PARALLEL PROCESSING APPARATUS AND METHOD IN DATABASE MANAGEMENT SYSTEM FOR SYNCHRONOUS REPLICATION

• US 20130198133A1
• Filed: 05/30/2012
• Published: 08/01/2013
• Est. Priority Date: 01/31/2012
• Status: Active Grant

• Current Assignee: Altibase Corp.
• Original Assignee: Altibase Corp.
• Inventors: Lee, Seung Won, Choi, Jun Ho
• Granted Patent: US 9,026,491 B2
• Time in Patent Office: Days
• US Class Current: 707/611
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06F 16/275 Synchronous replication

Abstract

A parallel-processing apparatus and method provide for synchronous replication in a database management system (DBMS). During synchronous replication into the active node and the standby node in the DBMS, replications of transactions are parallelized in units of transactions, thereby improving a performance of the DBMS, guaranteeing atomicity and consistency of the transaction, and solving the deadlock state which may occur in parallel-processing.

Claims

1. A parallel-processing apparatus for an active node in a database management system (DBMS) replicated into the active node and a standby node, the parallel-processing apparatus comprising:

a transaction allocation unit configured to allocate a plurality of master transactions to a plurality of master-transaction replication threads which are executed in parallel;

a master-transaction replication parallel-processing unit configured to execute in parallel, processes for replicating master-transaction logs which log a change in a local database of the active node caused by the plurality of master transactions through the allocated plurality of master-transaction replication threads and transmitting replications of the master-transaction logs to the standby node, to allow a plurality of replicated transactions corresponding to the plurality of master transactions to be executed in the standby node;

a deadlock detection unit configured to detect a deadlock state between a plurality of pairs of the master transactions and the replicated transactions; and

a master-transaction control unit configured to cancel at least one of the plurality of pairs of the master transactions and the replicated transactions for which the deadlock state is detected.

2. The parallel-processing apparatus of claim 1, further comprising a replicated-transaction result reception unit configured to receive an execution result comprising information of whether or not a task included in at least one of the plurality of replicated transactions fails from the standby node.

3. The parallel-processing apparatus of claim 2, wherein the deadlock detection unit is configured to detect a deadlock state between the plurality of pairs of master transactions and replicated transactions based on the received execution result.

4. The parallel-processing apparatus of claim 2, wherein the master-transaction control unit is configured to cancel a master transaction corresponding to a replicated transaction for which execution of the task has failed according to the received execution result and instruct the standby node to cancel the replicated transaction corresponding to the canceled master transaction.

5. The parallel-processing apparatus of claim 1, wherein the standby node is configured to operate the plurality of replicated-transaction execution threads corresponding to the plurality of master-transaction replication threads, respectively, and

the master-transaction replication parallel-processing unit is configured to transmit a replication of each master-transaction log to a replicated-transaction execution thread corresponding to a master-transaction replication thread which replicates the each master-transaction log.

6. The parallel-processing apparatus of claim 1, wherein the plurality of master transactions have identification numbers sequentially allocated thereto, and

the transaction allocation unit is configured to allocate the master transactions to the master-transaction replication threads by using the identification numbers and an operation result based on the number of master-transaction replication threads.

7. A parallel-processing apparatus for a standby node in a database management system (DBMS) replicated into an active node and the standby node, the parallel-processing apparatus comprising:

a replicated-transaction parallel-processing unit configured to execute in parallel, processes for receiving replications of master-transaction logs which log a change in a local database of the active node caused by the plurality of master transactions, from the active node and executing a plurality of replicated transactions which correspond to the plurality of master transactions on a local database of the standby node based on the replications of the master-transaction logs through a plurality of replicated-transaction execution threads which are executed in parallel;

a deadlock detection unit configured to detect a deadlock state between a plurality of pairs of the master transactions and the replicated transactions; and

a replicated-transaction control unit configured to process a task included in at least one of a plurality of pairs of master transactions and replicated transactions for which the deadlock state is detected as an execution failure.

8. The parallel-processing apparatus of claim 7, wherein the deadlock detection unit is configured to recognize that the deadlock state occurs between the plurality of pairs of the master transactions and the replicated transactions if execution of a task included in at least one of the plurality of replicated transactions is not completed after an elapse of a preset waiting time, and

the replicated-transaction control unit is configured to process a task of which an execution is not completed after an elapse of the preset waiting time as an execution failure.

9. The parallel-processing apparatus of claim 7, further comprising a replicated-transaction result reporting unit configured to report an execution result comprising information about whether or not a task included at least one of the plurality of replicated transactions is not completed after an elapse of the preset waiting time to the active node.

10. The parallel-processing apparatus of claim 8, wherein the replicated-transaction control unit is configured to receive instruction to cancel a replicated transaction corresponding to the master transaction canceled in the active node and cancel the instructed replicated transaction.

11. A parallel-processing method in a database management system (DBMS) replicated into an active node and a standby node, the parallel-processing method comprising:

executing in parallel, processes for replicating master-transaction logs which log a change in a local database of the active node caused by a plurality of master transactions through a plurality of allocated master-transaction replication threads and transmitting replications of the master-transaction logs to the standby node;

executing in parallel, processes for receiving replications of master-transaction logs which log a change in a local database of the active node caused by the plurality of master transactions, from the active node and executing a plurality of replicated transactions which correspond to the plurality of master transactions on a local database of the standby node based on the replications of the master-transaction logs through a plurality of replicated-transaction execution threads which are executed in parallel;

detecting, by the standby node, a deadlock state between a plurality of pairs between the master transactions and the replicated transactions, processing a task included in at least one of the plurality of pairs between the master transactions and the replicated transactions for which the deadlock state is detected as an execution failure, and reporting the replicated transaction having the failing task to the active node; and

cancelling, by the active node, the master transaction corresponding to the replicated transaction having the failing task.

12. The parallel-processing method of claim 11, wherein the standby node operates the plurality of replicated-transaction execution threads corresponding to the plurality of master-transaction replication threads, respectively, and

the active node transmits a replication of each master-transaction log to a replicated-transaction execution thread corresponding to a master-transaction replication thread which replications the each master-transaction log.

13. The parallel-processing method of claim 11, wherein the plurality of master transactions have identification numbers sequentially allocated thereto, and

the active node allocates the master transactions to the master-transaction replication threads by using the identification numbers and an operation result based on the number of master-transaction replication threads.

14. The parallel-processing method of claim 11, wherein the standby node recognizes that the deadlock state occurs between the plurality of pairs of the master transactions and the replicated transactions if execution of a task included in at least one of the plurality of replicated transactions is not completed after an elapse of a preset waiting time, and processes a task of which an execution is not completed after an elapse of the preset waiting time as an execution failure.

15. The parallel-processing method of claim 11, further comprising:

instructing, by the active node, the standby node to cancel a replicated transaction corresponding to the master transaction canceled in the active node; and

cancelling, by the standby node, the instructed replicated transaction from the active node.

16. A non-transitory computer readable storage medium storing a computer program product including computer instructions configured to cause a processor of a computer to manage an active node in a database management system (DBMS) replicated into the active node and a standby node, the computer program product comprising:

a transaction allocation unit configured to allocate a plurality of master transactions to a plurality of master-transaction replication threads which are executed in parallel;

a master-transaction replication parallel-processing unit configured to execute in parallel, processes for replicating master-transaction logs which log a change in a local database of the active node caused by the plurality of master transactions through the allocated plurality of master-transaction replication threads and transmitting replications of the master-transaction logs to the standby node, to allow a plurality of replicated transactions corresponding to the plurality of master transactions to be executed in the standby node;

a deadlock detection unit configured to detect a deadlock state between a plurality of pairs of the master transactions and the replicated transactions; and

a master-transaction control unit configured to cancel at least one of the plurality of pairs of the master transactions and the replicated transactions for which the deadlock state is detected.

17. The non-transitory computer readable storage medium of claim 16, further comprising a replicated-transaction result reception unit configured to receive an execution result comprising information of whether or not a task included in at least one of the plurality of replicated transactions fails from the standby node.

18. The non-transitory computer readable storage medium of claim 17, wherein the master-transaction control unit is configured to cancel a master transaction corresponding to a replicated transaction for which execution of the task has failed according to the received execution result and instruct the standby node to cancel the replicated transaction corresponding to the canceled master transaction.

19. A non-transitory computer readable storage medium storing a computer program product including computer instructions configured to cause a processor of a computer to manage a standby node in a database management system (DBMS) replicated into an active node and the standby node, the computer program product comprising:

a replicated-transaction parallel-processing unit configured to execute in parallel, processes for receiving replications of master-transaction logs which log a change in a local database of the active node caused by the plurality of master transactions, from the active node and executing a plurality of replicated transactions which correspond to the plurality of master transactions on a local database of the standby node based on the replications of the master-transaction logs through a plurality of replicated-transaction execution threads which are executed in parallel;

a deadlock detection unit configured to detect a deadlock state between a plurality of pairs of the master transactions and the replicated transactions; and

a replicated-transaction control unit configured to process a task included in at least one of a plurality of pairs of master transactions and replicated transactions for which the deadlock state is detected as an execution failure.

20. The non-transitory computer readable storage medium of claim 19, wherein the deadlock detection unit is configured to recognize that the deadlock state occurs between the plurality of pairs of the master transactions and the replicated transactions if execution of a task included in at least one of the plurality of replicated transactions is not completed after an elapse of a preset waiting time, and

the replicated-transaction control unit is configured to process a task of which an execution is not completed after an elapse of the preset waiting time as an execution failure.