2016-02-09 23:12:00 +00:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-15 23:03:42 +00:00
|
|
|
// 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).
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
#ifndef ROCKSDB_LITE
|
|
|
|
|
|
|
|
#include "utilities/transactions/transaction_lock_mgr.h"
|
|
|
|
|
2019-06-06 20:52:39 +00:00
|
|
|
#include <cinttypes>
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
#include <algorithm>
|
|
|
|
#include <mutex>
|
|
|
|
|
2017-11-06 18:52:17 +00:00
|
|
|
#include "monitoring/perf_context_imp.h"
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
#include "rocksdb/slice.h"
|
2015-09-08 19:36:48 +00:00
|
|
|
#include "rocksdb/utilities/transaction_db_mutex.h"
|
2019-05-31 00:39:43 +00:00
|
|
|
#include "test_util/sync_point.h"
|
2017-07-28 23:23:50 +00:00
|
|
|
#include "util/cast_util.h"
|
2019-04-08 20:24:29 +00:00
|
|
|
#include "util/hash.h"
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
#include "util/thread_local.h"
|
2017-08-06 00:17:48 +00:00
|
|
|
#include "utilities/transactions/pessimistic_transaction_db.h"
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
namespace ROCKSDB_NAMESPACE {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
struct LockInfo {
|
2016-12-06 01:18:14 +00:00
|
|
|
bool exclusive;
|
|
|
|
autovector<TransactionID> txn_ids;
|
2015-09-08 19:36:48 +00:00
|
|
|
|
|
|
|
// Transaction locks are not valid after this time in us
|
|
|
|
uint64_t expiration_time;
|
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
LockInfo(TransactionID id, uint64_t time, bool ex)
|
|
|
|
: exclusive(ex), expiration_time(time) {
|
|
|
|
txn_ids.push_back(id);
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
LockInfo(const LockInfo& lock_info)
|
2016-12-06 01:18:14 +00:00
|
|
|
: exclusive(lock_info.exclusive),
|
|
|
|
txn_ids(lock_info.txn_ids),
|
|
|
|
expiration_time(lock_info.expiration_time) {}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
struct LockMapStripe {
|
2015-09-08 19:36:48 +00:00
|
|
|
explicit LockMapStripe(std::shared_ptr<TransactionDBMutexFactory> factory) {
|
|
|
|
stripe_mutex = factory->AllocateMutex();
|
|
|
|
stripe_cv = factory->AllocateCondVar();
|
|
|
|
assert(stripe_mutex);
|
|
|
|
assert(stripe_cv);
|
|
|
|
}
|
|
|
|
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// Mutex must be held before modifying keys map
|
2015-09-08 19:36:48 +00:00
|
|
|
std::shared_ptr<TransactionDBMutex> stripe_mutex;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
// Condition Variable per stripe for waiting on a lock
|
2015-09-08 19:36:48 +00:00
|
|
|
std::shared_ptr<TransactionDBCondVar> stripe_cv;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
// Locked keys mapped to the info about the transactions that locked them.
|
|
|
|
// TODO(agiardullo): Explore performance of other data structures.
|
|
|
|
std::unordered_map<std::string, LockInfo> keys;
|
|
|
|
};
|
|
|
|
|
|
|
|
// Map of #num_stripes LockMapStripes
|
|
|
|
struct LockMap {
|
2015-09-08 19:36:48 +00:00
|
|
|
explicit LockMap(size_t num_stripes,
|
|
|
|
std::shared_ptr<TransactionDBMutexFactory> factory)
|
|
|
|
: num_stripes_(num_stripes) {
|
|
|
|
lock_map_stripes_.reserve(num_stripes);
|
|
|
|
for (size_t i = 0; i < num_stripes; i++) {
|
|
|
|
LockMapStripe* stripe = new LockMapStripe(factory);
|
|
|
|
lock_map_stripes_.push_back(stripe);
|
|
|
|
}
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
~LockMap() {
|
|
|
|
for (auto stripe : lock_map_stripes_) {
|
|
|
|
delete stripe;
|
|
|
|
}
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
// Number of sepearate LockMapStripes to create, each with their own Mutex
|
|
|
|
const size_t num_stripes_;
|
|
|
|
|
|
|
|
// Count of keys that are currently locked in this column family.
|
|
|
|
// (Only maintained if TransactionLockMgr::max_num_locks_ is positive.)
|
|
|
|
std::atomic<int64_t> lock_cnt{0};
|
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
std::vector<LockMapStripe*> lock_map_stripes_;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
size_t GetStripe(const std::string& key) const;
|
|
|
|
};
|
|
|
|
|
2017-08-18 01:49:30 +00:00
|
|
|
void DeadlockInfoBuffer::AddNewPath(DeadlockPath path) {
|
|
|
|
std::lock_guard<std::mutex> lock(paths_buffer_mutex_);
|
|
|
|
|
|
|
|
if (paths_buffer_.empty()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2018-10-23 17:33:55 +00:00
|
|
|
paths_buffer_[buffer_idx_] = std::move(path);
|
2017-08-18 01:49:30 +00:00
|
|
|
buffer_idx_ = (buffer_idx_ + 1) % paths_buffer_.size();
|
|
|
|
}
|
|
|
|
|
|
|
|
void DeadlockInfoBuffer::Resize(uint32_t target_size) {
|
|
|
|
std::lock_guard<std::mutex> lock(paths_buffer_mutex_);
|
|
|
|
|
|
|
|
paths_buffer_ = Normalize();
|
|
|
|
|
|
|
|
// Drop the deadlocks that will no longer be needed ater the normalize
|
|
|
|
if (target_size < paths_buffer_.size()) {
|
|
|
|
paths_buffer_.erase(
|
|
|
|
paths_buffer_.begin(),
|
|
|
|
paths_buffer_.begin() + (paths_buffer_.size() - target_size));
|
|
|
|
buffer_idx_ = 0;
|
|
|
|
}
|
|
|
|
// Resize the buffer to the target size and restore the buffer's idx
|
|
|
|
else {
|
|
|
|
auto prev_size = paths_buffer_.size();
|
|
|
|
paths_buffer_.resize(target_size);
|
|
|
|
buffer_idx_ = (uint32_t)prev_size;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<DeadlockPath> DeadlockInfoBuffer::Normalize() {
|
|
|
|
auto working = paths_buffer_;
|
|
|
|
|
|
|
|
if (working.empty()) {
|
|
|
|
return working;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Next write occurs at a nonexistent path's slot
|
|
|
|
if (paths_buffer_[buffer_idx_].empty()) {
|
|
|
|
working.resize(buffer_idx_);
|
|
|
|
} else {
|
|
|
|
std::rotate(working.begin(), working.begin() + buffer_idx_, working.end());
|
|
|
|
}
|
|
|
|
|
|
|
|
return working;
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<DeadlockPath> DeadlockInfoBuffer::PrepareBuffer() {
|
|
|
|
std::lock_guard<std::mutex> lock(paths_buffer_mutex_);
|
|
|
|
|
|
|
|
// Reversing the normalized vector returns the latest deadlocks first
|
|
|
|
auto working = Normalize();
|
|
|
|
std::reverse(working.begin(), working.end());
|
|
|
|
|
|
|
|
return working;
|
|
|
|
}
|
|
|
|
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
namespace {
|
|
|
|
void UnrefLockMapsCache(void* ptr) {
|
|
|
|
// Called when a thread exits or a ThreadLocalPtr gets destroyed.
|
|
|
|
auto lock_maps_cache =
|
|
|
|
static_cast<std::unordered_map<uint32_t, std::shared_ptr<LockMap>>*>(ptr);
|
|
|
|
delete lock_maps_cache;
|
|
|
|
}
|
|
|
|
} // anonymous namespace
|
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
TransactionLockMgr::TransactionLockMgr(
|
2016-02-02 01:07:05 +00:00
|
|
|
TransactionDB* txn_db, size_t default_num_stripes, int64_t max_num_locks,
|
2017-08-18 01:49:30 +00:00
|
|
|
uint32_t max_num_deadlocks,
|
2015-09-08 19:36:48 +00:00
|
|
|
std::shared_ptr<TransactionDBMutexFactory> mutex_factory)
|
2016-02-02 01:07:05 +00:00
|
|
|
: txn_db_impl_(nullptr),
|
|
|
|
default_num_stripes_(default_num_stripes),
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
max_num_locks_(max_num_locks),
|
2016-09-28 08:23:33 +00:00
|
|
|
lock_maps_cache_(new ThreadLocalPtr(&UnrefLockMapsCache)),
|
2017-08-18 01:49:30 +00:00
|
|
|
dlock_buffer_(max_num_deadlocks),
|
2016-09-28 08:23:33 +00:00
|
|
|
mutex_factory_(mutex_factory) {
|
2017-07-28 23:23:50 +00:00
|
|
|
assert(txn_db);
|
2020-04-29 20:06:27 +00:00
|
|
|
txn_db_impl_ = static_cast_with_check<PessimisticTransactionDB>(txn_db);
|
2016-02-02 01:07:05 +00:00
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
TransactionLockMgr::~TransactionLockMgr() {}
|
|
|
|
|
|
|
|
size_t LockMap::GetStripe(const std::string& key) const {
|
|
|
|
assert(num_stripes_ > 0);
|
2020-09-28 18:33:31 +00:00
|
|
|
return FastRange64(GetSliceNPHash64(key), num_stripes_);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void TransactionLockMgr::AddColumnFamily(uint32_t column_family_id) {
|
|
|
|
InstrumentedMutexLock l(&lock_map_mutex_);
|
|
|
|
|
|
|
|
if (lock_maps_.find(column_family_id) == lock_maps_.end()) {
|
2015-09-08 19:36:48 +00:00
|
|
|
lock_maps_.emplace(column_family_id,
|
2019-04-15 17:30:48 +00:00
|
|
|
std::make_shared<LockMap>(default_num_stripes_, mutex_factory_));
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
} else {
|
|
|
|
// column_family already exists in lock map
|
|
|
|
assert(false);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void TransactionLockMgr::RemoveColumnFamily(uint32_t column_family_id) {
|
|
|
|
// Remove lock_map for this column family. Since the lock map is stored
|
2016-04-28 09:30:44 +00:00
|
|
|
// as a shared ptr, concurrent transactions can still keep using it
|
|
|
|
// until they release their references to it.
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
{
|
|
|
|
InstrumentedMutexLock l(&lock_map_mutex_);
|
|
|
|
|
|
|
|
auto lock_maps_iter = lock_maps_.find(column_family_id);
|
2020-04-08 20:49:17 +00:00
|
|
|
if (lock_maps_iter == lock_maps_.end()) {
|
|
|
|
return;
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
lock_maps_.erase(lock_maps_iter);
|
|
|
|
} // lock_map_mutex_
|
|
|
|
|
|
|
|
// Clear all thread-local caches
|
|
|
|
autovector<void*> local_caches;
|
|
|
|
lock_maps_cache_->Scrape(&local_caches, nullptr);
|
|
|
|
for (auto cache : local_caches) {
|
|
|
|
delete static_cast<LockMaps*>(cache);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-11-09 19:17:34 +00:00
|
|
|
// Look up the LockMap std::shared_ptr for a given column_family_id.
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// Note: The LockMap is only valid as long as the caller is still holding on
|
2018-11-09 19:17:34 +00:00
|
|
|
// to the returned std::shared_ptr.
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
std::shared_ptr<LockMap> TransactionLockMgr::GetLockMap(
|
|
|
|
uint32_t column_family_id) {
|
|
|
|
// First check thread-local cache
|
|
|
|
if (lock_maps_cache_->Get() == nullptr) {
|
|
|
|
lock_maps_cache_->Reset(new LockMaps());
|
|
|
|
}
|
|
|
|
|
|
|
|
auto lock_maps_cache = static_cast<LockMaps*>(lock_maps_cache_->Get());
|
|
|
|
|
|
|
|
auto lock_map_iter = lock_maps_cache->find(column_family_id);
|
|
|
|
if (lock_map_iter != lock_maps_cache->end()) {
|
|
|
|
// Found lock map for this column family.
|
|
|
|
return lock_map_iter->second;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Not found in local cache, grab mutex and check shared LockMaps
|
|
|
|
InstrumentedMutexLock l(&lock_map_mutex_);
|
|
|
|
|
|
|
|
lock_map_iter = lock_maps_.find(column_family_id);
|
|
|
|
if (lock_map_iter == lock_maps_.end()) {
|
|
|
|
return std::shared_ptr<LockMap>(nullptr);
|
|
|
|
} else {
|
|
|
|
// Found lock map. Store in thread-local cache and return.
|
|
|
|
std::shared_ptr<LockMap>& lock_map = lock_map_iter->second;
|
|
|
|
lock_maps_cache->insert({column_family_id, lock_map});
|
|
|
|
|
|
|
|
return lock_map;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Returns true if this lock has expired and can be acquired by another
|
|
|
|
// transaction.
|
2015-09-08 19:36:48 +00:00
|
|
|
// If false, sets *expire_time to the expiration time of the lock according
|
|
|
|
// to Env->GetMicros() or 0 if no expiration.
|
2016-12-06 01:18:14 +00:00
|
|
|
bool TransactionLockMgr::IsLockExpired(TransactionID txn_id,
|
|
|
|
const LockInfo& lock_info, Env* env,
|
2015-09-08 19:36:48 +00:00
|
|
|
uint64_t* expire_time) {
|
2020-03-27 23:12:40 +00:00
|
|
|
if (lock_info.expiration_time == 0) {
|
|
|
|
*expire_time = 0;
|
|
|
|
return false;
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2020-03-27 23:12:40 +00:00
|
|
|
auto now = env->NowMicros();
|
|
|
|
bool expired = lock_info.expiration_time <= now;
|
|
|
|
if (!expired) {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// return how many microseconds until lock will be expired
|
2015-09-08 19:36:48 +00:00
|
|
|
*expire_time = lock_info.expiration_time;
|
|
|
|
} else {
|
2016-12-06 01:18:14 +00:00
|
|
|
for (auto id : lock_info.txn_ids) {
|
|
|
|
if (txn_id == id) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool success = txn_db_impl_->TryStealingExpiredTransactionLocks(id);
|
|
|
|
if (!success) {
|
|
|
|
expired = false;
|
2020-03-27 23:12:40 +00:00
|
|
|
*expire_time = 0;
|
2016-12-06 01:18:14 +00:00
|
|
|
break;
|
|
|
|
}
|
2016-02-02 01:07:05 +00:00
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
return expired;
|
|
|
|
}
|
|
|
|
|
2017-08-07 23:07:40 +00:00
|
|
|
Status TransactionLockMgr::TryLock(PessimisticTransaction* txn,
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
uint32_t column_family_id,
|
2016-12-06 01:18:14 +00:00
|
|
|
const std::string& key, Env* env,
|
|
|
|
bool exclusive) {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// Lookup lock map for this column family id
|
|
|
|
std::shared_ptr<LockMap> lock_map_ptr = GetLockMap(column_family_id);
|
|
|
|
LockMap* lock_map = lock_map_ptr.get();
|
|
|
|
if (lock_map == nullptr) {
|
|
|
|
char msg[255];
|
|
|
|
snprintf(msg, sizeof(msg), "Column family id not found: %" PRIu32,
|
|
|
|
column_family_id);
|
|
|
|
|
|
|
|
return Status::InvalidArgument(msg);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Need to lock the mutex for the stripe that this key hashes to
|
|
|
|
size_t stripe_num = lock_map->GetStripe(key);
|
|
|
|
assert(lock_map->lock_map_stripes_.size() > stripe_num);
|
2015-09-08 19:36:48 +00:00
|
|
|
LockMapStripe* stripe = lock_map->lock_map_stripes_.at(stripe_num);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
LockInfo lock_info(txn->GetID(), txn->GetExpirationTime(), exclusive);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
int64_t timeout = txn->GetLockTimeout();
|
|
|
|
|
2016-09-28 00:43:06 +00:00
|
|
|
return AcquireWithTimeout(txn, lock_map, stripe, column_family_id, key, env,
|
2019-04-10 22:54:53 +00:00
|
|
|
timeout, std::move(lock_info));
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Helper function for TryLock().
|
2016-09-28 00:43:06 +00:00
|
|
|
Status TransactionLockMgr::AcquireWithTimeout(
|
2017-08-07 23:07:40 +00:00
|
|
|
PessimisticTransaction* txn, LockMap* lock_map, LockMapStripe* stripe,
|
2016-09-28 00:43:06 +00:00
|
|
|
uint32_t column_family_id, const std::string& key, Env* env,
|
2019-04-10 22:54:53 +00:00
|
|
|
int64_t timeout, LockInfo&& lock_info) {
|
2015-09-08 19:36:48 +00:00
|
|
|
Status result;
|
|
|
|
uint64_t end_time = 0;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
if (timeout > 0) {
|
2018-02-16 00:43:23 +00:00
|
|
|
uint64_t start_time = env->NowMicros();
|
2015-09-08 19:36:48 +00:00
|
|
|
end_time = start_time + timeout;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
if (timeout < 0) {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// If timeout is negative, we wait indefinitely to acquire the lock
|
2015-09-08 19:36:48 +00:00
|
|
|
result = stripe->stripe_mutex->Lock();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
} else {
|
2015-09-08 19:36:48 +00:00
|
|
|
result = stripe->stripe_mutex->TryLockFor(timeout);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
if (!result.ok()) {
|
|
|
|
// failed to acquire mutex
|
|
|
|
return result;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
// Acquire lock if we are able to
|
2015-09-08 19:36:48 +00:00
|
|
|
uint64_t expire_time_hint = 0;
|
2016-12-06 01:18:14 +00:00
|
|
|
autovector<TransactionID> wait_ids;
|
2019-04-10 22:54:53 +00:00
|
|
|
result = AcquireLocked(lock_map, stripe, key, env, std::move(lock_info),
|
2016-12-06 01:18:14 +00:00
|
|
|
&expire_time_hint, &wait_ids);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2015-07-31 03:31:36 +00:00
|
|
|
if (!result.ok() && timeout != 0) {
|
2017-11-06 18:52:17 +00:00
|
|
|
PERF_TIMER_GUARD(key_lock_wait_time);
|
|
|
|
PERF_COUNTER_ADD(key_lock_wait_count, 1);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// If we weren't able to acquire the lock, we will keep retrying as long
|
2015-09-08 19:36:48 +00:00
|
|
|
// as the timeout allows.
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
bool timed_out = false;
|
|
|
|
do {
|
2015-09-08 19:36:48 +00:00
|
|
|
// Decide how long to wait
|
|
|
|
int64_t cv_end_time = -1;
|
2020-03-27 22:59:21 +00:00
|
|
|
if (expire_time_hint > 0 && end_time > 0) {
|
|
|
|
cv_end_time = std::min(expire_time_hint, end_time);
|
|
|
|
} else if (expire_time_hint > 0) {
|
2015-09-08 19:36:48 +00:00
|
|
|
cv_end_time = expire_time_hint;
|
2020-03-27 22:59:21 +00:00
|
|
|
} else if (end_time > 0) {
|
2015-09-08 19:36:48 +00:00
|
|
|
cv_end_time = end_time;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
assert(result.IsBusy() || wait_ids.size() != 0);
|
2016-09-28 08:23:33 +00:00
|
|
|
|
|
|
|
// We are dependent on a transaction to finish, so perform deadlock
|
|
|
|
// detection.
|
2016-12-06 01:18:14 +00:00
|
|
|
if (wait_ids.size() != 0) {
|
2016-09-28 08:23:33 +00:00
|
|
|
if (txn->IsDeadlockDetect()) {
|
2017-08-18 01:49:30 +00:00
|
|
|
if (IncrementWaiters(txn, wait_ids, key, column_family_id,
|
2018-06-27 19:22:17 +00:00
|
|
|
lock_info.exclusive, env)) {
|
2016-09-28 08:23:33 +00:00
|
|
|
result = Status::Busy(Status::SubCode::kDeadlock);
|
|
|
|
stripe->stripe_mutex->UnLock();
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
}
|
2016-12-06 01:18:14 +00:00
|
|
|
txn->SetWaitingTxn(wait_ids, column_family_id, &key);
|
2016-09-28 08:23:33 +00:00
|
|
|
}
|
|
|
|
|
2016-09-28 00:43:06 +00:00
|
|
|
TEST_SYNC_POINT("TransactionLockMgr::AcquireWithTimeout:WaitingTxn");
|
2015-09-08 19:36:48 +00:00
|
|
|
if (cv_end_time < 0) {
|
|
|
|
// Wait indefinitely
|
|
|
|
result = stripe->stripe_cv->Wait(stripe->stripe_mutex);
|
|
|
|
} else {
|
|
|
|
uint64_t now = env->NowMicros();
|
|
|
|
if (static_cast<uint64_t>(cv_end_time) > now) {
|
|
|
|
result = stripe->stripe_cv->WaitFor(stripe->stripe_mutex,
|
|
|
|
cv_end_time - now);
|
|
|
|
}
|
|
|
|
}
|
2016-09-28 08:23:33 +00:00
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
if (wait_ids.size() != 0) {
|
|
|
|
txn->ClearWaitingTxn();
|
2016-09-28 08:23:33 +00:00
|
|
|
if (txn->IsDeadlockDetect()) {
|
2016-12-06 01:18:14 +00:00
|
|
|
DecrementWaiters(txn, wait_ids);
|
2016-09-28 08:23:33 +00:00
|
|
|
}
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
if (result.IsTimedOut()) {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
timed_out = true;
|
|
|
|
// Even though we timed out, we will still make one more attempt to
|
|
|
|
// acquire lock below (it is possible the lock expired and we
|
|
|
|
// were never signaled).
|
|
|
|
}
|
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
if (result.ok() || result.IsTimedOut()) {
|
2019-04-10 22:54:53 +00:00
|
|
|
result = AcquireLocked(lock_map, stripe, key, env, std::move(lock_info),
|
2016-12-06 01:18:14 +00:00
|
|
|
&expire_time_hint, &wait_ids);
|
2015-09-08 19:36:48 +00:00
|
|
|
}
|
2015-07-31 03:31:36 +00:00
|
|
|
} while (!result.ok() && !timed_out);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
stripe->stripe_mutex->UnLock();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
void TransactionLockMgr::DecrementWaiters(
|
2017-08-16 23:49:11 +00:00
|
|
|
const PessimisticTransaction* txn,
|
|
|
|
const autovector<TransactionID>& wait_ids) {
|
2016-09-28 08:23:33 +00:00
|
|
|
std::lock_guard<std::mutex> lock(wait_txn_map_mutex_);
|
2016-12-06 01:18:14 +00:00
|
|
|
DecrementWaitersImpl(txn, wait_ids);
|
2016-09-28 08:23:33 +00:00
|
|
|
}
|
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
void TransactionLockMgr::DecrementWaitersImpl(
|
2017-08-16 23:49:11 +00:00
|
|
|
const PessimisticTransaction* txn,
|
|
|
|
const autovector<TransactionID>& wait_ids) {
|
2016-09-28 08:23:33 +00:00
|
|
|
auto id = txn->GetID();
|
2016-11-19 19:34:26 +00:00
|
|
|
assert(wait_txn_map_.Contains(id));
|
|
|
|
wait_txn_map_.Delete(id);
|
2016-09-28 08:23:33 +00:00
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
for (auto wait_id : wait_ids) {
|
|
|
|
rev_wait_txn_map_.Get(wait_id)--;
|
|
|
|
if (rev_wait_txn_map_.Get(wait_id) == 0) {
|
|
|
|
rev_wait_txn_map_.Delete(wait_id);
|
|
|
|
}
|
2016-09-28 08:23:33 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
bool TransactionLockMgr::IncrementWaiters(
|
2017-08-16 23:49:11 +00:00
|
|
|
const PessimisticTransaction* txn,
|
2017-08-18 01:49:30 +00:00
|
|
|
const autovector<TransactionID>& wait_ids, const std::string& key,
|
2018-06-27 19:22:17 +00:00
|
|
|
const uint32_t& cf_id, const bool& exclusive, Env* const env) {
|
2016-09-28 08:23:33 +00:00
|
|
|
auto id = txn->GetID();
|
2018-09-06 01:07:53 +00:00
|
|
|
std::vector<int> queue_parents(static_cast<size_t>(txn->GetDeadlockDetectDepth()));
|
|
|
|
std::vector<TransactionID> queue_values(static_cast<size_t>(txn->GetDeadlockDetectDepth()));
|
2016-09-28 08:23:33 +00:00
|
|
|
std::lock_guard<std::mutex> lock(wait_txn_map_mutex_);
|
2016-11-19 19:34:26 +00:00
|
|
|
assert(!wait_txn_map_.Contains(id));
|
2017-08-18 01:49:30 +00:00
|
|
|
|
2019-04-15 17:30:48 +00:00
|
|
|
wait_txn_map_.Insert(id, {wait_ids, cf_id, exclusive, key});
|
2016-11-19 19:34:26 +00:00
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
for (auto wait_id : wait_ids) {
|
|
|
|
if (rev_wait_txn_map_.Contains(wait_id)) {
|
|
|
|
rev_wait_txn_map_.Get(wait_id)++;
|
|
|
|
} else {
|
|
|
|
rev_wait_txn_map_.Insert(wait_id, 1);
|
|
|
|
}
|
2016-11-19 19:34:26 +00:00
|
|
|
}
|
2016-09-28 08:23:33 +00:00
|
|
|
|
|
|
|
// No deadlock if nobody is waiting on self.
|
2016-11-19 19:34:26 +00:00
|
|
|
if (!rev_wait_txn_map_.Contains(id)) {
|
2016-09-28 08:23:33 +00:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-12-06 01:18:14 +00:00
|
|
|
const auto* next_ids = &wait_ids;
|
2017-08-18 01:49:30 +00:00
|
|
|
int parent = -1;
|
2018-06-27 19:22:17 +00:00
|
|
|
int64_t deadlock_time = 0;
|
2016-12-06 01:18:14 +00:00
|
|
|
for (int tail = 0, head = 0; head < txn->GetDeadlockDetectDepth(); head++) {
|
2016-12-06 22:29:26 +00:00
|
|
|
int i = 0;
|
2016-12-06 01:18:14 +00:00
|
|
|
if (next_ids) {
|
2016-12-06 22:29:26 +00:00
|
|
|
for (; i < static_cast<int>(next_ids->size()) &&
|
|
|
|
tail + i < txn->GetDeadlockDetectDepth();
|
2016-12-06 01:18:14 +00:00
|
|
|
i++) {
|
2017-08-18 01:49:30 +00:00
|
|
|
queue_values[tail + i] = (*next_ids)[i];
|
|
|
|
queue_parents[tail + i] = parent;
|
2016-12-06 01:18:14 +00:00
|
|
|
}
|
|
|
|
tail += i;
|
|
|
|
}
|
|
|
|
|
|
|
|
// No more items in the list, meaning no deadlock.
|
|
|
|
if (tail == head) {
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2017-08-18 01:49:30 +00:00
|
|
|
auto next = queue_values[head];
|
2016-09-28 08:23:33 +00:00
|
|
|
if (next == id) {
|
2017-08-18 01:49:30 +00:00
|
|
|
std::vector<DeadlockInfo> path;
|
|
|
|
while (head != -1) {
|
|
|
|
assert(wait_txn_map_.Contains(queue_values[head]));
|
|
|
|
|
|
|
|
auto extracted_info = wait_txn_map_.Get(queue_values[head]);
|
|
|
|
path.push_back({queue_values[head], extracted_info.m_cf_id,
|
2018-10-23 17:33:55 +00:00
|
|
|
extracted_info.m_exclusive,
|
|
|
|
extracted_info.m_waiting_key});
|
2017-08-18 01:49:30 +00:00
|
|
|
head = queue_parents[head];
|
|
|
|
}
|
2018-06-27 19:22:17 +00:00
|
|
|
env->GetCurrentTime(&deadlock_time);
|
2017-08-18 01:49:30 +00:00
|
|
|
std::reverse(path.begin(), path.end());
|
2018-06-27 19:22:17 +00:00
|
|
|
dlock_buffer_.AddNewPath(DeadlockPath(path, deadlock_time));
|
|
|
|
deadlock_time = 0;
|
2016-12-06 01:18:14 +00:00
|
|
|
DecrementWaitersImpl(txn, wait_ids);
|
2016-09-28 08:23:33 +00:00
|
|
|
return true;
|
2016-11-19 19:34:26 +00:00
|
|
|
} else if (!wait_txn_map_.Contains(next)) {
|
2016-12-06 01:18:14 +00:00
|
|
|
next_ids = nullptr;
|
|
|
|
continue;
|
2016-09-28 08:23:33 +00:00
|
|
|
} else {
|
2017-08-18 01:49:30 +00:00
|
|
|
parent = head;
|
|
|
|
next_ids = &(wait_txn_map_.Get(next).m_neighbors);
|
2016-09-28 08:23:33 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Wait cycle too big, just assume deadlock.
|
2018-06-27 19:22:17 +00:00
|
|
|
env->GetCurrentTime(&deadlock_time);
|
|
|
|
dlock_buffer_.AddNewPath(DeadlockPath(deadlock_time, true));
|
2016-12-06 01:18:14 +00:00
|
|
|
DecrementWaitersImpl(txn, wait_ids);
|
2016-09-28 08:23:33 +00:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// Try to lock this key after we have acquired the mutex.
|
2015-09-08 19:36:48 +00:00
|
|
|
// Sets *expire_time to the expiration time in microseconds
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// or 0 if no expiration.
|
|
|
|
// REQUIRED: Stripe mutex must be held.
|
|
|
|
Status TransactionLockMgr::AcquireLocked(LockMap* lock_map,
|
|
|
|
LockMapStripe* stripe,
|
|
|
|
const std::string& key, Env* env,
|
2019-04-10 22:54:53 +00:00
|
|
|
LockInfo&& txn_lock_info,
|
2016-09-28 00:43:06 +00:00
|
|
|
uint64_t* expire_time,
|
2016-12-06 01:18:14 +00:00
|
|
|
autovector<TransactionID>* txn_ids) {
|
|
|
|
assert(txn_lock_info.txn_ids.size() == 1);
|
|
|
|
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
Status result;
|
|
|
|
// Check if this key is already locked
|
2018-02-21 01:34:44 +00:00
|
|
|
auto stripe_iter = stripe->keys.find(key);
|
|
|
|
if (stripe_iter != stripe->keys.end()) {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// Lock already held
|
2018-02-21 01:34:44 +00:00
|
|
|
LockInfo& lock_info = stripe_iter->second;
|
2016-12-06 01:18:14 +00:00
|
|
|
assert(lock_info.txn_ids.size() == 1 || !lock_info.exclusive);
|
|
|
|
|
|
|
|
if (lock_info.exclusive || txn_lock_info.exclusive) {
|
|
|
|
if (lock_info.txn_ids.size() == 1 &&
|
|
|
|
lock_info.txn_ids[0] == txn_lock_info.txn_ids[0]) {
|
|
|
|
// The list contains one txn and we're it, so just take it.
|
|
|
|
lock_info.exclusive = txn_lock_info.exclusive;
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
lock_info.expiration_time = txn_lock_info.expiration_time;
|
|
|
|
} else {
|
2016-12-06 01:18:14 +00:00
|
|
|
// Check if it's expired. Skips over txn_lock_info.txn_ids[0] in case
|
|
|
|
// it's there for a shared lock with multiple holders which was not
|
|
|
|
// caught in the first case.
|
|
|
|
if (IsLockExpired(txn_lock_info.txn_ids[0], lock_info, env,
|
|
|
|
expire_time)) {
|
|
|
|
// lock is expired, can steal it
|
|
|
|
lock_info.txn_ids = txn_lock_info.txn_ids;
|
|
|
|
lock_info.exclusive = txn_lock_info.exclusive;
|
|
|
|
lock_info.expiration_time = txn_lock_info.expiration_time;
|
|
|
|
// lock_cnt does not change
|
|
|
|
} else {
|
|
|
|
result = Status::TimedOut(Status::SubCode::kLockTimeout);
|
|
|
|
*txn_ids = lock_info.txn_ids;
|
|
|
|
}
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
2016-12-06 01:18:14 +00:00
|
|
|
} else {
|
|
|
|
// We are requesting shared access to a shared lock, so just grant it.
|
|
|
|
lock_info.txn_ids.push_back(txn_lock_info.txn_ids[0]);
|
|
|
|
// Using std::max means that expiration time never goes down even when
|
|
|
|
// a transaction is removed from the list. The correct solution would be
|
|
|
|
// to track expiry for every transaction, but this would also work for
|
|
|
|
// now.
|
|
|
|
lock_info.expiration_time =
|
|
|
|
std::max(lock_info.expiration_time, txn_lock_info.expiration_time);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
} else { // Lock not held.
|
|
|
|
// Check lock limit
|
|
|
|
if (max_num_locks_ > 0 &&
|
|
|
|
lock_map->lock_cnt.load(std::memory_order_acquire) >= max_num_locks_) {
|
2015-08-14 23:45:05 +00:00
|
|
|
result = Status::Busy(Status::SubCode::kLockLimit);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
} else {
|
|
|
|
// acquire lock
|
2019-04-10 22:54:53 +00:00
|
|
|
stripe->keys.emplace(key, std::move(txn_lock_info));
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
// Maintain lock count if there is a limit on the number of locks
|
|
|
|
if (max_num_locks_) {
|
|
|
|
lock_map->lock_cnt++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2017-08-07 23:07:40 +00:00
|
|
|
void TransactionLockMgr::UnLockKey(const PessimisticTransaction* txn,
|
2016-12-06 01:18:14 +00:00
|
|
|
const std::string& key,
|
|
|
|
LockMapStripe* stripe, LockMap* lock_map,
|
|
|
|
Env* env) {
|
2018-04-13 00:55:14 +00:00
|
|
|
#ifdef NDEBUG
|
|
|
|
(void)env;
|
|
|
|
#endif
|
2016-12-06 01:18:14 +00:00
|
|
|
TransactionID txn_id = txn->GetID();
|
|
|
|
|
|
|
|
auto stripe_iter = stripe->keys.find(key);
|
|
|
|
if (stripe_iter != stripe->keys.end()) {
|
|
|
|
auto& txns = stripe_iter->second.txn_ids;
|
|
|
|
auto txn_it = std::find(txns.begin(), txns.end(), txn_id);
|
|
|
|
// Found the key we locked. unlock it.
|
|
|
|
if (txn_it != txns.end()) {
|
|
|
|
if (txns.size() == 1) {
|
|
|
|
stripe->keys.erase(stripe_iter);
|
|
|
|
} else {
|
|
|
|
auto last_it = txns.end() - 1;
|
|
|
|
if (txn_it != last_it) {
|
|
|
|
*txn_it = *last_it;
|
|
|
|
}
|
|
|
|
txns.pop_back();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (max_num_locks_ > 0) {
|
|
|
|
// Maintain lock count if there is a limit on the number of locks.
|
|
|
|
assert(lock_map->lock_cnt.load(std::memory_order_relaxed) > 0);
|
|
|
|
lock_map->lock_cnt--;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
// This key is either not locked or locked by someone else. This should
|
|
|
|
// only happen if the unlocking transaction has expired.
|
|
|
|
assert(txn->GetExpirationTime() > 0 &&
|
|
|
|
txn->GetExpirationTime() < env->NowMicros());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-08-16 23:49:11 +00:00
|
|
|
void TransactionLockMgr::UnLock(PessimisticTransaction* txn,
|
|
|
|
uint32_t column_family_id,
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
const std::string& key, Env* env) {
|
|
|
|
std::shared_ptr<LockMap> lock_map_ptr = GetLockMap(column_family_id);
|
|
|
|
LockMap* lock_map = lock_map_ptr.get();
|
|
|
|
if (lock_map == nullptr) {
|
|
|
|
// Column Family must have been dropped.
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Lock the mutex for the stripe that this key hashes to
|
|
|
|
size_t stripe_num = lock_map->GetStripe(key);
|
|
|
|
assert(lock_map->lock_map_stripes_.size() > stripe_num);
|
2015-09-08 19:36:48 +00:00
|
|
|
LockMapStripe* stripe = lock_map->lock_map_stripes_.at(stripe_num);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
2015-09-08 19:36:48 +00:00
|
|
|
stripe->stripe_mutex->Lock();
|
2016-12-06 01:18:14 +00:00
|
|
|
UnLockKey(txn, key, stripe, lock_map, env);
|
2015-09-08 19:36:48 +00:00
|
|
|
stripe->stripe_mutex->UnLock();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
// Signal waiting threads to retry locking
|
2015-09-08 19:36:48 +00:00
|
|
|
stripe->stripe_cv->NotifyAll();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
|
2017-08-07 23:07:40 +00:00
|
|
|
void TransactionLockMgr::UnLock(const PessimisticTransaction* txn,
|
Replace tracked_keys with a new LockTracker interface in TransactionDB (#7013)
Summary:
We're going to support more locking protocols such as range lock in transaction.
However, in current design, `TransactionBase` has a member `tracked_keys` which assumes that point lock (lock a single key) is used, and is used in snapshot checking (isolation protocol). When using range lock, we may use read committed instead of snapshot checking as the isolation protocol.
The most significant usage scenarios of `tracked_keys` are:
1. pessimistic transaction uses it to track the locked keys, and unlock these keys when commit or rollback.
2. optimistic transaction does not lock keys upfront, it only tracks the lock intentions in tracked_keys, and do write conflict checking when commit.
3. each `SavePoint` tracks the keys that are locked since the `SavePoint`, `RollbackToSavePoint` or `PopSavePoint` relies on both the tracked keys in `SavePoint`s and `tracked_keys`.
Based on these scenarios, if we can abstract out a `LockTracker` interface to hold a set of tracked locks (can be keys or key ranges), and have methods that can be composed together to implement the scenarios, then `tracked_keys` can be an internal data structure of one implementation of `LockTracker`. See `utilities/transactions/lock/lock_tracker.h` for the detailed interface design, and `utilities/transactions/lock/point_lock_tracker.cc` for the implementation.
In the future, a `RangeLockTracker` can be implemented to track range locks without affecting other components.
After this PR, a clean interface for lock manager should be possible, and then ideally, we can have pluggable locking protocols.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7013
Test Plan: Run `transaction_test` and `optimistic_transaction_test`.
Reviewed By: ajkr
Differential Revision: D22163706
Pulled By: cheng-chang
fbshipit-source-id: f2860577b5334e31dd2994f5bc6d7c40d502b1b4
2020-08-06 19:36:48 +00:00
|
|
|
const LockTracker& tracker, Env* env) {
|
|
|
|
std::unique_ptr<LockTracker::ColumnFamilyIterator> cf_it(
|
|
|
|
tracker.GetColumnFamilyIterator());
|
|
|
|
assert(cf_it != nullptr);
|
|
|
|
while (cf_it->HasNext()) {
|
|
|
|
ColumnFamilyId cf = cf_it->Next();
|
|
|
|
std::shared_ptr<LockMap> lock_map_ptr = GetLockMap(cf);
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
LockMap* lock_map = lock_map_ptr.get();
|
Replace tracked_keys with a new LockTracker interface in TransactionDB (#7013)
Summary:
We're going to support more locking protocols such as range lock in transaction.
However, in current design, `TransactionBase` has a member `tracked_keys` which assumes that point lock (lock a single key) is used, and is used in snapshot checking (isolation protocol). When using range lock, we may use read committed instead of snapshot checking as the isolation protocol.
The most significant usage scenarios of `tracked_keys` are:
1. pessimistic transaction uses it to track the locked keys, and unlock these keys when commit or rollback.
2. optimistic transaction does not lock keys upfront, it only tracks the lock intentions in tracked_keys, and do write conflict checking when commit.
3. each `SavePoint` tracks the keys that are locked since the `SavePoint`, `RollbackToSavePoint` or `PopSavePoint` relies on both the tracked keys in `SavePoint`s and `tracked_keys`.
Based on these scenarios, if we can abstract out a `LockTracker` interface to hold a set of tracked locks (can be keys or key ranges), and have methods that can be composed together to implement the scenarios, then `tracked_keys` can be an internal data structure of one implementation of `LockTracker`. See `utilities/transactions/lock/lock_tracker.h` for the detailed interface design, and `utilities/transactions/lock/point_lock_tracker.cc` for the implementation.
In the future, a `RangeLockTracker` can be implemented to track range locks without affecting other components.
After this PR, a clean interface for lock manager should be possible, and then ideally, we can have pluggable locking protocols.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7013
Test Plan: Run `transaction_test` and `optimistic_transaction_test`.
Reviewed By: ajkr
Differential Revision: D22163706
Pulled By: cheng-chang
fbshipit-source-id: f2860577b5334e31dd2994f5bc6d7c40d502b1b4
2020-08-06 19:36:48 +00:00
|
|
|
if (!lock_map) {
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
// Column Family must have been dropped.
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Bucket keys by lock_map_ stripe
|
|
|
|
std::unordered_map<size_t, std::vector<const std::string*>> keys_by_stripe(
|
Replace tracked_keys with a new LockTracker interface in TransactionDB (#7013)
Summary:
We're going to support more locking protocols such as range lock in transaction.
However, in current design, `TransactionBase` has a member `tracked_keys` which assumes that point lock (lock a single key) is used, and is used in snapshot checking (isolation protocol). When using range lock, we may use read committed instead of snapshot checking as the isolation protocol.
The most significant usage scenarios of `tracked_keys` are:
1. pessimistic transaction uses it to track the locked keys, and unlock these keys when commit or rollback.
2. optimistic transaction does not lock keys upfront, it only tracks the lock intentions in tracked_keys, and do write conflict checking when commit.
3. each `SavePoint` tracks the keys that are locked since the `SavePoint`, `RollbackToSavePoint` or `PopSavePoint` relies on both the tracked keys in `SavePoint`s and `tracked_keys`.
Based on these scenarios, if we can abstract out a `LockTracker` interface to hold a set of tracked locks (can be keys or key ranges), and have methods that can be composed together to implement the scenarios, then `tracked_keys` can be an internal data structure of one implementation of `LockTracker`. See `utilities/transactions/lock/lock_tracker.h` for the detailed interface design, and `utilities/transactions/lock/point_lock_tracker.cc` for the implementation.
In the future, a `RangeLockTracker` can be implemented to track range locks without affecting other components.
After this PR, a clean interface for lock manager should be possible, and then ideally, we can have pluggable locking protocols.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/7013
Test Plan: Run `transaction_test` and `optimistic_transaction_test`.
Reviewed By: ajkr
Differential Revision: D22163706
Pulled By: cheng-chang
fbshipit-source-id: f2860577b5334e31dd2994f5bc6d7c40d502b1b4
2020-08-06 19:36:48 +00:00
|
|
|
lock_map->num_stripes_);
|
|
|
|
std::unique_ptr<LockTracker::KeyIterator> key_it(
|
|
|
|
tracker.GetKeyIterator(cf));
|
|
|
|
assert(key_it != nullptr);
|
|
|
|
while (key_it->HasNext()) {
|
|
|
|
const std::string& key = key_it->Next();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
size_t stripe_num = lock_map->GetStripe(key);
|
|
|
|
keys_by_stripe[stripe_num].push_back(&key);
|
|
|
|
}
|
|
|
|
|
|
|
|
// For each stripe, grab the stripe mutex and unlock all keys in this stripe
|
|
|
|
for (auto& stripe_iter : keys_by_stripe) {
|
|
|
|
size_t stripe_num = stripe_iter.first;
|
|
|
|
auto& stripe_keys = stripe_iter.second;
|
|
|
|
|
|
|
|
assert(lock_map->lock_map_stripes_.size() > stripe_num);
|
2015-09-08 19:36:48 +00:00
|
|
|
LockMapStripe* stripe = lock_map->lock_map_stripes_.at(stripe_num);
|
|
|
|
|
|
|
|
stripe->stripe_mutex->Lock();
|
|
|
|
|
|
|
|
for (const std::string* key : stripe_keys) {
|
2016-12-06 01:18:14 +00:00
|
|
|
UnLockKey(txn, *key, stripe, lock_map, env);
|
2015-09-08 19:36:48 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
stripe->stripe_mutex->UnLock();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
|
|
|
|
// Signal waiting threads to retry locking
|
2015-09-08 19:36:48 +00:00
|
|
|
stripe->stripe_cv->NotifyAll();
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-09-28 00:43:06 +00:00
|
|
|
TransactionLockMgr::LockStatusData TransactionLockMgr::GetLockStatusData() {
|
|
|
|
LockStatusData data;
|
|
|
|
// Lock order here is important. The correct order is lock_map_mutex_, then
|
|
|
|
// for every column family ID in ascending order lock every stripe in
|
|
|
|
// ascending order.
|
|
|
|
InstrumentedMutexLock l(&lock_map_mutex_);
|
|
|
|
|
2016-09-28 08:23:33 +00:00
|
|
|
std::vector<uint32_t> cf_ids;
|
|
|
|
for (const auto& map : lock_maps_) {
|
|
|
|
cf_ids.push_back(map.first);
|
|
|
|
}
|
|
|
|
std::sort(cf_ids.begin(), cf_ids.end());
|
2016-09-28 00:43:06 +00:00
|
|
|
|
2016-09-28 08:23:33 +00:00
|
|
|
for (auto i : cf_ids) {
|
|
|
|
const auto& stripes = lock_maps_[i]->lock_map_stripes_;
|
2016-09-28 00:43:06 +00:00
|
|
|
// Iterate and lock all stripes in ascending order.
|
2016-09-28 08:23:33 +00:00
|
|
|
for (const auto& j : stripes) {
|
|
|
|
j->stripe_mutex->Lock();
|
|
|
|
for (const auto& it : j->keys) {
|
2016-12-06 01:18:14 +00:00
|
|
|
struct KeyLockInfo info;
|
|
|
|
info.exclusive = it.second.exclusive;
|
|
|
|
info.key = it.first;
|
|
|
|
for (const auto& id : it.second.txn_ids) {
|
|
|
|
info.ids.push_back(id);
|
|
|
|
}
|
|
|
|
data.insert({i, info});
|
2016-09-28 00:43:06 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Unlock everything. Unlocking order is not important.
|
|
|
|
for (auto i : cf_ids) {
|
2016-09-28 08:23:33 +00:00
|
|
|
const auto& stripes = lock_maps_[i]->lock_map_stripes_;
|
|
|
|
for (const auto& j : stripes) {
|
2016-09-28 00:43:06 +00:00
|
|
|
j->stripe_mutex->UnLock();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return data;
|
|
|
|
}
|
2017-08-18 01:49:30 +00:00
|
|
|
std::vector<DeadlockPath> TransactionLockMgr::GetDeadlockInfoBuffer() {
|
|
|
|
return dlock_buffer_.PrepareBuffer();
|
|
|
|
}
|
|
|
|
|
|
|
|
void TransactionLockMgr::Resize(uint32_t target_size) {
|
|
|
|
dlock_buffer_.Resize(target_size);
|
|
|
|
}
|
2016-09-28 00:43:06 +00:00
|
|
|
|
2020-02-20 20:07:53 +00:00
|
|
|
} // namespace ROCKSDB_NAMESPACE
|
Pessimistic Transactions
Summary:
Initial implementation of Pessimistic Transactions. This diff contains the api changes discussed in D38913. This diff is pretty large, so let me know if people would prefer to meet up to discuss it.
MyRocks folks: please take a look at the API in include/rocksdb/utilities/transaction[_db].h and let me know if you have any issues.
Also, you'll notice a couple of TODOs in the implementation of RollbackToSavePoint(). After chatting with Siying, I'm going to send out a separate diff for an alternate implementation of this feature that implements the rollback inside of WriteBatch/WriteBatchWithIndex. We can then decide which route is preferable.
Next, I'm planning on doing some perf testing and then integrating this diff into MongoRocks for further testing.
Test Plan: Unit tests, db_bench parallel testing.
Reviewers: igor, rven, sdong, yhchiang, yoshinorim
Reviewed By: sdong
Subscribers: hermanlee4, maykov, spetrunia, leveldb, dhruba
Differential Revision: https://reviews.facebook.net/D40869
2015-05-26 00:37:33 +00:00
|
|
|
#endif // ROCKSDB_LITE
|