshot(ts);
}
void PessimisticTransactionDB::ReleaseTimestampedSnapshotsOlderThan(
TxnTimestamp ts) {
assert(db_impl_);
db_impl_->ReleaseTimestampedSnapshotsOlderThan(ts);
}
Status PessimisticTransactionDB::GetTimestampedSnapshots(
TxnTimestamp ts_lb, TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>& timestamped_snapshots) const {
assert(db_impl_);
return db_impl_->GetTimestampedSnapshots(ts_lb, ts_ub, timestamped_snapshots);
}
Status SnapshotCreationCallback::operator()(SequenceNumber seq,
bool disable_memtable) {
assert(db_impl_);
assert(commit_ts_ != kMaxTxnTimestamp);
const bool two_write_queues =
db_impl_->immutable_db_options().two_write_queues;
assert(!two_write_queues || !disable_memtable);
#ifdef NDEBUG
(void)two_write_queues;
(void)disable_memtable;
#endif
const bool seq_per_batch = db_impl_->seq_per_batch();
if (!seq_per_batch) {
assert(db_impl_->GetLastPublishedSequence() <= seq);
} else {
assert(db_impl_->GetLastPublishedSequence() < seq);
}
// Create a snapshot which can also be used for write conflict checking.
auto ret = db_impl_->CreateTimestampedSnapshot(seq, commit_ts_);
snapshot_creation_status_ = ret.first;
snapshot_ = ret.second;
if (snapshot_creation_status_.ok()) {
assert(snapshot_);
} else {
assert(!snapshot_);
}
if (snapshot_ && snapshot_notifier_) {
snapshot_notifier_->SnapshotCreated(snapshot_.get());
}
return Status::OK();
}
} // namespace ROCKSDB_NAMESPACE
#endif // ROCKSDB_LITE

40
utilities/transactions/pessimistic_transaction_db.h
View File

@ -150,6 +150,18 @@ class PessimisticTransactionDB : public TransactionDB {
return lock_manager_->GetLockTrackerFactory();
}
std::pair<Status, std::shared_ptr<const Snapshot>> CreateTimestampedSnapshot(
TxnTimestamp ts) override;
std::shared_ptr<const Snapshot> GetTimestampedSnapshot(
TxnTimestamp ts) const override;
void ReleaseTimestampedSnapshotsOlderThan(TxnTimestamp ts) override;
Status GetTimestampedSnapshots(TxnTimestamp ts_lb, TxnTimestamp ts_ub,
std::vector<std::shared_ptr<const Snapshot>>&
timestamped_snapshots) const override;
protected:
DBImpl* db_impl_;
std::shared_ptr<Logger> info_log_;
@ -255,5 +267,33 @@ inline Status PessimisticTransactionDB::FailIfCfEnablesTs(
return Status::OK();
}
class SnapshotCreationCallback : public PostMemTableCallback {
public:
explicit SnapshotCreationCallback(
DBImpl* dbi, TxnTimestamp commit_ts,
const std::shared_ptr<TransactionNotifier>& notifier,
std::shared_ptr<const Snapshot>& snapshot)
: db_impl_(dbi),
commit_ts_(commit_ts),
snapshot_notifier_(notifier),
snapshot_(snapshot) {
assert(db_impl_);
}
~SnapshotCreationCallback() override {
snapshot_creation_status_.PermitUncheckedError();
}
Status operator()(SequenceNumber seq, bool disable_memtable) override;
private:
DBImpl* const db_impl_;
const TxnTimestamp commit_ts_;
std::shared_ptr<TransactionNotifier> snapshot_notifier_;
std::shared_ptr<const Snapshot>& snapshot_;
Status snapshot_creation_status_;
};
} // namespace ROCKSDB_NAMESPACE
#endif // ROCKSDB_LITE

426
utilities/transactions/timestamped_snapshot_test.cc Normal file
View File

@ -0,0 +1,426 @@
// Copyright (c) Meta Platforms, Inc. and affiliates. 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).
#ifdef ROCKSDB_LITE
#include <cstdio>
int main(int /*argc*/, char** /*argv*/) {
fprintf(stderr, "SKIPPED as Transactions are not supported in LITE mode\n");
return 0;
}
#else // ROCKSDB_LITE
#include <cassert>
#include "util/cast_util.h"
#include "utilities/transactions/transaction_test.h"
namespace ROCKSDB_NAMESPACE {
INSTANTIATE_TEST_CASE_P(
Unsupported, TimestampedSnapshotWithTsSanityCheck,
::testing::Values(
std::make_tuple(false, false, WRITE_PREPARED, kOrderedWrite),
std::make_tuple(false, true, WRITE_PREPARED, kUnorderedWrite),
std::make_tuple(false, false, WRITE_UNPREPARED, kOrderedWrite)));
INSTANTIATE_TEST_CASE_P(WriteCommitted, TransactionTest,
::testing::Combine(::testing::Bool(), ::testing::Bool(),
::testing::Values(WRITE_COMMITTED),
::testing::Values(kOrderedWrite)));
namespace {
// Not thread-safe. Caller needs to provide external synchronization.
class TsCheckingTxnNotifier : public TransactionNotifier {
public:
explicit TsCheckingTxnNotifier() = default;
~TsCheckingTxnNotifier() override {}
void SnapshotCreated(const Snapshot* new_snapshot) override {
assert(new_snapshot);
if (prev_snapshot_seq_ != kMaxSequenceNumber) {
assert(prev_snapshot_seq_ <= new_snapshot->GetSequenceNumber());
}
prev_snapshot_seq_ = new_snapshot->GetSequenceNumber();
if (prev_snapshot_ts_ != kMaxTxnTimestamp) {
assert(prev_snapshot_ts_ <= new_snapshot->GetTimestamp());
}
prev_snapshot_ts_ = new_snapshot->GetTimestamp();
}
TxnTimestamp prev_snapshot_ts() const { return prev_snapshot_ts_; }
private:
SequenceNumber prev_snapshot_seq_ = kMaxSequenceNumber;
TxnTimestamp prev_snapshot_ts_ = kMaxTxnTimestamp;
};
} // anonymous namespace
TEST_P(TimestampedSnapshotWithTsSanityCheck, WithoutCommitTs) {
std::unique_ptr<Transaction> txn(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("a", "v"));
ASSERT_OK(txn->Prepare());
Status s = txn->CommitAndTryCreateSnapshot();
ASSERT_TRUE(s.IsInvalidArgument());
ASSERT_OK(txn->Rollback());
txn.reset(db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("a", "v"));
s = txn->CommitAndTryCreateSnapshot();
ASSERT_TRUE(s.IsInvalidArgument());
}
TEST_P(TimestampedSnapshotWithTsSanityCheck, SetCommitTs) {
std::unique_ptr<Transaction> txn(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("a", "v"));
ASSERT_OK(txn->Prepare());
std::shared_ptr<const Snapshot> snapshot;
Status s = txn->CommitAndTryCreateSnapshot(nullptr, 10, &snapshot);
ASSERT_TRUE(s.IsNotSupported());
ASSERT_OK(txn->Rollback());
txn.reset(db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("a", "v"));
s = txn->CommitAndTryCreateSnapshot(nullptr, 10, &snapshot);
ASSERT_TRUE(s.IsNotSupported());
}
TEST_P(TransactionTest, WithoutCommitTs) {
std::unique_ptr<Transaction> txn(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("a", "v"));
ASSERT_OK(txn->Prepare());
Status s = txn->CommitAndTryCreateSnapshot();
ASSERT_TRUE(s.IsInvalidArgument());
ASSERT_OK(txn->Rollback());
txn.reset(db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("a", "v"));
s = txn->CommitAndTryCreateSnapshot();
ASSERT_TRUE(s.IsInvalidArgument());
}
TEST_P(TransactionTest, CreateSnapshotWhenCommit) {
std::unique_ptr<Transaction> txn(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
assert(txn);
constexpr int batch_size = 10;
for (int i = 0; i < batch_size; ++i) {
ASSERT_OK(db->Put(WriteOptions(), "k" + std::to_string(i), "v0"));
}
const SequenceNumber seq0 = db->GetLatestSequenceNumber();
ASSERT_EQ(static_cast<SequenceNumber>(batch_size), seq0);
txn->SetSnapshot();
{
const Snapshot* const snapshot = txn->GetSnapshot();
assert(snapshot);
ASSERT_EQ(seq0, snapshot->GetSequenceNumber());
}
for (int i = 0; i < batch_size; ++i) {
ASSERT_OK(txn->Put("k" + std::to_string(i), "v1"));
}
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Prepare());
std::shared_ptr<const Snapshot> snapshot;
constexpr TxnTimestamp timestamp = 1;
auto notifier = std::make_shared<TsCheckingTxnNotifier>();
Status s = txn->CommitAndTryCreateSnapshot(notifier, timestamp, &snapshot);
ASSERT_OK(s);
ASSERT_LT(notifier->prev_snapshot_ts(), kMaxTxnTimestamp);
assert(snapshot);
ASSERT_EQ(timestamp, snapshot->GetTimestamp());
ASSERT_EQ(seq0 + batch_size, snapshot->GetSequenceNumber());
const Snapshot* const raw_snapshot_ptr = txn->GetSnapshot();
ASSERT_EQ(raw_snapshot_ptr, snapshot.get());
ASSERT_EQ(snapshot, txn->GetTimestampedSnapshot());
{
std::shared_ptr<const Snapshot> snapshot1 =
db->GetLatestTimestampedSnapshot();
ASSERT_EQ(snapshot, snapshot1);
}
{
std::shared_ptr<const Snapshot> snapshot1 =
db->GetTimestampedSnapshot(timestamp);
ASSERT_EQ(snapshot, snapshot1);
}
{
std::vector<std::shared_ptr<const Snapshot> > snapshots;
s = db->GetAllTimestampedSnapshots(snapshots);
ASSERT_OK(s);
ASSERT_EQ(std::vector<std::shared_ptr<const Snapshot> >{snapshot},
snapshots);
}
}
TEST_P(TransactionTest, CreateSnapshot) {
// First create a non-timestamped snapshot
ManagedSnapshot snapshot_guard(db);
for (int i = 0; i < 10; ++i) {
ASSERT_OK(db->Put(WriteOptions(), "k" + std::to_string(i),
"v0_" + std::to_string(i)));
}
{
auto ret = db->CreateTimestampedSnapshot(kMaxTxnTimestamp);
ASSERT_TRUE(ret.first.IsInvalidArgument());
auto snapshot = ret.second;
ASSERT_EQ(nullptr, snapshot.get());
}
constexpr TxnTimestamp timestamp = 100;
Status s;
std::shared_ptr<const Snapshot> ts_snap0;
std::tie(s, ts_snap0) = db->CreateTimestampedSnapshot(timestamp);
ASSERT_OK(s);
assert(ts_snap0);
ASSERT_EQ(timestamp, ts_snap0->GetTimestamp());
for (int i = 0; i < 10; ++i) {
ASSERT_OK(db->Delete(WriteOptions(), "k" + std::to_string(i)));
}
{
ReadOptions read_opts;
read_opts.snapshot = ts_snap0.get();
for (int i = 0; i < 10; ++i) {
std::string value;
s = db->Get(read_opts, "k" + std::to_string(i), &value);
ASSERT_OK(s);
ASSERT_EQ("v0_" + std::to_string(i), value);
}
}
{
std::shared_ptr<const Snapshot> snapshot =
db->GetLatestTimestampedSnapshot();
ASSERT_EQ(ts_snap0, snapshot);
}
{
std::shared_ptr<const Snapshot> snapshot =
db->GetTimestampedSnapshot(timestamp);
ASSERT_OK(s);
ASSERT_EQ(ts_snap0, snapshot);
}
{
std::vector<std::shared_ptr<const Snapshot> > snapshots;
s = db->GetAllTimestampedSnapshots(snapshots);
ASSERT_OK(s);
ASSERT_EQ(std::vector<std::shared_ptr<const Snapshot> >{ts_snap0},
snapshots);
}
}
TEST_P(TransactionTest, SequenceAndTsOrder) {
Status s;
std::shared_ptr<const Snapshot> snapshot;
std::tie(s, snapshot) = db->CreateTimestampedSnapshot(100);
ASSERT_OK(s);
assert(snapshot);
{
// Cannot request smaller timestamp for the new timestamped snapshot.
std::shared_ptr<const Snapshot> tmp_snapshot;
std::tie(s, tmp_snapshot) = db->CreateTimestampedSnapshot(50);
ASSERT_TRUE(s.IsInvalidArgument());
ASSERT_EQ(nullptr, tmp_snapshot.get());
}
// If requesting a new timestamped snapshot with the same timestamp and
// sequence number, we avoid creating new snapshot object but reuse
// exisisting one.
std::shared_ptr<const Snapshot> snapshot1;
std::tie(s, snapshot1) = db->CreateTimestampedSnapshot(100);
ASSERT_OK(s);
ASSERT_EQ(snapshot.get(), snapshot1.get());
// If there is no write, but we request a larger timestamp, we still create
// a new snapshot object.
std::shared_ptr<const Snapshot> snapshot2;
std::tie(s, snapshot2) = db->CreateTimestampedSnapshot(200);
ASSERT_OK(s);
assert(snapshot2);
ASSERT_NE(snapshot.get(), snapshot2.get());
ASSERT_EQ(snapshot2->GetSequenceNumber(), snapshot->GetSequenceNumber());
ASSERT_EQ(200, snapshot2->GetTimestamp());
// Increase sequence number.
ASSERT_OK(db->Put(WriteOptions(), "foo", "v0"));
{
// We are requesting the same timestamp for a larger sequence number, thus
// we cannot create timestamped snapshot.
std::shared_ptr<const Snapshot> tmp_snapshot;
std::tie(s, tmp_snapshot) = db->CreateTimestampedSnapshot(200);
ASSERT_TRUE(s.IsInvalidArgument());
ASSERT_EQ(nullptr, tmp_snapshot.get());
}
{
std::unique_ptr<Transaction> txn1(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
ASSERT_OK(txn1->Put("bar", "v0"));
std::shared_ptr<const Snapshot> ss;
ASSERT_OK(txn1->CommitAndTryCreateSnapshot(nullptr, 200, &ss));
// Cannot create snapshot because requested timestamp is the same as the
// latest timestamped snapshot while sequence number is strictly higher.
ASSERT_EQ(nullptr, ss);
}
{
std::unique_ptr<Transaction> txn2(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
ASSERT_OK(txn2->Put("bar", "v0"));
std::shared_ptr<const Snapshot> ss;
// Application should never do this. This is just to demonstrate error
// handling.
ASSERT_OK(txn2->CommitAndTryCreateSnapshot(nullptr, 100, &ss));
// Cannot create snapshot because requested timestamp is smaller than
// latest timestamped snapshot.
ASSERT_EQ(nullptr, ss);
}
}
TEST_P(TransactionTest, CloseDbWithSnapshots) {
std::unique_ptr<Transaction> txn(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
ASSERT_OK(txn->SetName("txn0"));
ASSERT_OK(txn->Put("foo", "v"));
ASSERT_OK(txn->Prepare());
std::shared_ptr<const Snapshot> snapshot;
constexpr TxnTimestamp timestamp = 121;
auto notifier = std::make_shared<TsCheckingTxnNotifier>();
ASSERT_OK(txn->CommitAndTryCreateSnapshot(notifier, timestamp, &snapshot));
assert(snapshot);
ASSERT_LT(notifier->prev_snapshot_ts(), kMaxTxnTimestamp);
ASSERT_EQ(timestamp, snapshot->GetTimestamp());
ASSERT_TRUE(db->Close().IsAborted());
}
TEST_P(TransactionTest, MultipleTimestampedSnapshots) {
auto* dbimpl = static_cast_with_check<DBImpl>(db->GetRootDB());
assert(dbimpl);
const bool seq_per_batch = dbimpl->seq_per_batch();
// TODO: remove the following assert(!seq_per_batch) once timestamped snapshot
// is supported in write-prepared/write-unprepared transactions.
assert(!seq_per_batch);
constexpr size_t txn_size = 10;
constexpr TxnTimestamp ts_delta = 10;
constexpr size_t num_txns = 100;
std::vector<std::shared_ptr<const Snapshot> > snapshots(num_txns);
constexpr TxnTimestamp start_ts = 10000;
auto notifier = std::make_shared<TsCheckingTxnNotifier>();
for (size_t i = 0; i < num_txns; ++i) {
std::unique_ptr<Transaction> txn(
db->BeginTransaction(WriteOptions(), TransactionOptions()));
ASSERT_OK(txn->SetName("txn" + std::to_string(i)));
for (size_t j = 0; j < txn_size; ++j) {
ASSERT_OK(txn->Put("k" + std::to_string(j),
"v" + std::to_string(j) + "_" + std::to_string(i)));
}
if (0 == (i % 2)) {
ASSERT_OK(txn->Prepare());
}
ASSERT_OK(txn->CommitAndTryCreateSnapshot(notifier, start_ts + i * ts_delta,
&snapshots[i]));
assert(snapshots[i]);
ASSERT_LT(notifier->prev_snapshot_ts(), kMaxTxnTimestamp);
ASSERT_EQ(start_ts + i * ts_delta, snapshots[i]->GetTimestamp());
}
{
auto snapshot = db->GetTimestampedSnapshot(start_ts + 1);
ASSERT_EQ(nullptr, snapshot);
}
constexpr TxnTimestamp max_ts = start_ts + num_txns * ts_delta;
for (size_t i = 0; i < num_txns; ++i) {
auto snapshot = db->GetTimestampedSnapshot(start_ts + i * ts_delta);
ASSERT_EQ(snapshots[i], snapshot);
std::vector<std::shared_ptr<const Snapshot> > tmp_snapshots;
Status s = db->GetTimestampedSnapshots(max_ts, start_ts + i * ts_delta,
tmp_snapshots);
ASSERT_TRUE(s.IsInvalidArgument());
ASSERT_TRUE(tmp_snapshots.empty());
for (size_t j = i; j < num_txns; ++j) {
std::vector<std::shared_ptr<const Snapshot> > expected_snapshots(
snapshots.begin() + i, snapshots.begin() + j);
tmp_snapshots.clear();
s = db->GetTimestampedSnapshots(start_ts + i * ts_delta,
start_ts + j * ts_delta, tmp_snapshots);
if (i < j) {
ASSERT_OK(s);
} else {
ASSERT_TRUE(s.IsInvalidArgument());
}
ASSERT_EQ(expected_snapshots, tmp_snapshots);
}
}
{
std::vector<std::shared_ptr<const Snapshot> > tmp_snapshots;
const Status s = db->GetAllTimestampedSnapshots(tmp_snapshots);
ASSERT_OK(s);
ASSERT_EQ(snapshots, tmp_snapshots);
const std::shared_ptr<const Snapshot> latest_snapshot =
db->GetLatestTimestampedSnapshot();
ASSERT_EQ(snapshots.back(), latest_snapshot);
}
for (size_t i = 0; i <= num_txns; ++i) {
std::vector<std::shared_ptr<const Snapshot> > snapshots1(
snapshots.begin() + i, snapshots.end());
if (i > 0) {
auto snapshot1 =
db->GetTimestampedSnapshot(start_ts + (i - 1) * ts_delta);
assert(snapshot1);
ASSERT_EQ(start_ts + (i - 1) * ts_delta, snapshot1->GetTimestamp());
}
db->ReleaseTimestampedSnapshotsOlderThan(start_ts + i * ts_delta);
if (i > 0) {
auto snapshot1 =
db->GetTimestampedSnapshot(start_ts + (i - 1) * ts_delta);
ASSERT_EQ(nullptr, snapshot1);
}
std::vector<std::shared_ptr<const Snapshot> > tmp_snapshots;
const Status s = db->GetAllTimestampedSnapshots(tmp_snapshots);
ASSERT_OK(s);
ASSERT_EQ(snapshots1, tmp_snapshots);
}
// Even after released by db, the applications still hold reference to shared
// snapshots.
for (size_t i = 0; i < num_txns; ++i) {
assert(snapshots[i]);
ASSERT_EQ(start_ts + i * ts_delta, snapshots[i]->GetTimestamp());
}
snapshots.clear();
ASSERT_OK(db->Close());
delete db;
db = nullptr;
}
} // namespace ROCKSDB_NAMESPACE
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}
#endif // !ROCKSDB_LITE

37
utilities/transactions/transaction_base.cc
View File

@ -21,6 +21,43 @@
namespace ROCKSDB_NAMESPACE {
Status Transaction::CommitAndTryCreateSnapshot(
std::shared_ptr<TransactionNotifier> notifier, TxnTimestamp ts,
std::shared_ptr<const Snapshot>* snapshot) {
if (snapshot) {
snapshot->reset();
}
TxnTimestamp commit_ts = GetCommitTimestamp();
if (commit_ts == kMaxTxnTimestamp) {
if (ts == kMaxTxnTimestamp) {
return Status::InvalidArgument("Commit timestamp unset");
} else {
const Status s = SetCommitTimestamp(ts);
if (!s.ok()) {
return s;
}
}
} else if (ts != kMaxTxnTimestamp) {
if (ts != commit_ts) {
// For now we treat this as error.
return Status::InvalidArgument("Different commit ts specified");
}
}
SetSnapshotOnNextOperation(notifier);
Status s = Commit();
if (!s.ok()) {
return s;
}
assert(s.ok());
// If we reach here, we must return ok status for this function.
std::shared_ptr<const Snapshot> new_snapshot = GetTimestampedSnapshot();
if (snapshot) {
*snapshot = new_snapshot;
}
return Status::OK();
}
TransactionBaseImpl::TransactionBaseImpl(
DB* db, const WriteOptions& write_options,
const LockTrackerFactory& lock_tracker_factory)

6
utilities/transactions/transaction_base.h
View File

@ -206,6 +206,10 @@ class TransactionBaseImpl : public Transaction {
return snapshot_.get();
}
std::shared_ptr<const Snapshot> GetTimestampedSnapshot() const override {
return snapshot_;
}
virtual void SetSnapshot() override;
void SetSnapshotOnNextOperation(
std::shared_ptr<TransactionNotifier> notifier = nullptr) override;
@ -346,7 +350,9 @@ class TransactionBaseImpl : public Transaction {
save_points_;
private:
friend class WriteCommittedTxn;
friend class WritePreparedTxn;
// Extra data to be persisted with the commit. Note this is only used when
// prepare phase is not skipped.
WriteBatch commit_time_batch_;

30
utilities/transactions/transaction_test.cc
View File

@ -31,8 +31,6 @@
#include "utilities/merge_operators/string_append/stringappend.h"
#include "utilities/transactions/pessimistic_transaction_db.h"
using std::string;
namespace ROCKSDB_NAMESPACE {
INSTANTIATE_TEST_CASE_P(
@ -333,7 +331,7 @@ TEST_P(TransactionTest, WaitingTxn) {
WriteOptions write_options;
ReadOptions read_options;
TransactionOptions txn_options;
string value;
std::string value;
Status s;
txn_options.lock_timeout = 1;
@ -994,7 +992,7 @@ TEST_P(TransactionTest, SimpleTwoPhaseTransactionTest) {
TransactionOptions txn_options;
txn_options.use_only_the_last_commit_time_batch_for_recovery = cwb4recovery;
string value;
std::string value;
Status s;
DBImpl* db_impl = static_cast_with_check<DBImpl>(db->GetRootDB());
@ -2165,7 +2163,7 @@ TEST_P(TransactionTest, WriteOptionsTest) {
TEST_P(TransactionTest, WriteConflictTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
Status s;
ASSERT_OK(db->Put(write_options, "foo", "A"));
@ -2370,7 +2368,7 @@ TEST_P(TransactionTest, FlushTest2) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
TransactionOptions txn_options;
string value;
std::string value;
DBImpl* db_impl = static_cast_with_check<DBImpl>(db->GetRootDB());
@ -2647,7 +2645,7 @@ TEST_P(TransactionTest, ColumnFamiliesTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
TransactionOptions txn_options;
string value;
std::string value;
Status s;
ColumnFamilyHandle *cfa, *cfb;
@ -2814,7 +2812,7 @@ TEST_P(TransactionTest, MultiGetBatchedTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
TransactionOptions txn_options;
string value;
std::string value;
Status s;
ColumnFamilyHandle* cf;
@ -2904,7 +2902,7 @@ TEST_P(TransactionTest, MultiGetBatchedTest) {
TEST_P(TransactionTest, MultiGetLargeBatchedTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
string value;
std::string value;
Status s;
ColumnFamilyHandle* cf;
@ -3037,7 +3035,7 @@ TEST_P(TransactionTest, MultiGetSnapshot) {
TEST_P(TransactionTest, ColumnFamiliesTest2) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
string value;
std::string value;
Status s;
ColumnFamilyHandle *one, *two;
@ -3113,7 +3111,7 @@ TEST_P(TransactionTest, ColumnFamiliesTest2) {
TEST_P(TransactionTest, EmptyTest) {
WriteOptions write_options;
ReadOptions read_options;
string value;
std::string value;
Status s;
s = db->Put(write_options, "aaa", "aaa");
@ -3156,7 +3154,7 @@ TEST_P(TransactionTest, PredicateManyPreceders) {
WriteOptions write_options;
ReadOptions read_options1, read_options2;
TransactionOptions txn_options;
string value;
std::string value;
Status s;
txn_options.set_snapshot = true;
@ -3400,7 +3398,7 @@ TEST_P(TransactionTest, ExpiredTransaction) {
WriteOptions write_options;
ReadOptions read_options;
TransactionOptions txn_options;
string value;
std::string value;
Status s;
// Set txn expiration timeout to 0 microseconds (expires instantly)
@ -3589,7 +3587,7 @@ TEST_P(TransactionTest, LockLimitTest) {
WriteOptions write_options;
ReadOptions read_options, snapshot_read_options;
TransactionOptions txn_options;
string value;
std::string value;
Status s;
delete db;
@ -5414,7 +5412,7 @@ TEST_P(TransactionStressTest, ExpiredTransactionDataRace1) {
ASSERT_OK(s);
ReadOptions read_options;
string value;
std::string value;
s = db->Get(read_options, "X", &value);
ASSERT_OK(s);
ASSERT_EQ("1", value);
@ -6021,7 +6019,7 @@ TEST_P(TransactionTest, DuplicateKeys) {
ASSERT_OK(batch.Merge(cf_handle, Slice("key"), Slice("4")));
ASSERT_OK(db->Write(write_options, &batch));
ReadOptions read_options;
string value;
std::string value;
ASSERT_OK(db->Get(read_options, cf_handle, "key", &value));
ASSERT_EQ(value, "value,1,2,3,4");
delete cf_handle;

18
utilities/transactions/transaction_test.h
View File

@ -562,4 +562,22 @@ class WriteCommittedTxnWithTsTest
std::vector<ColumnFamilyHandle*> handles_{};
};
class TimestampedSnapshotWithTsSanityCheck
: public TransactionTestBase,
public ::testing::WithParamInterface<
std::tuple<bool, bool, TxnDBWritePolicy, WriteOrdering>> {
public:
explicit TimestampedSnapshotWithTsSanityCheck()
: TransactionTestBase(std::get<0>(GetParam()), std::get<1>(GetParam()),
std::get<2>(GetParam()), std::get<3>(GetParam())) {}
~TimestampedSnapshotWithTsSanityCheck() override {
for (auto* h : handles_) {
delete h;
}
}
protected:
std::vector<ColumnFamilyHandle*> handles_{};
};
} // namespace ROCKSDB_NAMESPACE