2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-16 01:03:42 +02:00
|
|
|
// 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 02:37:33 +02:00
|
|
|
|
|
|
|
#ifndef ROCKSDB_LITE
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
#include "utilities/transactions/pessimistic_transaction.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 <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"
|
2017-07-29 01:23:50 +02:00
|
|
|
#include "util/cast_util.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 "util/string_util.h"
|
2016-02-02 02:07:05 +01:00
|
|
|
#include "util/sync_point.h"
|
2017-08-06 02:17:48 +02:00
|
|
|
#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 02:37:33 +02:00
|
|
|
#include "utilities/transactions/transaction_util.h"
|
|
|
|
|
|
|
|
namespace rocksdb {
|
|
|
|
|
|
|
|
struct WriteOptions;
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
std::atomic<TransactionID> PessimisticTransaction::txn_id_counter_(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 02:37:33 +02:00
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
TransactionID PessimisticTransaction::GenTxnID() {
|
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 txn_id_counter_.fetch_add(1);
|
|
|
|
}
|
|
|
|
|
2017-08-17 01:49:11 +02:00
|
|
|
PessimisticTransaction::PessimisticTransaction(
|
|
|
|
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),
|
2017-08-03 17:46:47 +02:00
|
|
|
expiration_time_(0),
|
2016-02-03 04:19:17 +01:00
|
|
|
txn_id_(0),
|
2016-09-28 02:43:06 +02:00
|
|
|
waiting_cf_id_(0),
|
|
|
|
waiting_key_(nullptr),
|
2016-09-28 10:23:33 +02:00
|
|
|
lock_timeout_(0),
|
|
|
|
deadlock_detect_(false),
|
|
|
|
deadlock_detect_depth_(0) {
|
2017-07-29 01:23:50 +02:00
|
|
|
txn_db_impl_ =
|
2017-08-06 02:17:48 +02:00
|
|
|
static_cast_with_check<PessimisticTransactionDB, TransactionDB>(txn_db);
|
2017-08-03 17:46:47 +02:00
|
|
|
db_impl_ = static_cast_with_check<DBImpl, DB>(db_);
|
2016-02-03 04:19:17 +01:00
|
|
|
Initialize(txn_options);
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
void PessimisticTransaction::Initialize(const TransactionOptions& txn_options) {
|
2016-02-03 04:19:17 +01:00
|
|
|
txn_id_ = GenTxnID();
|
|
|
|
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_ = STARTED;
|
2016-02-03 04:19:17 +01:00
|
|
|
|
2016-09-28 10:23:33 +02:00
|
|
|
deadlock_detect_ = txn_options.deadlock_detect;
|
|
|
|
deadlock_detect_depth_ = txn_options.deadlock_detect_depth;
|
2017-04-11 00:38:34 +02:00
|
|
|
write_batch_.SetMaxBytes(txn_options.max_write_batch_size);
|
2016-09-28 10:23:33 +02:00
|
|
|
|
2016-02-03 04:19:17 +01:00
|
|
|
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);
|
|
|
|
}
|
2017-11-02 01:23:52 +01:00
|
|
|
use_only_the_last_commit_time_batch_for_recovery_ =
|
|
|
|
txn_options.use_only_the_last_commit_time_batch_for_recovery;
|
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
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
PessimisticTransaction::~PessimisticTransaction() {
|
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_);
|
|
|
|
}
|
2016-10-05 22:39:00 +02:00
|
|
|
if (!name_.empty() && txn_state_ != COMMITED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
void PessimisticTransaction::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
|
|
|
}
|
|
|
|
|
2017-08-17 01:49:11 +02:00
|
|
|
void PessimisticTransaction::Reinitialize(
|
|
|
|
TransactionDB* txn_db, const WriteOptions& write_options,
|
|
|
|
const TransactionOptions& txn_options) {
|
2016-10-05 22:39:00 +02:00
|
|
|
if (!name_.empty() && txn_state_ != COMMITED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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);
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
bool PessimisticTransaction::IsExpired() const {
|
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 (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;
|
|
|
|
}
|
|
|
|
|
2017-08-17 01:49:11 +02:00
|
|
|
WriteCommittedTxn::WriteCommittedTxn(TransactionDB* txn_db,
|
|
|
|
const WriteOptions& write_options,
|
|
|
|
const TransactionOptions& txn_options)
|
2017-08-08 01:07:40 +02:00
|
|
|
: PessimisticTransaction(txn_db, write_options, txn_options){};
|
2017-08-03 17:46:47 +02:00
|
|
|
|
2017-09-13 20:56:27 +02:00
|
|
|
Status PessimisticTransaction::CommitBatch(WriteBatch* batch) {
|
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
|
|
|
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) {
|
2016-10-05 22:39:00 +02:00
|
|
|
TransactionState expected = STARTED;
|
|
|
|
can_commit = std::atomic_compare_exchange_strong(&txn_state_, &expected,
|
2016-04-18 20:15:50 +02:00
|
|
|
AWAITING_COMMIT);
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == STARTED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
// lock stealing is not a concern
|
|
|
|
can_commit = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (can_commit) {
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(AWAITING_COMMIT);
|
2017-09-09 00:53:51 +02:00
|
|
|
s = CommitBatchInternal(batch);
|
2016-04-18 20:15:50 +02:00
|
|
|
if (s.ok()) {
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(COMMITED);
|
2016-04-18 20:15:50 +02:00
|
|
|
}
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == LOCKS_STOLEN) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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;
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
Status PessimisticTransaction::Prepare() {
|
2016-04-18 20:15:50 +02:00
|
|
|
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
|
2016-10-05 22:39:00 +02:00
|
|
|
TransactionState expected = STARTED;
|
|
|
|
can_prepare = std::atomic_compare_exchange_strong(&txn_state_, &expected,
|
2016-04-18 20:15:50 +02:00
|
|
|
AWAITING_PREPARE);
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == STARTED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
// expiration and lock stealing is not possible
|
|
|
|
can_prepare = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (can_prepare) {
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(AWAITING_PREPARE);
|
2016-04-18 20:15:50 +02:00
|
|
|
// transaction can't expire after preparation
|
|
|
|
expiration_time_ = 0;
|
2017-08-08 01:07:40 +02:00
|
|
|
s = PrepareInternal();
|
2016-04-18 20:15:50 +02:00
|
|
|
if (s.ok()) {
|
|
|
|
assert(log_number_ != 0);
|
|
|
|
dbimpl_->MarkLogAsContainingPrepSection(log_number_);
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(PREPARED);
|
2016-04-18 20:15:50 +02:00
|
|
|
}
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == LOCKS_STOLEN) {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::Expired();
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == PREPARED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::InvalidArgument("Transaction has already been prepared.");
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == COMMITED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::InvalidArgument("Transaction has already been committed.");
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == ROLLEDBACK) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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;
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
Status WriteCommittedTxn::PrepareInternal() {
|
|
|
|
WriteOptions write_options = write_options_;
|
|
|
|
write_options.disableWAL = false;
|
|
|
|
WriteBatchInternal::MarkEndPrepare(GetWriteBatch()->GetWriteBatch(), name_);
|
|
|
|
Status s =
|
|
|
|
db_impl_->WriteImpl(write_options, GetWriteBatch()->GetWriteBatch(),
|
|
|
|
/*callback*/ nullptr, &log_number_, /*log ref*/ 0,
|
|
|
|
/* disable_memtable*/ true);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
|
|
|
Status PessimisticTransaction::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;
|
2017-08-08 01:07:40 +02:00
|
|
|
bool commit_without_prepare = false;
|
2016-04-18 20:15:50 +02:00
|
|
|
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_.
|
2016-10-05 22:39:00 +02:00
|
|
|
TransactionState expected = STARTED;
|
2017-08-08 01:07:40 +02:00
|
|
|
commit_without_prepare = std::atomic_compare_exchange_strong(
|
|
|
|
&txn_state_, &expected, AWAITING_COMMIT);
|
2016-02-02 02:07:05 +01:00
|
|
|
TEST_SYNC_POINT("TransactionTest::ExpirableTransactionDataRace:1");
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == PREPARED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
// expiration and lock stealing is not a concern
|
|
|
|
commit_prepared = true;
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == STARTED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
// expiration and lock stealing is not a concern
|
2017-08-08 01:07:40 +02:00
|
|
|
commit_without_prepare = true;
|
|
|
|
// TODO(myabandeh): what if the user mistakenly forgets prepare? We should
|
|
|
|
// add an option so that the user explictly express the intention of
|
|
|
|
// skipping the prepare phase.
|
2016-04-18 20:15:50 +02:00
|
|
|
}
|
2016-02-02 02:07:05 +01:00
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
if (commit_without_prepare) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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 {
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(AWAITING_COMMIT);
|
2017-08-08 01:07:40 +02:00
|
|
|
s = CommitWithoutPrepareInternal();
|
2016-04-18 20:15:50 +02:00
|
|
|
Clear();
|
|
|
|
if (s.ok()) {
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(COMMITED);
|
2016-04-18 20:15:50 +02:00
|
|
|
}
|
2016-02-02 02:07:05 +01:00
|
|
|
}
|
2016-04-18 20:15:50 +02:00
|
|
|
} else if (commit_prepared) {
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(AWAITING_COMMIT);
|
2016-04-18 20:15:50 +02:00
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
s = CommitInternal();
|
Add facility to write only a portion of WriteBatch to WAL
Summary:
When constructing a write batch a client may now call MarkWalTerminationPoint() on that batch. No batch operations after this call will be added written to the WAL but will still be inserted into the Memtable. This facility is used to remove one of the three WriteImpl calls in 2PC transactions. This produces a ~1% perf improvement.
```
RocksDB - unoptimized 2pc, sync_binlog=1, disable_2pc=off
INFO 2016-08-31 14:30:38,814 [main]: REQUEST PHASE COMPLETED. 75000000 requests done in 2619 seconds. Requests/second = 28628
RocksDB - optimized 2pc , sync_binlog=1, disable_2pc=off
INFO 2016-08-31 16:26:59,442 [main]: REQUEST PHASE COMPLETED. 75000000 requests done in 2581 seconds. Requests/second = 29054
```
Test Plan: Two unit tests added.
Reviewers: sdong, yiwu, IslamAbdelRahman
Reviewed By: yiwu
Subscribers: hermanlee4, dhruba, andrewkr
Differential Revision: https://reviews.facebook.net/D64599
2016-10-07 20:31:26 +02:00
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (!s.ok()) {
|
2017-06-24 23:06:43 +02:00
|
|
|
ROCKS_LOG_WARN(db_impl_->immutable_db_options().info_log,
|
|
|
|
"Commit write failed");
|
2016-04-18 20:15:50 +02:00
|
|
|
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();
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(COMMITED);
|
|
|
|
} else if (txn_state_ == LOCKS_STOLEN) {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::Expired();
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == COMMITED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::InvalidArgument("Transaction has already been committed.");
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == ROLLEDBACK) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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;
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
Status WriteCommittedTxn::CommitWithoutPrepareInternal() {
|
|
|
|
Status s = db_->Write(write_options_, GetWriteBatch()->GetWriteBatch());
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2017-09-09 00:53:51 +02:00
|
|
|
Status WriteCommittedTxn::CommitBatchInternal(WriteBatch* batch) {
|
|
|
|
Status s = db_->Write(write_options_, batch);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
Status WriteCommittedTxn::CommitInternal() {
|
|
|
|
// We take the commit-time batch and append the Commit marker.
|
|
|
|
// The Memtable will ignore the Commit marker in non-recovery mode
|
|
|
|
WriteBatch* working_batch = GetCommitTimeWriteBatch();
|
|
|
|
WriteBatchInternal::MarkCommit(working_batch, name_);
|
|
|
|
|
|
|
|
// any operations appended to this working_batch will be ignored from WAL
|
|
|
|
working_batch->MarkWalTerminationPoint();
|
|
|
|
|
|
|
|
// insert prepared batch into Memtable only skipping WAL.
|
|
|
|
// Memtable will ignore BeginPrepare/EndPrepare markers
|
|
|
|
// in non recovery mode and simply insert the values
|
|
|
|
WriteBatchInternal::Append(working_batch, GetWriteBatch()->GetWriteBatch());
|
|
|
|
|
|
|
|
auto s = db_impl_->WriteImpl(write_options_, working_batch, nullptr, nullptr,
|
|
|
|
log_number_);
|
|
|
|
return s;
|
|
|
|
}
|
|
|
|
|
2017-10-03 04:46:42 +02:00
|
|
|
Status PessimisticTransaction::Rollback() {
|
2016-04-18 20:15:50 +02:00
|
|
|
Status s;
|
2016-10-05 22:39:00 +02:00
|
|
|
if (txn_state_ == PREPARED) {
|
|
|
|
txn_state_.store(AWAITING_ROLLBACK);
|
2017-10-03 04:46:42 +02:00
|
|
|
|
|
|
|
s = RollbackInternal();
|
|
|
|
|
2016-04-18 20:15:50 +02:00
|
|
|
if (s.ok()) {
|
|
|
|
// we do not need to keep our prepared section around
|
|
|
|
assert(log_number_ > 0);
|
|
|
|
dbimpl_->MarkLogAsHavingPrepSectionFlushed(log_number_);
|
|
|
|
Clear();
|
2016-10-05 22:39:00 +02:00
|
|
|
txn_state_.store(ROLLEDBACK);
|
2016-04-18 20:15:50 +02:00
|
|
|
}
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == STARTED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
// prepare couldn't have taken place
|
|
|
|
Clear();
|
2016-10-05 22:39:00 +02:00
|
|
|
} else if (txn_state_ == COMMITED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
s = Status::InvalidArgument("This transaction has already been committed.");
|
|
|
|
} else {
|
|
|
|
s = Status::InvalidArgument(
|
|
|
|
"Two phase transaction is not in state for rollback.");
|
|
|
|
}
|
|
|
|
|
|
|
|
return s;
|
|
|
|
}
|
2017-10-03 04:46:42 +02:00
|
|
|
|
|
|
|
Status WriteCommittedTxn::RollbackInternal() {
|
|
|
|
WriteBatch rollback_marker;
|
|
|
|
WriteBatchInternal::MarkRollback(&rollback_marker, name_);
|
|
|
|
auto s = db_impl_->WriteImpl(write_options_, &rollback_marker);
|
|
|
|
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
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
Status PessimisticTransaction::RollbackToSavePoint() {
|
2016-10-05 22:39:00 +02:00
|
|
|
if (txn_state_ != STARTED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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
|
2017-08-08 01:07:40 +02:00
|
|
|
Status PessimisticTransaction::LockBatch(WriteBatch* batch,
|
2017-08-17 01:49:11 +02:00
|
|
|
TransactionKeyMap* 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
|
|
|
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,
|
2017-08-08 01:07:40 +02:00
|
|
|
const Slice& /* unused */) override {
|
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
|
|
|
RecordKey(column_family_id, key);
|
|
|
|
return Status::OK();
|
|
|
|
}
|
|
|
|
virtual Status MergeCF(uint32_t column_family_id, const Slice& key,
|
2017-08-08 01:07:40 +02:00
|
|
|
const Slice& /* unused */) override {
|
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
|
|
|
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;
|
|
|
|
|
2016-12-06 02:18:14 +01:00
|
|
|
s = txn_db_impl_->TryLock(this, cfh_id, key, true /* 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 02:37:33 +02:00
|
|
|
if (!s.ok()) {
|
|
|
|
break;
|
|
|
|
}
|
2015-09-15 02:11:52 +02:00
|
|
|
TrackKey(keys_to_unlock, cfh_id, std::move(key), kMaxSequenceNumber,
|
2017-04-11 00:47:20 +02:00
|
|
|
false, true /* 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 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.
|
2017-08-08 01:07:40 +02:00
|
|
|
Status PessimisticTransaction::TryLock(ColumnFamilyHandle* column_family,
|
2017-08-17 01:49:11 +02:00
|
|
|
const Slice& key, bool read_only,
|
2017-11-11 22:08:22 +01:00
|
|
|
bool exclusive, bool skip_validate) {
|
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;
|
2017-04-11 00:47:20 +02:00
|
|
|
bool lock_upgrade = 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
|
|
|
Status s;
|
|
|
|
|
2015-09-28 21:12:17 +02:00
|
|
|
// lock this key if this transactions hasn't already locked it
|
2017-11-11 22:08:22 +01:00
|
|
|
SequenceNumber tracked_at_seq = 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 {
|
2017-04-11 00:47:20 +02:00
|
|
|
if (!iter->second.exclusive && exclusive) {
|
|
|
|
lock_upgrade = true;
|
|
|
|
}
|
2015-09-30 02:00:16 +02:00
|
|
|
previously_locked = true;
|
2017-11-11 22:08:22 +01:00
|
|
|
tracked_at_seq = 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
|
|
|
|
2017-04-11 00:47:20 +02:00
|
|
|
// Lock this key if this transactions hasn't already locked it or we require
|
|
|
|
// an upgrade.
|
|
|
|
if (!previously_locked || lock_upgrade) {
|
2016-12-06 02:18:14 +01:00
|
|
|
s = txn_db_impl_->TryLock(this, cfh_id, key_str, 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 02:37:33 +02:00
|
|
|
}
|
|
|
|
|
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
|
2017-11-11 22:08:22 +01:00
|
|
|
if (skip_validate || snapshot_ == nullptr) {
|
2015-09-28 21:12:17 +02:00
|
|
|
// 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.
|
2017-11-11 22:08:22 +01:00
|
|
|
if (tracked_at_seq == kMaxSequenceNumber) {
|
2015-09-28 21:12:17 +02:00
|
|
|
// Since we haven't checked a snapshot, we only know this key has not
|
|
|
|
// been modified since after we locked it.
|
2017-11-11 22:08:22 +01:00
|
|
|
// Note: when allocate_seq_only_for_data_==false this is less than the
|
|
|
|
// latest allocated seq but it is ok since i) this is just a heuristic
|
|
|
|
// used only as a hint to avoid actual check for conflicts, ii) this would
|
|
|
|
// cause a false positive only if the snapthot is taken right after the
|
|
|
|
// lock, which would be an unusual sequence.
|
|
|
|
tracked_at_seq = db_->GetLatestSequenceNumber();
|
2015-09-28 21:12:17 +02:00
|
|
|
}
|
|
|
|
} 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.
|
2017-11-11 22:08:22 +01:00
|
|
|
// If we already have validated an earilier snapshot it must has been
|
|
|
|
// reflected in tracked_at_seq and ValidateSnapshot will return OK.
|
2015-09-28 21:12:17 +02:00
|
|
|
if (s.ok()) {
|
2017-11-11 22:08:22 +01:00
|
|
|
s = ValidateSnapshot(column_family, key, &tracked_at_seq);
|
2015-09-28 21:12:17 +02:00
|
|
|
|
|
|
|
if (!s.ok()) {
|
|
|
|
// Failed to validate key
|
|
|
|
if (!previously_locked) {
|
|
|
|
// Unlock key we just locked
|
2017-04-11 00:47:20 +02:00
|
|
|
if (lock_upgrade) {
|
|
|
|
s = txn_db_impl_->TryLock(this, cfh_id, key_str,
|
|
|
|
false /* exclusive */);
|
|
|
|
assert(s.ok());
|
|
|
|
} else {
|
|
|
|
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()) {
|
2017-11-11 22:08:22 +01:00
|
|
|
// We must track all the locked keys so that we can unlock them later. If
|
|
|
|
// the key is already locked, this func will update some stats on the
|
|
|
|
// tracked key. It could also update the tracked_at_seq if it is lower than
|
|
|
|
// the existing trackey seq.
|
|
|
|
TrackKey(cfh_id, key_str, tracked_at_seq, read_only, exclusive);
|
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_.
|
2017-11-11 22:08:22 +01:00
|
|
|
// tracked_at_seq is the global seq at which we either locked the key or already
|
|
|
|
// have done ValidateSnapshot.
|
2017-08-17 01:49:11 +02:00
|
|
|
Status PessimisticTransaction::ValidateSnapshot(
|
|
|
|
ColumnFamilyHandle* column_family, const Slice& key,
|
2017-11-11 22:08:22 +01:00
|
|
|
SequenceNumber* tracked_at_seq) {
|
2015-09-28 21:12:17 +02:00
|
|
|
assert(snapshot_);
|
|
|
|
|
2017-11-11 22:08:22 +01:00
|
|
|
SequenceNumber snap_seq = snapshot_->GetSequenceNumber();
|
|
|
|
if (*tracked_at_seq <= snap_seq) {
|
|
|
|
// If the key has been previous validated (or locked) at a sequence number
|
|
|
|
// earlier than the current snapshot's sequence number, we already know it
|
|
|
|
// has not been modified aftter snap_seq either.
|
2015-09-28 21:12:17 +02:00
|
|
|
return Status::OK();
|
|
|
|
}
|
2017-11-11 22:08:22 +01:00
|
|
|
// Otherwise we have either
|
|
|
|
// 1: tracked_at_seq == kMaxSequenceNumber, i.e., first time tracking the key
|
|
|
|
// 2: snap_seq < tracked_at_seq: last time we lock the key was via
|
|
|
|
// skip_validate option which means we had skipped ValidateSnapshot. In both
|
|
|
|
// cases we should do ValidateSnapshot now.
|
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
|
|
|
|
2017-11-11 22:08:22 +01:00
|
|
|
*tracked_at_seq = snap_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
|
|
|
ColumnFamilyHandle* cfh =
|
2017-08-03 17:46:47 +02:00
|
|
|
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
|
|
|
|
2017-11-11 22:08:22 +01:00
|
|
|
return TransactionUtil::CheckKeyForConflicts(
|
|
|
|
db_impl_, cfh, key.ToString(), snap_seq, 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
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
bool PessimisticTransaction::TryStealingLocks() {
|
2016-02-02 02:07:05 +01:00
|
|
|
assert(IsExpired());
|
2016-10-05 22:39:00 +02:00
|
|
|
TransactionState expected = STARTED;
|
|
|
|
return std::atomic_compare_exchange_strong(&txn_state_, &expected,
|
2016-02-02 02:07:05 +01:00
|
|
|
LOCKS_STOLEN);
|
|
|
|
}
|
|
|
|
|
2017-08-17 01:49:11 +02:00
|
|
|
void PessimisticTransaction::UnlockGetForUpdate(
|
|
|
|
ColumnFamilyHandle* column_family, const Slice& key) {
|
2015-09-15 02:11:52 +02:00
|
|
|
txn_db_impl_->UnLock(this, GetColumnFamilyID(column_family), key.ToString());
|
|
|
|
}
|
|
|
|
|
2017-08-08 01:07:40 +02:00
|
|
|
Status PessimisticTransaction::SetName(const TransactionName& name) {
|
2016-04-18 20:15:50 +02:00
|
|
|
Status s;
|
2016-10-05 22:39:00 +02:00
|
|
|
if (txn_state_ == STARTED) {
|
2016-04-18 20:15:50 +02:00
|
|
|
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
|