Revision 499ebb3ab5ea4207950fc95acf102b8f58add1c5 authored by Maysam Yabandeh on 24 June 2017, 21:06:43 UTC, committed by Facebook Github Bot on 24 June 2017, 21:11:29 UTC
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
1 parent 0ac4afb
db_io_failure_test.cc
// 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.
#include "db/db_test_util.h"
#include "port/stack_trace.h"
namespace rocksdb {
class DBIOFailureTest : public DBTestBase {
public:
DBIOFailureTest() : DBTestBase("/db_io_failure_test") {}
};
#ifndef ROCKSDB_LITE
// Check that number of files does not grow when writes are dropped
TEST_F(DBIOFailureTest, DropWrites) {
do {
Options options = CurrentOptions();
options.env = env_;
options.paranoid_checks = false;
Reopen(options);
ASSERT_OK(Put("foo", "v1"));
ASSERT_EQ("v1", Get("foo"));
Compact("a", "z");
const size_t num_files = CountFiles();
// Force out-of-space errors
env_->drop_writes_.store(true, std::memory_order_release);
env_->sleep_counter_.Reset();
env_->no_slowdown_ = true;
for (int i = 0; i < 5; i++) {
if (option_config_ != kUniversalCompactionMultiLevel &&
option_config_ != kUniversalSubcompactions) {
for (int level = 0; level < dbfull()->NumberLevels(); level++) {
if (level > 0 && level == dbfull()->NumberLevels() - 1) {
break;
}
dbfull()->TEST_CompactRange(level, nullptr, nullptr, nullptr,
true /* disallow trivial move */);
}
} else {
dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
}
}
std::string property_value;
ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
ASSERT_EQ("5", property_value);
env_->drop_writes_.store(false, std::memory_order_release);
ASSERT_LT(CountFiles(), num_files + 3);
// Check that compaction attempts slept after errors
// TODO @krad: Figure out why ASSERT_EQ 5 keeps failing in certain compiler
// versions
ASSERT_GE(env_->sleep_counter_.Read(), 4);
} while (ChangeCompactOptions());
}
// Check background error counter bumped on flush failures.
TEST_F(DBIOFailureTest, DropWritesFlush) {
do {
Options options = CurrentOptions();
options.env = env_;
options.max_background_flushes = 1;
Reopen(options);
ASSERT_OK(Put("foo", "v1"));
// Force out-of-space errors
env_->drop_writes_.store(true, std::memory_order_release);
std::string property_value;
// Background error count is 0 now.
ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
ASSERT_EQ("0", property_value);
dbfull()->TEST_FlushMemTable(true);
ASSERT_TRUE(db_->GetProperty("rocksdb.background-errors", &property_value));
ASSERT_EQ("1", property_value);
env_->drop_writes_.store(false, std::memory_order_release);
} while (ChangeCompactOptions());
}
#endif // ROCKSDB_LITE
// Check that CompactRange() returns failure if there is not enough space left
// on device
TEST_F(DBIOFailureTest, NoSpaceCompactRange) {
do {
Options options = CurrentOptions();
options.env = env_;
options.disable_auto_compactions = true;
Reopen(options);
// generate 5 tables
for (int i = 0; i < 5; ++i) {
ASSERT_OK(Put(Key(i), Key(i) + "v"));
ASSERT_OK(Flush());
}
// Force out-of-space errors
env_->no_space_.store(true, std::memory_order_release);
Status s = dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr,
true /* disallow trivial move */);
ASSERT_TRUE(s.IsIOError());
ASSERT_TRUE(s.IsNoSpace());
env_->no_space_.store(false, std::memory_order_release);
} while (ChangeCompactOptions());
}
TEST_F(DBIOFailureTest, NonWritableFileSystem) {
do {
Options options = CurrentOptions();
options.write_buffer_size = 4096;
options.arena_block_size = 4096;
options.env = env_;
Reopen(options);
ASSERT_OK(Put("foo", "v1"));
env_->non_writeable_rate_.store(100);
std::string big(100000, 'x');
int errors = 0;
for (int i = 0; i < 20; i++) {
if (!Put("foo", big).ok()) {
errors++;
env_->SleepForMicroseconds(100000);
}
}
ASSERT_GT(errors, 0);
env_->non_writeable_rate_.store(0);
} while (ChangeCompactOptions());
}
#ifndef ROCKSDB_LITE
TEST_F(DBIOFailureTest, ManifestWriteError) {
// Test for the following problem:
// (a) Compaction produces file F
// (b) Log record containing F is written to MANIFEST file, but Sync() fails
// (c) GC deletes F
// (d) After reopening DB, reads fail since deleted F is named in log record
// We iterate twice. In the second iteration, everything is the
// same except the log record never makes it to the MANIFEST file.
for (int iter = 0; iter < 2; iter++) {
std::atomic<bool>* error_type = (iter == 0) ? &env_->manifest_sync_error_
: &env_->manifest_write_error_;
// Insert foo=>bar mapping
Options options = CurrentOptions();
options.env = env_;
options.create_if_missing = true;
options.error_if_exists = false;
options.paranoid_checks = true;
DestroyAndReopen(options);
ASSERT_OK(Put("foo", "bar"));
ASSERT_EQ("bar", Get("foo"));
// Memtable compaction (will succeed)
Flush();
ASSERT_EQ("bar", Get("foo"));
const int last = 2;
MoveFilesToLevel(2);
ASSERT_EQ(NumTableFilesAtLevel(last), 1); // foo=>bar is now in last level
// Merging compaction (will fail)
error_type->store(true, std::memory_order_release);
dbfull()->TEST_CompactRange(last, nullptr, nullptr); // Should fail
ASSERT_EQ("bar", Get("foo"));
error_type->store(false, std::memory_order_release);
// Since paranoid_checks=true, writes should fail
ASSERT_NOK(Put("foo2", "bar2"));
// Recovery: should not lose data
ASSERT_EQ("bar", Get("foo"));
// Try again with paranoid_checks=false
Close();
options.paranoid_checks = false;
Reopen(options);
// Merging compaction (will fail)
error_type->store(true, std::memory_order_release);
dbfull()->TEST_CompactRange(last, nullptr, nullptr); // Should fail
ASSERT_EQ("bar", Get("foo"));
// Recovery: should not lose data
error_type->store(false, std::memory_order_release);
Reopen(options);
ASSERT_EQ("bar", Get("foo"));
// Since paranoid_checks=false, writes should succeed
ASSERT_OK(Put("foo2", "bar2"));
ASSERT_EQ("bar", Get("foo"));
ASSERT_EQ("bar2", Get("foo2"));
}
}
#endif // ROCKSDB_LITE
TEST_F(DBIOFailureTest, PutFailsParanoid) {
// Test the following:
// (a) A random put fails in paranoid mode (simulate by sync fail)
// (b) All other puts have to fail, even if writes would succeed
// (c) All of that should happen ONLY if paranoid_checks = true
Options options = CurrentOptions();
options.env = env_;
options.create_if_missing = true;
options.error_if_exists = false;
options.paranoid_checks = true;
DestroyAndReopen(options);
CreateAndReopenWithCF({"pikachu"}, options);
Status s;
ASSERT_OK(Put(1, "foo", "bar"));
ASSERT_OK(Put(1, "foo1", "bar1"));
// simulate error
env_->log_write_error_.store(true, std::memory_order_release);
s = Put(1, "foo2", "bar2");
ASSERT_TRUE(!s.ok());
env_->log_write_error_.store(false, std::memory_order_release);
s = Put(1, "foo3", "bar3");
// the next put should fail, too
ASSERT_TRUE(!s.ok());
// but we're still able to read
ASSERT_EQ("bar", Get(1, "foo"));
// do the same thing with paranoid checks off
options.paranoid_checks = false;
DestroyAndReopen(options);
CreateAndReopenWithCF({"pikachu"}, options);
ASSERT_OK(Put(1, "foo", "bar"));
ASSERT_OK(Put(1, "foo1", "bar1"));
// simulate error
env_->log_write_error_.store(true, std::memory_order_release);
s = Put(1, "foo2", "bar2");
ASSERT_TRUE(!s.ok());
env_->log_write_error_.store(false, std::memory_order_release);
s = Put(1, "foo3", "bar3");
// the next put should NOT fail
ASSERT_TRUE(s.ok());
}
} // namespace rocksdb
int main(int argc, char** argv) {
rocksdb::port::InstallStackTraceHandler();
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}
Computing file changes ...