rocksdb/db/fault_injection_test.cc

550 lines
17 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 2014 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.
// This test uses a custom Env to keep track of the state of a filesystem as of
// the last "sync". It then checks for data loss errors by purposely dropping
// file data (or entire files) not protected by a "sync".
#include "db/db_impl.h"
#include "db/log_format.h"
#include "db/version_set.h"
#include "env/mock_env.h"
#include "rocksdb/cache.h"
#include "rocksdb/db.h"
#include "rocksdb/env.h"
#include "rocksdb/table.h"
#include "rocksdb/write_batch.h"
#include "util/fault_injection_test_env.h"
#include "util/filename.h"
#include "util/logging.h"
#include "util/mutexlock.h"
#include "util/sync_point.h"
#include "util/testharness.h"
#include "util/testutil.h"
namespace rocksdb {
static const int kValueSize = 1000;
static const int kMaxNumValues = 2000;
static const size_t kNumIterations = 3;
class FaultInjectionTest : public testing::Test,
public testing::WithParamInterface<bool> {
protected:
enum OptionConfig {
kDefault,
kDifferentDataDir,
kWalDir,
kSyncWal,
kWalDirSyncWal,
kMultiLevels,
kEnd,
};
int option_config_;
// When need to make sure data is persistent, sync WAL
bool sync_use_wal_;
// When need to make sure data is persistent, call DB::CompactRange()
bool sync_use_compact_;
bool sequential_order_;
protected:
public:
enum ExpectedVerifResult { kValExpectFound, kValExpectNoError };
enum ResetMethod {
kResetDropUnsyncedData,
kResetDropRandomUnsyncedData,
kResetDeleteUnsyncedFiles,
kResetDropAndDeleteUnsynced
};
std::unique_ptr<Env> base_env_;
FaultInjectionTestEnv* env_;
std::string dbname_;
shared_ptr<Cache> tiny_cache_;
Options options_;
DB* db_;
FaultInjectionTest()
: option_config_(kDefault),
sync_use_wal_(false),
sync_use_compact_(true),
base_env_(nullptr),
env_(NULL),
db_(NULL) {
}
~FaultInjectionTest() {
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
}
bool ChangeOptions() {
option_config_++;
if (option_config_ >= kEnd) {
return false;
} else {
if (option_config_ == kMultiLevels) {
base_env_.reset(new MockEnv(Env::Default()));
}
return true;
}
}
// Return the current option configuration.
Options CurrentOptions() {
sync_use_wal_ = false;
sync_use_compact_ = true;
Options options;
switch (option_config_) {
case kWalDir:
options.wal_dir = test::TmpDir(env_) + "/fault_test_wal";
break;
case kDifferentDataDir:
options.db_paths.emplace_back(test::TmpDir(env_) + "/fault_test_data",
1000000U);
break;
case kSyncWal:
sync_use_wal_ = true;
sync_use_compact_ = false;
break;
case kWalDirSyncWal:
options.wal_dir = test::TmpDir(env_) + "/fault_test_wal";
sync_use_wal_ = true;
sync_use_compact_ = false;
break;
case kMultiLevels:
options.write_buffer_size = 64 * 1024;
options.target_file_size_base = 64 * 1024;
options.level0_file_num_compaction_trigger = 2;
options.level0_slowdown_writes_trigger = 2;
options.level0_stop_writes_trigger = 4;
options.max_bytes_for_level_base = 128 * 1024;
options.max_write_buffer_number = 2;
options.max_background_compactions = 8;
options.max_background_flushes = 8;
sync_use_wal_ = true;
sync_use_compact_ = false;
break;
default:
break;
}
return options;
}
Status NewDB() {
assert(db_ == NULL);
assert(tiny_cache_ == nullptr);
assert(env_ == NULL);
env_ =
new FaultInjectionTestEnv(base_env_ ? base_env_.get() : Env::Default());
options_ = CurrentOptions();
options_.env = env_;
options_.paranoid_checks = true;
BlockBasedTableOptions table_options;
tiny_cache_ = NewLRUCache(100);
table_options.block_cache = tiny_cache_;
options_.table_factory.reset(NewBlockBasedTableFactory(table_options));
dbname_ = test::TmpDir() + "/fault_test";
rocksdb: Replace ASSERT* with EXPECT* in functions that does not return void value Summary: gtest does not use exceptions to fail a unit test by design, and `ASSERT*`s are implemented using `return`. As a consequence we cannot use `ASSERT*` in a function that does not return `void` value ([[ https://code.google.com/p/googletest/wiki/AdvancedGuide#Assertion_Placement | 1]]), and have to fix our existing code. This diff does this in a generic way, with no manual changes. In order to detect all existing `ASSERT*` that are used in functions that doesn't return void value, I change the code to generate compile errors for such cases. In `util/testharness.h` I defined `EXPECT*` assertions, the same way as `ASSERT*`, and redefined `ASSERT*` to return `void`. Then executed: ```lang=bash % USE_CLANG=1 make all -j55 -k 2> build.log % perl -naF: -e 'print "-- -number=".$F[1]." ".$F[0]."\n" if /: error:/' \ build.log | xargs -L 1 perl -spi -e 's/ASSERT/EXPECT/g if $. == $number' % make format ``` After that I reverted back change to `ASSERT*` in `util/testharness.h`. But preserved introduced `EXPECT*`, which is the same as `ASSERT*`. This will be deleted once switched to gtest. This diff is independent and contains manual changes only in `util/testharness.h`. Test Plan: Make sure all tests are passing. ```lang=bash % USE_CLANG=1 make check ``` Reviewers: igor, lgalanis, sdong, yufei.zhu, rven, meyering Reviewed By: meyering Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33333
2015-03-17 03:52:32 +00:00
EXPECT_OK(DestroyDB(dbname_, options_));
options_.create_if_missing = true;
Status s = OpenDB();
options_.create_if_missing = false;
return s;
}
void SetUp() override {
sequential_order_ = GetParam();
ASSERT_OK(NewDB());
}
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 21:08:00 +00:00
void TearDown() override {
CloseDB();
Status s = DestroyDB(dbname_, options_);
delete env_;
env_ = NULL;
tiny_cache_.reset();
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 21:08:00 +00:00
ASSERT_OK(s);
}
void Build(const WriteOptions& write_options, int start_idx, int num_vals) {
std::string key_space, value_space;
WriteBatch batch;
for (int i = start_idx; i < start_idx + num_vals; i++) {
Slice key = Key(i, &key_space);
batch.Clear();
batch.Put(key, Value(i, &value_space));
ASSERT_OK(db_->Write(write_options, &batch));
}
}
Status ReadValue(int i, std::string* val) const {
std::string key_space, value_space;
Slice key = Key(i, &key_space);
Value(i, &value_space);
ReadOptions options;
return db_->Get(options, key, val);
}
Status Verify(int start_idx, int num_vals,
ExpectedVerifResult expected) const {
std::string val;
std::string value_space;
Status s;
for (int i = start_idx; i < start_idx + num_vals && s.ok(); i++) {
Value(i, &value_space);
s = ReadValue(i, &val);
if (s.ok()) {
rocksdb: Replace ASSERT* with EXPECT* in functions that does not return void value Summary: gtest does not use exceptions to fail a unit test by design, and `ASSERT*`s are implemented using `return`. As a consequence we cannot use `ASSERT*` in a function that does not return `void` value ([[ https://code.google.com/p/googletest/wiki/AdvancedGuide#Assertion_Placement | 1]]), and have to fix our existing code. This diff does this in a generic way, with no manual changes. In order to detect all existing `ASSERT*` that are used in functions that doesn't return void value, I change the code to generate compile errors for such cases. In `util/testharness.h` I defined `EXPECT*` assertions, the same way as `ASSERT*`, and redefined `ASSERT*` to return `void`. Then executed: ```lang=bash % USE_CLANG=1 make all -j55 -k 2> build.log % perl -naF: -e 'print "-- -number=".$F[1]." ".$F[0]."\n" if /: error:/' \ build.log | xargs -L 1 perl -spi -e 's/ASSERT/EXPECT/g if $. == $number' % make format ``` After that I reverted back change to `ASSERT*` in `util/testharness.h`. But preserved introduced `EXPECT*`, which is the same as `ASSERT*`. This will be deleted once switched to gtest. This diff is independent and contains manual changes only in `util/testharness.h`. Test Plan: Make sure all tests are passing. ```lang=bash % USE_CLANG=1 make check ``` Reviewers: igor, lgalanis, sdong, yufei.zhu, rven, meyering Reviewed By: meyering Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33333
2015-03-17 03:52:32 +00:00
EXPECT_EQ(value_space, val);
}
if (expected == kValExpectFound) {
if (!s.ok()) {
fprintf(stderr, "Error when read %dth record (expect found): %s\n", i,
s.ToString().c_str());
return s;
}
} else if (!s.ok() && !s.IsNotFound()) {
fprintf(stderr, "Error when read %dth record: %s\n", i,
s.ToString().c_str());
return s;
}
}
return Status::OK();
}
#ifdef ROCKSDB_UBSAN_RUN
#if defined(__clang__)
__attribute__((__no_sanitize__("shift"), no_sanitize("signed-integer-overflow")))
#elif defined(__GNUC__)
__attribute__((__no_sanitize_undefined__))
#endif
#endif
// Return the ith key
Slice Key(int i, std::string* storage) const {
int num = i;
if (!sequential_order_) {
// random transfer
const int m = 0x5bd1e995;
num *= m;
num ^= num << 24;
}
char buf[100];
snprintf(buf, sizeof(buf), "%016d", num);
storage->assign(buf, strlen(buf));
return Slice(*storage);
}
// Return the value to associate with the specified key
Slice Value(int k, std::string* storage) const {
Random r(k);
return test::RandomString(&r, kValueSize, storage);
}
void CloseDB() {
delete db_;
db_ = nullptr;
}
Status OpenDB() {
CloseDB();
env_->ResetState();
Status s = DB::Open(options_, dbname_, &db_);
assert(db_ != nullptr);
return s;
}
void DeleteAllData() {
Iterator* iter = db_->NewIterator(ReadOptions());
WriteOptions options;
for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
ASSERT_OK(db_->Delete(WriteOptions(), iter->key()));
}
delete iter;
FlushOptions flush_options;
flush_options.wait = true;
db_->Flush(flush_options);
}
// rnd cannot be null for kResetDropRandomUnsyncedData
void ResetDBState(ResetMethod reset_method, Random* rnd = nullptr) {
env_->AssertNoOpenFile();
switch (reset_method) {
case kResetDropUnsyncedData:
ASSERT_OK(env_->DropUnsyncedFileData());
break;
case kResetDropRandomUnsyncedData:
ASSERT_OK(env_->DropRandomUnsyncedFileData(rnd));
break;
case kResetDeleteUnsyncedFiles:
ASSERT_OK(env_->DeleteFilesCreatedAfterLastDirSync());
break;
case kResetDropAndDeleteUnsynced:
ASSERT_OK(env_->DropUnsyncedFileData());
ASSERT_OK(env_->DeleteFilesCreatedAfterLastDirSync());
break;
default:
assert(false);
}
}
void PartialCompactTestPreFault(int num_pre_sync, int num_post_sync) {
DeleteAllData();
WriteOptions write_options;
write_options.sync = sync_use_wal_;
Build(write_options, 0, num_pre_sync);
if (sync_use_compact_) {
db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
}
write_options.sync = false;
Build(write_options, num_pre_sync, num_post_sync);
}
void PartialCompactTestReopenWithFault(ResetMethod reset_method,
int num_pre_sync, int num_post_sync,
Random* rnd = nullptr) {
env_->SetFilesystemActive(false);
CloseDB();
ResetDBState(reset_method, rnd);
ASSERT_OK(OpenDB());
ASSERT_OK(Verify(0, num_pre_sync, FaultInjectionTest::kValExpectFound));
ASSERT_OK(Verify(num_pre_sync, num_post_sync,
FaultInjectionTest::kValExpectNoError));
WaitCompactionFinish();
ASSERT_OK(Verify(0, num_pre_sync, FaultInjectionTest::kValExpectFound));
ASSERT_OK(Verify(num_pre_sync, num_post_sync,
FaultInjectionTest::kValExpectNoError));
}
void NoWriteTestPreFault() {
}
void NoWriteTestReopenWithFault(ResetMethod reset_method) {
CloseDB();
ResetDBState(reset_method);
ASSERT_OK(OpenDB());
}
void WaitCompactionFinish() {
static_cast<DBImpl*>(db_->GetRootDB())->TEST_WaitForCompact();
ASSERT_OK(db_->Put(WriteOptions(), "", ""));
}
};
TEST_P(FaultInjectionTest, FaultTest) {
do {
Random rnd(301);
for (size_t idx = 0; idx < kNumIterations; idx++) {
int num_pre_sync = rnd.Uniform(kMaxNumValues);
int num_post_sync = rnd.Uniform(kMaxNumValues);
PartialCompactTestPreFault(num_pre_sync, num_post_sync);
PartialCompactTestReopenWithFault(kResetDropUnsyncedData, num_pre_sync,
num_post_sync);
NoWriteTestPreFault();
NoWriteTestReopenWithFault(kResetDropUnsyncedData);
PartialCompactTestPreFault(num_pre_sync, num_post_sync);
PartialCompactTestReopenWithFault(kResetDropRandomUnsyncedData,
num_pre_sync, num_post_sync, &rnd);
NoWriteTestPreFault();
NoWriteTestReopenWithFault(kResetDropUnsyncedData);
// Setting a separate data path won't pass the test as we don't sync
// it after creating new files,
PartialCompactTestPreFault(num_pre_sync, num_post_sync);
PartialCompactTestReopenWithFault(kResetDropAndDeleteUnsynced,
num_pre_sync, num_post_sync);
NoWriteTestPreFault();
NoWriteTestReopenWithFault(kResetDropAndDeleteUnsynced);
PartialCompactTestPreFault(num_pre_sync, num_post_sync);
// No new files created so we expect all values since no files will be
// dropped.
PartialCompactTestReopenWithFault(kResetDeleteUnsyncedFiles, num_pre_sync,
num_post_sync);
NoWriteTestPreFault();
NoWriteTestReopenWithFault(kResetDeleteUnsyncedFiles);
}
} while (ChangeOptions());
}
// Previous log file is not fsynced if sync is forced after log rolling.
TEST_P(FaultInjectionTest, WriteOptionSyncTest) {
test::SleepingBackgroundTask sleeping_task_low;
env_->SetBackgroundThreads(1, Env::HIGH);
// Block the job queue to prevent flush job from running.
env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
Env::Priority::HIGH);
sleeping_task_low.WaitUntilSleeping();
WriteOptions write_options;
write_options.sync = false;
std::string key_space, value_space;
ASSERT_OK(
db_->Put(write_options, Key(1, &key_space), Value(1, &value_space)));
FlushOptions flush_options;
flush_options.wait = false;
ASSERT_OK(db_->Flush(flush_options));
write_options.sync = true;
ASSERT_OK(
db_->Put(write_options, Key(2, &key_space), Value(2, &value_space)));
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
db_->FlushWAL(false);
env_->SetFilesystemActive(false);
NoWriteTestReopenWithFault(kResetDropAndDeleteUnsynced);
sleeping_task_low.WakeUp();
sleeping_task_low.WaitUntilDone();
ASSERT_OK(OpenDB());
std::string val;
Value(2, &value_space);
ASSERT_OK(ReadValue(2, &val));
ASSERT_EQ(value_space, val);
Value(1, &value_space);
ASSERT_OK(ReadValue(1, &val));
ASSERT_EQ(value_space, val);
}
TEST_P(FaultInjectionTest, UninstalledCompaction) {
options_.target_file_size_base = 32 * 1024;
options_.write_buffer_size = 100 << 10; // 100KB
options_.level0_file_num_compaction_trigger = 6;
options_.level0_stop_writes_trigger = 1 << 10;
options_.level0_slowdown_writes_trigger = 1 << 10;
options_.max_background_compactions = 1;
OpenDB();
if (!sequential_order_) {
rocksdb::SyncPoint::GetInstance()->LoadDependency({
{"FaultInjectionTest::FaultTest:0", "DBImpl::BGWorkCompaction"},
{"CompactionJob::Run():End", "FaultInjectionTest::FaultTest:1"},
{"FaultInjectionTest::FaultTest:2",
"DBImpl::BackgroundCompaction:NonTrivial:AfterRun"},
});
}
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
int kNumKeys = 1000;
Build(WriteOptions(), 0, kNumKeys);
FlushOptions flush_options;
flush_options.wait = true;
db_->Flush(flush_options);
ASSERT_OK(db_->Put(WriteOptions(), "", ""));
TEST_SYNC_POINT("FaultInjectionTest::FaultTest:0");
TEST_SYNC_POINT("FaultInjectionTest::FaultTest:1");
env_->SetFilesystemActive(false);
TEST_SYNC_POINT("FaultInjectionTest::FaultTest:2");
CloseDB();
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
ResetDBState(kResetDropUnsyncedData);
std::atomic<bool> opened(false);
rocksdb::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::Open:Opened", [&](void* arg) { opened.store(true); });
rocksdb::SyncPoint::GetInstance()->SetCallBack(
"DBImpl::BGWorkCompaction",
[&](void* arg) { ASSERT_TRUE(opened.load()); });
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
ASSERT_OK(OpenDB());
ASSERT_OK(Verify(0, kNumKeys, FaultInjectionTest::kValExpectFound));
WaitCompactionFinish();
ASSERT_OK(Verify(0, kNumKeys, FaultInjectionTest::kValExpectFound));
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
}
TEST_P(FaultInjectionTest, ManualLogSyncTest) {
test::SleepingBackgroundTask sleeping_task_low;
env_->SetBackgroundThreads(1, Env::HIGH);
// Block the job queue to prevent flush job from running.
env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
Env::Priority::HIGH);
sleeping_task_low.WaitUntilSleeping();
WriteOptions write_options;
write_options.sync = false;
std::string key_space, value_space;
ASSERT_OK(
db_->Put(write_options, Key(1, &key_space), Value(1, &value_space)));
FlushOptions flush_options;
flush_options.wait = false;
ASSERT_OK(db_->Flush(flush_options));
ASSERT_OK(
db_->Put(write_options, Key(2, &key_space), Value(2, &value_space)));
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
ASSERT_OK(db_->FlushWAL(true));
env_->SetFilesystemActive(false);
NoWriteTestReopenWithFault(kResetDropAndDeleteUnsynced);
sleeping_task_low.WakeUp();
sleeping_task_low.WaitUntilDone();
ASSERT_OK(OpenDB());
std::string val;
Value(2, &value_space);
ASSERT_OK(ReadValue(2, &val));
ASSERT_EQ(value_space, val);
Value(1, &value_space);
ASSERT_OK(ReadValue(1, &val));
ASSERT_EQ(value_space, val);
}
TEST_P(FaultInjectionTest, WriteBatchWalTerminationTest) {
ReadOptions ro;
Options options = CurrentOptions();
options.env = env_;
WriteOptions wo;
wo.sync = true;
wo.disableWAL = false;
WriteBatch batch;
batch.Put("cats", "dogs");
batch.MarkWalTerminationPoint();
batch.Put("boys", "girls");
ASSERT_OK(db_->Write(wo, &batch));
env_->SetFilesystemActive(false);
NoWriteTestReopenWithFault(kResetDropAndDeleteUnsynced);
ASSERT_OK(OpenDB());
std::string val;
ASSERT_OK(db_->Get(ro, "cats", &val));
ASSERT_EQ("dogs", val);
ASSERT_EQ(db_->Get(ro, "boys", &val), Status::NotFound());
}
INSTANTIATE_TEST_CASE_P(FaultTest, FaultInjectionTest, ::testing::Bool());
} // namespace rocksdb
int main(int argc, char** argv) {
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 21:08:00 +00:00
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}