rocksdb/db/write_callback_test.cc

453 lines
15 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
#ifndef ROCKSDB_LITE
#include <atomic>
#include <functional>
#include <string>
#include <utility>
#include <vector>
#include "db/db_impl/db_impl.h"
#include "db/write_callback.h"
#include "port/port.h"
#include "rocksdb/db.h"
#include "rocksdb/write_batch.h"
#include "test_util/sync_point.h"
#include "test_util/testharness.h"
#include "util/random.h"
using std::string;
namespace ROCKSDB_NAMESPACE {
class WriteCallbackTest : public testing::Test {
public:
string dbname;
WriteCallbackTest() {
dbname = test::PerThreadDBPath("write_callback_testdb");
}
};
class WriteCallbackTestWriteCallback1 : public WriteCallback {
public:
bool was_called = false;
Status Callback(DB *db) override {
was_called = true;
// Make sure db is a DBImpl
DBImpl* db_impl = dynamic_cast<DBImpl*> (db);
if (db_impl == nullptr) {
return Status::InvalidArgument("");
}
return Status::OK();
}
bool AllowWriteBatching() override { return true; }
};
class WriteCallbackTestWriteCallback2 : public WriteCallback {
public:
Status Callback(DB* /*db*/) override { return Status::Busy(); }
bool AllowWriteBatching() override { return true; }
};
class MockWriteCallback : public WriteCallback {
public:
bool should_fail_ = false;
bool allow_batching_ = false;
std::atomic<bool> was_called_{false};
MockWriteCallback() {}
MockWriteCallback(const MockWriteCallback& other) {
should_fail_ = other.should_fail_;
allow_batching_ = other.allow_batching_;
was_called_.store(other.was_called_.load());
}
Status Callback(DB* /*db*/) override {
was_called_.store(true);
if (should_fail_) {
return Status::Busy();
} else {
return Status::OK();
}
}
bool AllowWriteBatching() override { return allow_batching_; }
};
TEST_F(WriteCallbackTest, WriteWithCallbackTest) {
struct WriteOP {
WriteOP(bool should_fail = false) { callback_.should_fail_ = should_fail; }
void Put(const string& key, const string& val) {
kvs_.push_back(std::make_pair(key, val));
write_batch_.Put(key, val);
}
void Clear() {
kvs_.clear();
write_batch_.Clear();
callback_.was_called_.store(false);
}
MockWriteCallback callback_;
WriteBatch write_batch_;
std::vector<std::pair<string, string>> kvs_;
};
// In each scenario we'll launch multiple threads to write.
// The size of each array equals to number of threads, and
// each boolean in it denote whether callback of corresponding
// thread should succeed or fail.
std::vector<std::vector<WriteOP>> write_scenarios = {
{true},
{false},
{false, false},
{true, true},
{true, false},
{false, true},
{false, false, false},
{true, true, true},
{false, true, false},
{true, false, true},
{true, false, false, false, false},
{false, false, false, false, true},
{false, false, true, false, true},
};
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
for (auto& unordered_write : {true, false}) {
for (auto& seq_per_batch : {true, false}) {
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 23:06:43 +02:00
for (auto& two_queues : {true, false}) {
for (auto& allow_parallel : {true, false}) {
for (auto& allow_batching : {true, false}) {
for (auto& enable_WAL : {true, false}) {
for (auto& enable_pipelined_write : {true, false}) {
for (auto& write_group : write_scenarios) {
Options options;
options.create_if_missing = true;
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
options.unordered_write = unordered_write;
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 23:06:43 +02:00
options.allow_concurrent_memtable_write = allow_parallel;
options.enable_pipelined_write = enable_pipelined_write;
options.two_write_queues = two_queues;
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
// Skip unsupported combinations
if (options.enable_pipelined_write && seq_per_batch) {
continue;
}
if (options.enable_pipelined_write && options.two_write_queues) {
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
continue;
}
if (options.unordered_write &&
!options.allow_concurrent_memtable_write) {
continue;
}
if (options.unordered_write && options.enable_pipelined_write) {
continue;
}
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 23:06:43 +02:00
ReadOptions read_options;
DB* db;
DBImpl* db_impl;
DestroyDB(dbname, options);
DBOptions db_options(options);
ColumnFamilyOptions cf_options(options);
std::vector<ColumnFamilyDescriptor> column_families;
column_families.push_back(
ColumnFamilyDescriptor(kDefaultColumnFamilyName, cf_options));
std::vector<ColumnFamilyHandle*> handles;
auto open_s =
DBImpl::Open(db_options, dbname, column_families, &handles,
&db, seq_per_batch, true /* batch_per_txn */);
ASSERT_OK(open_s);
assert(handles.size() == 1);
delete handles[0];
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 23:06:43 +02:00
db_impl = dynamic_cast<DBImpl*>(db);
ASSERT_TRUE(db_impl);
// Writers that have called JoinBatchGroup.
std::atomic<uint64_t> threads_joining(0);
// Writers that have linked to the queue
std::atomic<uint64_t> threads_linked(0);
// Writers that pass WriteThread::JoinBatchGroup:Wait sync-point.
std::atomic<uint64_t> threads_verified(0);
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 23:06:43 +02:00
std::atomic<uint64_t> seq(db_impl->GetLatestSequenceNumber());
ASSERT_EQ(db_impl->GetLatestSequenceNumber(), 0);
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
"WriteThread::JoinBatchGroup:Start", [&](void*) {
uint64_t cur_threads_joining = threads_joining.fetch_add(1);
// Wait for the last joined writer to link to the queue.
// In this way the writers link to the queue one by one.
// This allows us to confidently detect the first writer
// who increases threads_linked as the leader.
while (threads_linked.load() < cur_threads_joining) {
}
});
// Verification once writers call JoinBatchGroup.
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
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 23:06:43 +02:00
"WriteThread::JoinBatchGroup:Wait", [&](void* arg) {
uint64_t cur_threads_linked = threads_linked.fetch_add(1);
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 23:06:43 +02:00
bool is_leader = false;
bool is_last = false;
// who am i
is_leader = (cur_threads_linked == 0);
is_last = (cur_threads_linked == write_group.size() - 1);
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 23:06:43 +02:00
// check my state
auto* writer = reinterpret_cast<WriteThread::Writer*>(arg);
if (is_leader) {
ASSERT_TRUE(writer->state ==
WriteThread::State::STATE_GROUP_LEADER);
} else {
ASSERT_TRUE(writer->state ==
WriteThread::State::STATE_INIT);
}
// (meta test) the first WriteOP should indeed be the first
// and the last should be the last (all others can be out of
// order)
if (is_leader) {
ASSERT_TRUE(writer->callback->Callback(nullptr).ok() ==
!write_group.front().callback_.should_fail_);
} else if (is_last) {
ASSERT_TRUE(writer->callback->Callback(nullptr).ok() ==
!write_group.back().callback_.should_fail_);
}
threads_verified.fetch_add(1);
// Wait here until all verification in this sync-point
// callback finish for all writers.
while (threads_verified.load() < write_group.size()) {
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 23:06:43 +02:00
}
});
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
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 23:06:43 +02:00
"WriteThread::JoinBatchGroup:DoneWaiting", [&](void* arg) {
// check my state
auto* writer = reinterpret_cast<WriteThread::Writer*>(arg);
if (!allow_batching) {
// no batching so everyone should be a leader
ASSERT_TRUE(writer->state ==
WriteThread::State::STATE_GROUP_LEADER);
} else if (!allow_parallel) {
ASSERT_TRUE(writer->state ==
WriteThread::State::STATE_COMPLETED ||
(enable_pipelined_write &&
writer->state ==
WriteThread::State::
STATE_MEMTABLE_WRITER_LEADER));
}
});
std::atomic<uint32_t> thread_num(0);
std::atomic<char> dummy_key(0);
// Each write thread create a random write batch and write to DB
// with a write callback.
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 23:06:43 +02:00
std::function<void()> write_with_callback_func = [&]() {
uint32_t i = thread_num.fetch_add(1);
Random rnd(i);
// leaders gotta lead
while (i > 0 && threads_verified.load() < 1) {
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 23:06:43 +02:00
}
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 23:06:43 +02:00
// loser has to lose
while (i == write_group.size() - 1 &&
threads_verified.load() < write_group.size() - 1) {
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 23:06:43 +02:00
}
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 23:06:43 +02:00
auto& write_op = write_group.at(i);
write_op.Clear();
write_op.callback_.allow_batching_ = allow_batching;
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 23:06:43 +02:00
// insert some keys
for (uint32_t j = 0; j < rnd.Next() % 50; j++) {
// grab unique key
char my_key = dummy_key.fetch_add(1);
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 23:06:43 +02:00
string skey(5, my_key);
string sval(10, my_key);
write_op.Put(skey, sval);
if (!write_op.callback_.should_fail_ && !seq_per_batch) {
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 23:06:43 +02:00
seq.fetch_add(1);
}
}
if (!write_op.callback_.should_fail_ && seq_per_batch) {
seq.fetch_add(1);
}
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 23:06:43 +02:00
WriteOptions woptions;
woptions.disableWAL = !enable_WAL;
woptions.sync = enable_WAL;
Status s;
if (seq_per_batch) {
class PublishSeqCallback : public PreReleaseCallback {
public:
PublishSeqCallback(DBImpl* db_impl_in)
: db_impl_(db_impl_in) {}
Status Callback(SequenceNumber last_seq, bool /*not used*/,
uint64_t, size_t /*index*/,
size_t /*total*/) override {
db_impl_->SetLastPublishedSequence(last_seq);
return Status::OK();
}
DBImpl* db_impl_;
} publish_seq_callback(db_impl);
// seq_per_batch requires a natural batch separator or Noop
WriteBatchInternal::InsertNoop(&write_op.write_batch_);
const size_t ONE_BATCH = 1;
s = db_impl->WriteImpl(
woptions, &write_op.write_batch_, &write_op.callback_,
nullptr, 0, false, nullptr, ONE_BATCH,
two_queues ? &publish_seq_callback : nullptr);
} else {
s = db_impl->WriteWithCallback(
woptions, &write_op.write_batch_, &write_op.callback_);
}
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 23:06:43 +02:00
if (write_op.callback_.should_fail_) {
ASSERT_TRUE(s.IsBusy());
} else {
ASSERT_OK(s);
}
};
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
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 23:06:43 +02:00
// do all the writes
std::vector<port::Thread> threads;
for (uint32_t i = 0; i < write_group.size(); i++) {
threads.emplace_back(write_with_callback_func);
}
for (auto& t : threads) {
t.join();
}
ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->DisableProcessing();
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 23:06:43 +02:00
// check for keys
string value;
for (auto& w : write_group) {
ASSERT_TRUE(w.callback_.was_called_.load());
for (auto& kvp : w.kvs_) {
if (w.callback_.should_fail_) {
ASSERT_TRUE(
db->Get(read_options, kvp.first, &value).IsNotFound());
} else {
ASSERT_OK(db->Get(read_options, kvp.first, &value));
ASSERT_EQ(value, kvp.second);
}
}
}
ASSERT_EQ(seq.load(), db_impl->TEST_GetLastVisibleSequence());
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 23:06:43 +02:00
delete db;
DestroyDB(dbname, options);
}
}
}
}
}
Unordered Writes (#5218) Summary: Performing unordered writes in rocksdb when unordered_write option is set to true. When enabled the writes to memtable are done without joining any write thread. This offers much higher write throughput since the upcoming writes would not have to wait for the slowest memtable write to finish. The tradeoff is that the writes visible to a snapshot might change over time. If the application cannot tolerate that, it should implement its own mechanisms to work around that. Using TransactionDB with WRITE_PREPARED write policy is one way to achieve that. Doing so increases the max throughput by 2.2x without however compromising the snapshot guarantees. The patch is prepared based on an original by siying Existing unit tests are extended to include unordered_write option. Benchmark Results: ``` TEST_TMPDIR=/dev/shm/ ./db_bench_unordered --benchmarks=fillrandom --threads=32 --num=10000000 -max_write_buffer_number=16 --max_background_jobs=64 --batch_size=8 --writes=3000000 -level0_file_num_compaction_trigger=99999 --level0_slowdown_writes_trigger=99999 --level0_stop_writes_trigger=99999 -enable_pipelined_write=false -disable_auto_compactions --unordered_write=1 ``` With WAL - Vanilla RocksDB: 78.6 MB/s - WRITER_PREPARED with unordered_write: 177.8 MB/s (2.2x) - unordered_write: 368.9 MB/s (4.7x with relaxed snapshot guarantees) Without WAL - Vanilla RocksDB: 111.3 MB/s - WRITER_PREPARED with unordered_write: 259.3 MB/s MB/s (2.3x) - unordered_write: 645.6 MB/s (5.8x with relaxed snapshot guarantees) - WRITER_PREPARED with unordered_write disable concurrency control: 185.3 MB/s MB/s (2.35x) Limitations: - The feature is not yet extended to `max_successive_merges` > 0. The feature is also incompatible with `enable_pipelined_write` = true as well as with `allow_concurrent_memtable_write` = false. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5218 Differential Revision: D15219029 Pulled By: maysamyabandeh fbshipit-source-id: 38f2abc4af8780148c6128acdba2b3227bc81759
2019-05-14 02:43:47 +02:00
}
}
}
}
TEST_F(WriteCallbackTest, WriteCallBackTest) {
Options options;
WriteOptions write_options;
ReadOptions read_options;
string value;
DB* db;
DBImpl* db_impl;
DestroyDB(dbname, options);
options.create_if_missing = true;
Status s = DB::Open(options, dbname, &db);
ASSERT_OK(s);
db_impl = dynamic_cast<DBImpl*> (db);
ASSERT_TRUE(db_impl);
WriteBatch wb;
wb.Put("a", "value.a");
wb.Delete("x");
// Test a simple Write
s = db->Write(write_options, &wb);
ASSERT_OK(s);
s = db->Get(read_options, "a", &value);
ASSERT_OK(s);
ASSERT_EQ("value.a", value);
// Test WriteWithCallback
WriteCallbackTestWriteCallback1 callback1;
WriteBatch wb2;
wb2.Put("a", "value.a2");
s = db_impl->WriteWithCallback(write_options, &wb2, &callback1);
ASSERT_OK(s);
ASSERT_TRUE(callback1.was_called);
s = db->Get(read_options, "a", &value);
ASSERT_OK(s);
ASSERT_EQ("value.a2", value);
// Test WriteWithCallback for a callback that fails
WriteCallbackTestWriteCallback2 callback2;
WriteBatch wb3;
wb3.Put("a", "value.a3");
s = db_impl->WriteWithCallback(write_options, &wb3, &callback2);
ASSERT_NOK(s);
s = db->Get(read_options, "a", &value);
ASSERT_OK(s);
ASSERT_EQ("value.a2", value);
delete db;
DestroyDB(dbname, options);
}
} // namespace ROCKSDB_NAMESPACE
int main(int argc, char** argv) {
::testing::InitGoogleTest(&argc, argv);
return RUN_ALL_TESTS();
}
#else
#include <stdio.h>
int main(int /*argc*/, char** /*argv*/) {
fprintf(stderr,
"SKIPPED as WriteWithCallback is not supported in ROCKSDB_LITE\n");
return 0;
}
#endif // !ROCKSDB_LITE