2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
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 02:37:33 +02:00
|
|
|
// This source code is licensed under the BSD-style license found in the
|
|
|
|
// LICENSE file in the root directory of this source tree. An additional grant
|
|
|
|
// of patent rights can be found in the PATENTS file in the same directory.
|
|
|
|
|
|
|
|
#ifndef ROCKSDB_LITE
|
|
|
|
|
|
|
|
#include "utilities/transactions/transaction_impl.h"
|
|
|
|
|
|
|
|
#include <map>
|
|
|
|
#include <set>
|
|
|
|
#include <string>
|
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
#include "db/column_family.h"
|
|
|
|
#include "db/db_impl.h"
|
|
|
|
#include "rocksdb/comparator.h"
|
|
|
|
#include "rocksdb/db.h"
|
2015-08-01 00:18:27 +02:00
|
|
|
#include "rocksdb/snapshot.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 02:37:33 +02:00
|
|
|
#include "rocksdb/status.h"
|
|
|
|
#include "rocksdb/utilities/transaction_db.h"
|
|
|
|
#include "util/string_util.h"
|
2016-02-02 02:07:05 +01:00
|
|
|
#include "util/sync_point.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 02:37:33 +02:00
|
|
|
#include "utilities/transactions/transaction_db_impl.h"
|
|
|
|
#include "utilities/transactions/transaction_util.h"
|
|
|
|
|
|
|
|
namespace rocksdb {
|
|
|
|
|
|
|
|
struct WriteOptions;
|
|
|
|
|
|
|
|
std::atomic<TransactionID> TransactionImpl::txn_id_counter_(1);
|
|
|
|
|
|
|
|
TransactionID TransactionImpl::GenTxnID() {
|
|
|
|
return txn_id_counter_.fetch_add(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
TransactionImpl::TransactionImpl(TransactionDB* txn_db,
|
|
|
|
const WriteOptions& write_options,
|
|
|
|
const TransactionOptions& txn_options)
|
2016-08-11 23:19:33 +02:00
|
|
|
: TransactionBaseImpl(txn_db->GetRootDB(), write_options),
|
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 02:37:33 +02:00
|
|
|
txn_db_impl_(nullptr),
|
2016-02-03 04:19:17 +01:00
|
|
|
txn_id_(0),
|
2016-09-28 02:43:06 +02:00
|
|
|
waiting_txn_id_(0),
|
|
|
|
waiting_cf_id_(0),
|
|
|
|
waiting_key_(nullptr),
|
2016-02-03 04:19:17 +01:00
|
|
|
expiration_time_(0),
|
2016-04-18 20:15:50 +02:00
|
|
|
lock_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 02:37:33 +02:00
|
|
|
txn_db_impl_ = dynamic_cast<TransactionDBImpl*>(txn_db);
|
|
|
|
assert(txn_db_impl_);
|
2016-08-11 23:19:33 +02:00
|
|
|
db_impl_ = dynamic_cast<DBImpl*>(txn_db->GetRootDB());
|
2016-04-18 20:15:50 +02:00
|
|
|
assert(db_impl_);
|
2016-02-03 04:19:17 +01:00
|
|
|
Initialize(txn_options);
|
|
|
|
}
|
|
|
|
|
|
|
|
void TransactionImpl::Initialize(const TransactionOptions& txn_options) {
|
|
|
|
txn_id_ = GenTxnID();
|
|
|
|
|
|
|
|
exec_status_ = STARTED;
|
|
|
|
|
|
|
|
lock_timeout_ = txn_options.lock_timeout * 1000;
|
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 02:37:33 +02:00
|
|
|
if (lock_timeout_ < 0) {
|
|
|
|
// Lock timeout not set, use default
|
2015-09-08 21:36:48 +02:00
|
|
|
lock_timeout_ =
|
|
|
|
txn_db_impl_->GetTxnDBOptions().transaction_lock_timeout * 1000;
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
2016-02-03 04:19:17 +01:00
|
|
|
if (txn_options.expiration >= 0) {
|
|
|
|
expiration_time_ = start_time_ + txn_options.expiration * 1000;
|
|
|
|
} else {
|
|
|
|
expiration_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 02:37:33 +02:00
|
|
|
if (txn_options.set_snapshot) {
|
|
|
|
SetSnapshot();
|
|
|
|
}
|
2016-03-04 00:36:26 +01:00
|
|
|
|
2016-02-02 02:07:05 +01:00
|
|
|
if (expiration_time_ > 0) {
|
|
|
|
txn_db_impl_->InsertExpirableTransaction(txn_id_, this);
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
TransactionImpl::~TransactionImpl() {
|
2015-09-12 03:10:50 +02:00
|
|
|
txn_db_impl_->UnLock(this, &GetTrackedKeys());
|
2016-02-02 02:07:05 +01:00
|
|
|
if (expiration_time_ > 0) {
|
|
|
|
txn_db_impl_->RemoveExpirableTransaction(txn_id_);
|
|
|
|
}
|
2016-04-18 20:15:50 +02:00
|
|
|
if (!name_.empty() && exec_status_ != COMMITED) {
|
|
|
|
txn_db_impl_->UnregisterTransaction(this);
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
2015-08-25 04:13:18 +02:00
|
|
|
void TransactionImpl::Clear() {
|
2015-09-12 03:10:50 +02:00
|
|
|
txn_db_impl_->UnLock(this, &GetTrackedKeys());
|
2015-08-25 04:13:18 +02:00
|
|
|
TransactionBaseImpl::Clear();
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
2016-03-04 00:36:26 +01:00
|
|
|
void TransactionImpl::Reinitialize(TransactionDB* txn_db,
|
|
|
|
const WriteOptions& write_options,
|
2016-02-03 04:19:17 +01:00
|
|
|
const TransactionOptions& txn_options) {
|
2016-04-18 20:15:50 +02:00
|
|
|
if (!name_.empty() && exec_status_ != COMMITED) {
|
|
|
|
txn_db_impl_->UnregisterTransaction(this);
|
|
|
|
}
|
2016-08-11 23:19:33 +02:00
|
|
|
TransactionBaseImpl::Reinitialize(txn_db->GetRootDB(), write_options);
|
2016-02-03 04:19:17 +01:00
|
|
|
Initialize(txn_options);
|
|
|
|
}
|
|
|
|
|
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 02:37:33 +02:00
|
|
|
bool TransactionImpl::IsExpired() const {
|
|
|
|
if (expiration_time_ > 0) {
|
2015-09-08 21:36:48 +02:00
|
|
|
if (db_->GetEnv()->NowMicros() >= 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 02:37:33 +02:00
|
|
|
// Transaction is expired.
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status TransactionImpl::CommitBatch(WriteBatch* batch) {
|
|
|
|
TransactionKeyMap keys_to_unlock;
|
|
|
|
Status s = LockBatch(batch, &keys_to_unlock);
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
bool can_commit = false;
|
|
|
|
|
|
|
|
if (IsExpired()) {
|
|
|
|
s = Status::Expired();
|
|
|
|
} else if (expiration_time_ > 0) {
|
|
|
|
ExecutionStatus expected = STARTED;
|
|
|
|
can_commit = std::atomic_compare_exchange_strong(&exec_status_, &expected,
|
|
|
|
AWAITING_COMMIT);
|
|
|
|
} else if (exec_status_ == STARTED) {
|
|
|
|
// lock stealing is not a concern
|
|
|
|
can_commit = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (can_commit) {
|
|
|
|
exec_status_.store(AWAITING_COMMIT);
|
|
|
|
s = db_->Write(write_options_, batch);
|
|
|
|
if (s.ok()) {
|
|
|
|
exec_status_.store(COMMITED);
|
|
|
|
}
|
|
|
|
} else if (exec_status_ == LOCKS_STOLEN) {
|
|
|
|
s = Status::Expired();
|
|
|
|
} else {
|
|
|
|
s = Status::InvalidArgument("Transaction is not in state for commit.");
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
txn_db_impl_->UnLock(this, &keys_to_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 02:37:33 +02:00
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
Status TransactionImpl::Prepare() {
|
|
|
|
Status s;
|
|
|
|
|
|
|
|
if (name_.empty()) {
|
|
|
|
return Status::InvalidArgument(
|
|
|
|
"Cannot prepare a transaction that has not been named.");
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (IsExpired()) {
|
|
|
|
return Status::Expired();
|
|
|
|
}
|
|
|
|
|
|
|
|
bool can_prepare = false;
|
|
|
|
|
|
|
|
if (expiration_time_ > 0) {
|
|
|
|
// must concern ourselves with expiraton and/or lock stealing
|
|
|
|
// need to compare/exchange bc locks could be stolen under us here
|
|
|
|
ExecutionStatus expected = STARTED;
|
|
|
|
can_prepare = std::atomic_compare_exchange_strong(&exec_status_, &expected,
|
|
|
|
AWAITING_PREPARE);
|
|
|
|
} else if (exec_status_ == STARTED) {
|
|
|
|
// expiration and lock stealing is not possible
|
|
|
|
can_prepare = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (can_prepare) {
|
|
|
|
exec_status_.store(AWAITING_PREPARE);
|
|
|
|
// transaction can't expire after preparation
|
|
|
|
expiration_time_ = 0;
|
|
|
|
WriteOptions write_options = write_options_;
|
|
|
|
write_options.disableWAL = false;
|
|
|
|
WriteBatchInternal::MarkEndPrepare(GetWriteBatch()->GetWriteBatch(), name_);
|
|
|
|
s = db_impl_->WriteImpl(write_options, GetWriteBatch()->GetWriteBatch(),
|
|
|
|
/*callback*/ nullptr, &log_number_, /*log ref*/ 0,
|
|
|
|
/* disable_memtable*/ true);
|
|
|
|
if (s.ok()) {
|
|
|
|
assert(log_number_ != 0);
|
|
|
|
dbimpl_->MarkLogAsContainingPrepSection(log_number_);
|
|
|
|
exec_status_.store(PREPARED);
|
|
|
|
}
|
|
|
|
} else if (exec_status_ == LOCKS_STOLEN) {
|
|
|
|
s = Status::Expired();
|
|
|
|
} else if (exec_status_ == PREPARED) {
|
|
|
|
s = Status::InvalidArgument("Transaction has already been prepared.");
|
|
|
|
} else if (exec_status_ == COMMITED) {
|
|
|
|
s = Status::InvalidArgument("Transaction has already been committed.");
|
|
|
|
} else if (exec_status_ == ROLLEDBACK) {
|
|
|
|
s = Status::InvalidArgument("Transaction has already been rolledback.");
|
|
|
|
} else {
|
|
|
|
s = Status::InvalidArgument("Transaction is not in state for commit.");
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
Status TransactionImpl::Commit() {
|
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 02:37:33 +02:00
|
|
|
Status s;
|
2016-04-18 20:15:50 +02:00
|
|
|
bool commit_single = false;
|
|
|
|
bool commit_prepared = 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 02:37:33 +02:00
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (IsExpired()) {
|
|
|
|
return Status::Expired();
|
|
|
|
}
|
2016-02-02 02:07:05 +01:00
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (expiration_time_ > 0) {
|
|
|
|
// we must atomicaly compare and exchange the state here because at
|
|
|
|
// this state in the transaction it is possible for another thread
|
|
|
|
// to change our state out from under us in the even that we expire and have
|
|
|
|
// our locks stolen. In this case the only valid state is STARTED because
|
|
|
|
// a state of PREPARED would have a cleared expiration_time_.
|
2016-02-02 02:07:05 +01:00
|
|
|
ExecutionStatus expected = STARTED;
|
2016-04-18 20:15:50 +02:00
|
|
|
commit_single = std::atomic_compare_exchange_strong(
|
|
|
|
&exec_status_, &expected, AWAITING_COMMIT);
|
2016-02-02 02:07:05 +01:00
|
|
|
TEST_SYNC_POINT("TransactionTest::ExpirableTransactionDataRace:1");
|
2016-04-18 20:15:50 +02:00
|
|
|
} else if (exec_status_ == PREPARED) {
|
|
|
|
// expiration and lock stealing is not a concern
|
|
|
|
commit_prepared = true;
|
|
|
|
} else if (exec_status_ == STARTED) {
|
|
|
|
// expiration and lock stealing is not a concern
|
|
|
|
commit_single = true;
|
|
|
|
}
|
2016-02-02 02:07:05 +01:00
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (commit_single) {
|
|
|
|
assert(!commit_prepared);
|
|
|
|
if (WriteBatchInternal::Count(GetCommitTimeWriteBatch()) > 0) {
|
|
|
|
s = Status::InvalidArgument(
|
|
|
|
"Commit-time batch contains values that will not be committed.");
|
2016-02-02 02:07:05 +01:00
|
|
|
} else {
|
2016-04-18 20:15:50 +02:00
|
|
|
exec_status_.store(AWAITING_COMMIT);
|
|
|
|
s = db_->Write(write_options_, GetWriteBatch()->GetWriteBatch());
|
|
|
|
Clear();
|
|
|
|
if (s.ok()) {
|
|
|
|
exec_status_.store(COMMITED);
|
|
|
|
}
|
2016-02-02 02:07:05 +01:00
|
|
|
}
|
2016-04-18 20:15:50 +02:00
|
|
|
} else if (commit_prepared) {
|
|
|
|
exec_status_.store(AWAITING_COMMIT);
|
|
|
|
WriteOptions write_options = write_options_;
|
|
|
|
|
|
|
|
// insert prepared batch into Memtable only.
|
|
|
|
// Memtable will ignore BeginPrepare/EndPrepare markers
|
|
|
|
// in non recovery mode and simply insert the values
|
|
|
|
write_options.disableWAL = true;
|
|
|
|
assert(log_number_ > 0);
|
|
|
|
s = db_impl_->WriteImpl(write_options, GetWriteBatch()->GetWriteBatch(),
|
|
|
|
nullptr, nullptr, log_number_);
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
// We take the commit-time batch and append the Commit marker.
|
|
|
|
// We then write this batch to both WAL and Memtable.
|
|
|
|
// The Memtable will ignore the Commit marker in non-recovery mode
|
|
|
|
write_options.disableWAL = false;
|
|
|
|
WriteBatchInternal::MarkCommit(GetCommitTimeWriteBatch(), name_);
|
|
|
|
s = db_impl_->WriteImpl(write_options, GetCommitTimeWriteBatch());
|
|
|
|
if (!s.ok()) {
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
// FindObsoleteFiles must now look to the memtables
|
|
|
|
// to determine what prep logs must be kept around,
|
|
|
|
// not the prep section heap.
|
|
|
|
assert(log_number_ > 0);
|
|
|
|
dbimpl_->MarkLogAsHavingPrepSectionFlushed(log_number_);
|
|
|
|
txn_db_impl_->UnregisterTransaction(this);
|
|
|
|
|
|
|
|
Clear();
|
|
|
|
exec_status_.store(COMMITED);
|
|
|
|
} else if (exec_status_ == LOCKS_STOLEN) {
|
|
|
|
s = Status::Expired();
|
|
|
|
} else if (exec_status_ == COMMITED) {
|
|
|
|
s = Status::InvalidArgument("Transaction has already been committed.");
|
|
|
|
} else if (exec_status_ == ROLLEDBACK) {
|
|
|
|
s = Status::InvalidArgument("Transaction has already been rolledback.");
|
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 02:37:33 +02:00
|
|
|
} else {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::InvalidArgument("Transaction is not in state for commit.");
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
Status TransactionImpl::Rollback() {
|
|
|
|
Status s;
|
|
|
|
if (exec_status_ == PREPARED) {
|
|
|
|
WriteBatch rollback_marker;
|
|
|
|
WriteBatchInternal::MarkRollback(&rollback_marker, name_);
|
|
|
|
exec_status_.store(AWAITING_ROLLBACK);
|
|
|
|
s = db_impl_->WriteImpl(write_options_, &rollback_marker);
|
|
|
|
if (s.ok()) {
|
|
|
|
// we do not need to keep our prepared section around
|
|
|
|
assert(log_number_ > 0);
|
|
|
|
dbimpl_->MarkLogAsHavingPrepSectionFlushed(log_number_);
|
|
|
|
Clear();
|
|
|
|
exec_status_.store(ROLLEDBACK);
|
|
|
|
}
|
|
|
|
} else if (exec_status_ == STARTED) {
|
|
|
|
// prepare couldn't have taken place
|
|
|
|
Clear();
|
|
|
|
} else if (exec_status_ == COMMITED) {
|
|
|
|
s = Status::InvalidArgument("This transaction has already been committed.");
|
|
|
|
} else {
|
|
|
|
s = Status::InvalidArgument(
|
|
|
|
"Two phase transaction is not in state for rollback.");
|
|
|
|
}
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
|
2015-09-12 03:10:50 +02:00
|
|
|
Status TransactionImpl::RollbackToSavePoint() {
|
2016-04-18 20:15:50 +02:00
|
|
|
if (exec_status_ != STARTED) {
|
|
|
|
return Status::InvalidArgument("Transaction is beyond state for rollback.");
|
|
|
|
}
|
|
|
|
|
2015-09-12 03:10:50 +02:00
|
|
|
// Unlock any keys locked since last transaction
|
2015-09-15 02:11:52 +02:00
|
|
|
const std::unique_ptr<TransactionKeyMap>& keys =
|
|
|
|
GetTrackedKeysSinceSavePoint();
|
|
|
|
|
2015-09-12 03:10:50 +02:00
|
|
|
if (keys) {
|
2015-09-15 02:11:52 +02:00
|
|
|
txn_db_impl_->UnLock(this, keys.get());
|
2015-09-12 03:10:50 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return TransactionBaseImpl::RollbackToSavePoint();
|
|
|
|
}
|
|
|
|
|
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 02:37:33 +02:00
|
|
|
// Lock all keys in this batch.
|
|
|
|
// On success, caller should unlock keys_to_unlock
|
|
|
|
Status TransactionImpl::LockBatch(WriteBatch* batch,
|
|
|
|
TransactionKeyMap* keys_to_unlock) {
|
|
|
|
class Handler : public WriteBatch::Handler {
|
|
|
|
public:
|
|
|
|
// Sorted map of column_family_id to sorted set of keys.
|
|
|
|
// Since LockBatch() always locks keys in sorted order, it cannot deadlock
|
|
|
|
// with itself. We're not using a comparator here since it doesn't matter
|
|
|
|
// what the sorting is as long as it's consistent.
|
|
|
|
std::map<uint32_t, std::set<std::string>> keys_;
|
|
|
|
|
|
|
|
Handler() {}
|
|
|
|
|
|
|
|
void RecordKey(uint32_t column_family_id, const Slice& key) {
|
|
|
|
std::string key_str = key.ToString();
|
|
|
|
|
|
|
|
auto iter = (keys_)[column_family_id].find(key_str);
|
|
|
|
if (iter == (keys_)[column_family_id].end()) {
|
|
|
|
// key not yet seen, store it.
|
|
|
|
(keys_)[column_family_id].insert({std::move(key_str)});
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
virtual Status PutCF(uint32_t column_family_id, const Slice& key,
|
|
|
|
const Slice& value) override {
|
|
|
|
RecordKey(column_family_id, key);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
virtual Status MergeCF(uint32_t column_family_id, const Slice& key,
|
|
|
|
const Slice& value) override {
|
|
|
|
RecordKey(column_family_id, key);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
virtual Status DeleteCF(uint32_t column_family_id,
|
|
|
|
const Slice& key) override {
|
|
|
|
RecordKey(column_family_id, key);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
// Iterating on this handler will add all keys in this batch into keys
|
|
|
|
Handler handler;
|
|
|
|
batch->Iterate(&handler);
|
|
|
|
|
|
|
|
Status s;
|
|
|
|
|
|
|
|
// Attempt to lock all keys
|
|
|
|
for (const auto& cf_iter : handler.keys_) {
|
|
|
|
uint32_t cfh_id = cf_iter.first;
|
|
|
|
auto& cfh_keys = cf_iter.second;
|
|
|
|
|
|
|
|
for (const auto& key_iter : cfh_keys) {
|
|
|
|
const std::string& key = key_iter;
|
|
|
|
|
|
|
|
s = txn_db_impl_->TryLock(this, cfh_id, key);
|
|
|
|
if (!s.ok()) {
|
|
|
|
break;
|
|
|
|
}
|
2015-09-15 02:11:52 +02:00
|
|
|
TrackKey(keys_to_unlock, cfh_id, std::move(key), kMaxSequenceNumber,
|
|
|
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (!s.ok()) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!s.ok()) {
|
|
|
|
txn_db_impl_->UnLock(this, keys_to_unlock);
|
|
|
|
}
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Attempt to lock this key.
|
|
|
|
// Returns OK if the key has been successfully locked. Non-ok, otherwise.
|
|
|
|
// If check_shapshot is true and this transaction has a snapshot set,
|
|
|
|
// this key will only be locked if there have been no writes to this key since
|
|
|
|
// the snapshot time.
|
|
|
|
Status TransactionImpl::TryLock(ColumnFamilyHandle* column_family,
|
2015-09-15 02:11:52 +02:00
|
|
|
const Slice& key, bool read_only,
|
|
|
|
bool untracked) {
|
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 02:37:33 +02:00
|
|
|
uint32_t cfh_id = GetColumnFamilyID(column_family);
|
|
|
|
std::string key_str = key.ToString();
|
|
|
|
bool previously_locked;
|
|
|
|
Status s;
|
|
|
|
|
2015-09-28 21:12:17 +02:00
|
|
|
// lock this key if this transactions hasn't already locked it
|
|
|
|
SequenceNumber current_seqno = kMaxSequenceNumber;
|
|
|
|
SequenceNumber new_seqno = kMaxSequenceNumber;
|
2015-09-30 02:00:16 +02:00
|
|
|
|
|
|
|
const auto& tracked_keys = GetTrackedKeys();
|
|
|
|
const auto tracked_keys_cf = tracked_keys.find(cfh_id);
|
|
|
|
if (tracked_keys_cf == tracked_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 02:37:33 +02:00
|
|
|
previously_locked = false;
|
2015-09-30 02:00:16 +02:00
|
|
|
} else {
|
|
|
|
auto iter = tracked_keys_cf->second.find(key_str);
|
|
|
|
if (iter == tracked_keys_cf->second.end()) {
|
|
|
|
previously_locked = false;
|
|
|
|
} else {
|
|
|
|
previously_locked = true;
|
2015-09-15 02:11:52 +02:00
|
|
|
current_seqno = iter->second.seq;
|
2015-09-30 02:00:16 +02:00
|
|
|
}
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
|
2015-09-30 02:00:16 +02:00
|
|
|
// lock this key if this transactions hasn't already locked it
|
|
|
|
if (!previously_locked) {
|
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 02:37:33 +02:00
|
|
|
s = txn_db_impl_->TryLock(this, cfh_id, key_str);
|
|
|
|
}
|
|
|
|
|
2015-09-28 21:12:17 +02:00
|
|
|
SetSnapshotIfNeeded();
|
|
|
|
|
|
|
|
// Even though we do not care about doing conflict checking for this write,
|
|
|
|
// we still need to take a lock to make sure we do not cause a conflict with
|
|
|
|
// some other write. However, we do not need to check if there have been
|
|
|
|
// any writes since this transaction's snapshot.
|
|
|
|
// TODO(agiardullo): could optimize by supporting shared txn locks in the
|
|
|
|
// future
|
|
|
|
if (untracked || snapshot_ == nullptr) {
|
|
|
|
// Need to remember the earliest sequence number that we know that this
|
|
|
|
// key has not been modified after. This is useful if this same
|
|
|
|
// transaction
|
|
|
|
// later tries to lock this key again.
|
|
|
|
if (current_seqno == kMaxSequenceNumber) {
|
|
|
|
// Since we haven't checked a snapshot, we only know this key has not
|
|
|
|
// been modified since after we locked it.
|
|
|
|
new_seqno = db_->GetLatestSequenceNumber();
|
|
|
|
} else {
|
|
|
|
new_seqno = current_seqno;
|
|
|
|
}
|
|
|
|
} else {
|
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 02:37:33 +02:00
|
|
|
// If a snapshot is set, we need to make sure the key hasn't been modified
|
|
|
|
// since the snapshot. This must be done after we locked the key.
|
2015-09-28 21:12:17 +02:00
|
|
|
if (s.ok()) {
|
|
|
|
s = ValidateSnapshot(column_family, key, current_seqno, &new_seqno);
|
|
|
|
|
|
|
|
if (!s.ok()) {
|
|
|
|
// Failed to validate key
|
|
|
|
if (!previously_locked) {
|
|
|
|
// Unlock key we just locked
|
|
|
|
txn_db_impl_->UnLock(this, cfh_id, key.ToString());
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-09-12 03:10:50 +02:00
|
|
|
if (s.ok()) {
|
|
|
|
// Let base class know we've conflict checked this key.
|
2015-09-15 02:11:52 +02:00
|
|
|
TrackKey(cfh_id, key_str, new_seqno, read_only);
|
2015-09-12 03:10:50 +02:00
|
|
|
}
|
|
|
|
|
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 02:37:33 +02:00
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Return OK() if this key has not been modified more recently than the
|
|
|
|
// transaction snapshot_.
|
2015-09-28 21:12:17 +02:00
|
|
|
Status TransactionImpl::ValidateSnapshot(ColumnFamilyHandle* column_family,
|
|
|
|
const Slice& key,
|
|
|
|
SequenceNumber prev_seqno,
|
|
|
|
SequenceNumber* new_seqno) {
|
|
|
|
assert(snapshot_);
|
|
|
|
|
2016-01-28 02:11:44 +01:00
|
|
|
SequenceNumber seq = snapshot_->GetSequenceNumber();
|
2015-09-28 21:12:17 +02:00
|
|
|
if (prev_seqno <= seq) {
|
|
|
|
// If the key has been previous validated at a sequence number earlier
|
|
|
|
// than the curent snapshot's sequence number, we already know it has not
|
|
|
|
// been modified.
|
|
|
|
return Status::OK();
|
|
|
|
}
|
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 02:37:33 +02:00
|
|
|
|
2015-09-28 21:12:17 +02:00
|
|
|
*new_seqno = seq;
|
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 02:37:33 +02:00
|
|
|
|
2015-09-28 21:12:17 +02:00
|
|
|
assert(dynamic_cast<DBImpl*>(db_) != nullptr);
|
|
|
|
auto db_impl = reinterpret_cast<DBImpl*>(db_);
|
|
|
|
|
|
|
|
ColumnFamilyHandle* cfh =
|
|
|
|
column_family ? column_family : db_impl->DefaultColumnFamily();
|
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 02:37:33 +02:00
|
|
|
|
2016-01-28 02:11:44 +01:00
|
|
|
return TransactionUtil::CheckKeyForConflicts(db_impl, cfh, key.ToString(),
|
|
|
|
snapshot_->GetSequenceNumber(),
|
|
|
|
false /* cache_only */);
|
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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
2016-02-02 02:07:05 +01:00
|
|
|
bool TransactionImpl::TryStealingLocks() {
|
|
|
|
assert(IsExpired());
|
|
|
|
ExecutionStatus expected = STARTED;
|
|
|
|
return std::atomic_compare_exchange_strong(&exec_status_, &expected,
|
|
|
|
LOCKS_STOLEN);
|
|
|
|
}
|
|
|
|
|
2015-09-15 02:11:52 +02:00
|
|
|
void TransactionImpl::UnlockGetForUpdate(ColumnFamilyHandle* column_family,
|
|
|
|
const Slice& key) {
|
|
|
|
txn_db_impl_->UnLock(this, GetColumnFamilyID(column_family), key.ToString());
|
|
|
|
}
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
Status TransactionImpl::SetName(const TransactionName& name) {
|
|
|
|
Status s;
|
|
|
|
if (exec_status_ == STARTED) {
|
|
|
|
if (name_.length()) {
|
|
|
|
s = Status::InvalidArgument("Transaction has already been named.");
|
|
|
|
} else if (txn_db_impl_->GetTransactionByName(name) != nullptr) {
|
|
|
|
s = Status::InvalidArgument("Transaction name must be unique.");
|
|
|
|
} else if (name.length() < 1 || name.length() > 512) {
|
|
|
|
s = Status::InvalidArgument(
|
|
|
|
"Transaction name length must be between 1 and 512 chars.");
|
|
|
|
} else {
|
|
|
|
name_ = name;
|
|
|
|
txn_db_impl_->RegisterTransaction(this);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
s = Status::InvalidArgument("Transaction is beyond state for naming.");
|
|
|
|
}
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
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 02:37:33 +02:00
|
|
|
} // namespace rocksdb
|
|
|
|
|
|
|
|
#endif // ROCKSDB_LITE
|