rocksdb/db/db_log_iter_test.cc

297 lines
9.8 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
// This source code is also licensed under the GPLv2 license found in the
// COPYING file in the root directory of this source tree.
//
// 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.
// Introduction of SyncPoint effectively disabled building and running this test
// in Release build.
// which is a pity, it is a good test
#if !defined(ROCKSDB_LITE)
#include "db/db_test_util.h"
#include "port/stack_trace.h"
namespace rocksdb {
class DBTestXactLogIterator : public DBTestBase {
public:
DBTestXactLogIterator() : DBTestBase("/db_log_iter_test") {}
std::unique_ptr<TransactionLogIterator> OpenTransactionLogIter(
const SequenceNumber seq) {
unique_ptr<TransactionLogIterator> iter;
Status status = dbfull()->GetUpdatesSince(seq, &iter);
EXPECT_OK(status);
EXPECT_TRUE(iter->Valid());
return iter;
}
};
namespace {
SequenceNumber ReadRecords(
std::unique_ptr<TransactionLogIterator>& iter,
int& count) {
count = 0;
SequenceNumber lastSequence = 0;
BatchResult res;
while (iter->Valid()) {
res = iter->GetBatch();
EXPECT_TRUE(res.sequence > lastSequence);
++count;
lastSequence = res.sequence;
EXPECT_OK(iter->status());
iter->Next();
}
return res.sequence;
}
void ExpectRecords(
const int expected_no_records,
std::unique_ptr<TransactionLogIterator>& iter) {
int num_records;
ReadRecords(iter, num_records);
ASSERT_EQ(num_records, expected_no_records);
}
} // namespace
TEST_F(DBTestXactLogIterator, TransactionLogIterator) {
do {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({"pikachu"}, options);
Put(0, "key1", DummyString(1024));
Put(1, "key2", DummyString(1024));
Put(1, "key2", DummyString(1024));
ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 3U);
{
auto iter = OpenTransactionLogIter(0);
ExpectRecords(3, iter);
}
ReopenWithColumnFamilies({"default", "pikachu"}, options);
env_->SleepForMicroseconds(2 * 1000 * 1000);
{
Put(0, "key4", DummyString(1024));
Put(1, "key5", DummyString(1024));
Put(0, "key6", DummyString(1024));
}
{
auto iter = OpenTransactionLogIter(0);
ExpectRecords(6, iter);
}
} while (ChangeCompactOptions());
}
#ifndef NDEBUG // sync point is not included with DNDEBUG build
TEST_F(DBTestXactLogIterator, TransactionLogIteratorRace) {
static const int LOG_ITERATOR_RACE_TEST_COUNT = 2;
static const char* sync_points[LOG_ITERATOR_RACE_TEST_COUNT][4] = {
{"WalManager::GetSortedWalFiles:1", "WalManager::PurgeObsoleteFiles:1",
"WalManager::PurgeObsoleteFiles:2", "WalManager::GetSortedWalFiles:2"},
{"WalManager::GetSortedWalsOfType:1",
"WalManager::PurgeObsoleteFiles:1",
"WalManager::PurgeObsoleteFiles:2",
"WalManager::GetSortedWalsOfType:2"}};
for (int test = 0; test < LOG_ITERATOR_RACE_TEST_COUNT; ++test) {
// Setup sync point dependency to reproduce the race condition of
// a log file moved to archived dir, in the middle of GetSortedWalFiles
rocksdb::SyncPoint::GetInstance()->LoadDependency(
{ { sync_points[test][0], sync_points[test][1] },
{ sync_points[test][2], sync_points[test][3] },
});
do {
rocksdb::SyncPoint::GetInstance()->ClearTrace();
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
Put("key1", DummyString(1024));
dbfull()->Flush(FlushOptions());
Put("key2", DummyString(1024));
dbfull()->Flush(FlushOptions());
Put("key3", DummyString(1024));
dbfull()->Flush(FlushOptions());
Put("key4", DummyString(1024));
ASSERT_EQ(dbfull()->GetLatestSequenceNumber(), 4U);
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 21:06:43 +00:00
dbfull()->FlushWAL(false);
{
auto iter = OpenTransactionLogIter(0);
ExpectRecords(4, iter);
}
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
// trigger async flush, and log move. Well, log move will
// wait until the GetSortedWalFiles:1 to reproduce the race
// condition
FlushOptions flush_options;
flush_options.wait = false;
dbfull()->Flush(flush_options);
// "key5" would be written in a new memtable and log
Put("key5", DummyString(1024));
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 21:06:43 +00:00
dbfull()->FlushWAL(false);
{
// this iter would miss "key4" if not fixed
auto iter = OpenTransactionLogIter(0);
ExpectRecords(5, iter);
}
} while (ChangeCompactOptions());
}
}
#endif
TEST_F(DBTestXactLogIterator, TransactionLogIteratorStallAtLastRecord) {
do {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
Put("key1", DummyString(1024));
auto iter = OpenTransactionLogIter(0);
ASSERT_OK(iter->status());
ASSERT_TRUE(iter->Valid());
iter->Next();
ASSERT_TRUE(!iter->Valid());
ASSERT_OK(iter->status());
Put("key2", DummyString(1024));
iter->Next();
ASSERT_OK(iter->status());
ASSERT_TRUE(iter->Valid());
} while (ChangeCompactOptions());
}
TEST_F(DBTestXactLogIterator, TransactionLogIteratorCheckAfterRestart) {
do {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
Put("key1", DummyString(1024));
Put("key2", DummyString(1023));
dbfull()->Flush(FlushOptions());
Reopen(options);
auto iter = OpenTransactionLogIter(0);
ExpectRecords(2, iter);
} while (ChangeCompactOptions());
}
TEST_F(DBTestXactLogIterator, TransactionLogIteratorCorruptedLog) {
do {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
for (int i = 0; i < 1024; i++) {
Put("key"+ToString(i), DummyString(10));
}
dbfull()->Flush(FlushOptions());
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 21:06:43 +00:00
dbfull()->FlushWAL(false);
// Corrupt this log to create a gap
rocksdb::VectorLogPtr wal_files;
ASSERT_OK(dbfull()->GetSortedWalFiles(wal_files));
const auto logfile_path = dbname_ + "/" + wal_files.front()->PathName();
if (mem_env_) {
mem_env_->Truncate(logfile_path, wal_files.front()->SizeFileBytes() / 2);
} else {
ASSERT_EQ(0, truncate(logfile_path.c_str(),
wal_files.front()->SizeFileBytes() / 2));
}
// Insert a new entry to a new log file
Put("key1025", DummyString(10));
Optimize for serial commits in 2PC Summary: Throughput: 46k tps in our sysbench settings (filling the details later) The idea is to have the simplest change that gives us a reasonable boost in 2PC throughput. Major design changes: 1. The WAL file internal buffer is not flushed after each write. Instead it is flushed before critical operations (WAL copy via fs) or when FlushWAL is called by MySQL. Flushing the WAL buffer is also protected via mutex_. 2. Use two sequence numbers: last seq, and last seq for write. Last seq is the last visible sequence number for reads. Last seq for write is the next sequence number that should be used to write to WAL/memtable. This allows to have a memtable write be in parallel to WAL writes. 3. BatchGroup is not used for writes. This means that we can have parallel writers which changes a major assumption in the code base. To accommodate for that i) allow only 1 WriteImpl that intends to write to memtable via mem_mutex_--which is fine since in 2PC almost all of the memtable writes come via group commit phase which is serial anyway, ii) make all the parts in the code base that assumed to be the only writer (via EnterUnbatched) to also acquire mem_mutex_, iii) stat updates are protected via a stat_mutex_. Note: the first commit has the approach figured out but is not clean. Submitting the PR anyway to get the early feedback on the approach. If we are ok with the approach I will go ahead with this updates: 0) Rebase with Yi's pipelining changes 1) Currently batching is disabled by default to make sure that it will be consistent with all unit tests. Will make this optional via a config. 2) A couple of unit tests are disabled. They need to be updated with the serial commit of 2PC taken into account. 3) Replacing BatchGroup with mem_mutex_ got a bit ugly as it requires releasing mutex_ beforehand (the same way EnterUnbatched does). This needs to be cleaned up. Closes https://github.com/facebook/rocksdb/pull/2345 Differential Revision: D5210732 Pulled By: maysamyabandeh fbshipit-source-id: 78653bd95a35cd1e831e555e0e57bdfd695355a4
2017-06-24 21:06:43 +00:00
dbfull()->FlushWAL(false);
// Try to read from the beginning. Should stop before the gap and read less
// than 1025 entries
auto iter = OpenTransactionLogIter(0);
int count;
SequenceNumber last_sequence_read = ReadRecords(iter, count);
ASSERT_LT(last_sequence_read, 1025U);
// Try to read past the gap, should be able to seek to key1025
auto iter2 = OpenTransactionLogIter(last_sequence_read + 1);
ExpectRecords(1, iter2);
} while (ChangeCompactOptions());
}
TEST_F(DBTestXactLogIterator, TransactionLogIteratorBatchOperations) {
do {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({"pikachu"}, options);
WriteBatch batch;
batch.Put(handles_[1], "key1", DummyString(1024));
batch.Put(handles_[0], "key2", DummyString(1024));
batch.Put(handles_[1], "key3", DummyString(1024));
batch.Delete(handles_[0], "key2");
dbfull()->Write(WriteOptions(), &batch);
Flush(1);
Flush(0);
ReopenWithColumnFamilies({"default", "pikachu"}, options);
Put(1, "key4", DummyString(1024));
auto iter = OpenTransactionLogIter(3);
ExpectRecords(2, iter);
} while (ChangeCompactOptions());
}
TEST_F(DBTestXactLogIterator, TransactionLogIteratorBlobs) {
Options options = OptionsForLogIterTest();
DestroyAndReopen(options);
CreateAndReopenWithCF({"pikachu"}, options);
{
WriteBatch batch;
batch.Put(handles_[1], "key1", DummyString(1024));
batch.Put(handles_[0], "key2", DummyString(1024));
batch.PutLogData(Slice("blob1"));
batch.Put(handles_[1], "key3", DummyString(1024));
batch.PutLogData(Slice("blob2"));
batch.Delete(handles_[0], "key2");
dbfull()->Write(WriteOptions(), &batch);
ReopenWithColumnFamilies({"default", "pikachu"}, options);
}
auto res = OpenTransactionLogIter(0)->GetBatch();
struct Handler : public WriteBatch::Handler {
std::string seen;
virtual Status PutCF(uint32_t cf, const Slice& key,
const Slice& value) override {
seen += "Put(" + ToString(cf) + ", " + key.ToString() + ", " +
ToString(value.size()) + ")";
return Status::OK();
}
virtual Status MergeCF(uint32_t cf, const Slice& key,
const Slice& value) override {
seen += "Merge(" + ToString(cf) + ", " + key.ToString() + ", " +
ToString(value.size()) + ")";
return Status::OK();
}
virtual void LogData(const Slice& blob) override {
seen += "LogData(" + blob.ToString() + ")";
}
virtual Status DeleteCF(uint32_t cf, const Slice& key) override {
seen += "Delete(" + ToString(cf) + ", " + key.ToString() + ")";
return Status::OK();
}
} handler;
res.writeBatchPtr->Iterate(&handler);
ASSERT_EQ(
"Put(1, key1, 1024)"
"Put(0, key2, 1024)"
"LogData(blob1)"
"Put(1, key3, 1024)"
"LogData(blob2)"
"Delete(0, key2)",
handler.seen);
}
} // namespace rocksdb
#endif // !defined(ROCKSDB_LITE)
int main(int argc, char** argv) {
#if !defined(ROCKSDB_LITE)
rocksdb::port::InstallStackTraceHandler();
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
#else
return 0;
#endif
}