Parallel recovery by non-failed nodes

• US 8,234,517 B2
• Filed: 04/23/2004
• Issued: 07/31/2012
• Est. Priority Date: 08/01/2003
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Bamford, Roger J., Chandrasekaran, Sashikanth, Pruscino, Angelo
• Primary Examiner(s): Baderman, Scott
• Assistant Examiner(s): TRUONG, LOAN
• Application Number:
  US10/831,413
• Publication Number:
  US 20040215640A1
• Time in Patent Office: 3,021 Days
• Field of Search: 714/3, 714/4.1, 714/15, 714/33, 714/42
• US Class Current: 714/4.1
• CPC Class Codes: G06F 16/27 Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor

Abstract

Various techniques are described for improving the performance of a shared-nothing database system in which at least two of the nodes that are running the shared-nothing database system have shared access to a disk. Specifically, techniques are provided for recovering the data owned by a failed node using multiple recovery nodes operating in parallel. The data owned by a failed node is reassigned to recovery nodes that have access to the shared disk on which the data resides. The recovery logs of the failed node are read by the recovery nodes, or by a coordinator process that distributes the recovery tasks to the recovery nodes.

Claims

1. A method comprising computer-implemented steps of:

determining that a particular node has failed, wherein the particular node is in a shared-nothing database system in which at least some database data, on which the particular node performed operations, is stored on a particular storage device; and

in response to determining that the failure has occurred, performing parallel recovery, by two or more non-failed nodes that have physical access to the particular storage device, of at least some of the database data on which the particular node performed operations.

2. The method of claim 1, further comprising computer-implemented steps of:

maintaining a plurality of persistent data items on the particular storage device;

wherein the particular storage device is accessible to a plurality of nodes;

wherein the persistent data items including a particular data item stored at a particular location on said particular storage device;

assigning exclusive ownership of each of the persistent data items to one of the nodes;

wherein a particular node of said plurality of nodes is assigned exclusive ownership of said particular data item;

wherein when any node wants an operation to be performed that involves said particular data item, the node that desires the operation to be performed ships the operation to the particular node for the particular node to perform the operation on the particular data item as said particular data item is exclusively owned by said particular node;

wherein performing parallel recovery comprises;

(a) assigning, to each of two or more recovery nodes, exclusive ownership of a subset of a set of persistent data items that were involved in the failure, and (b) each recovery node of the two or more recovery nodes performing a recovery operation on the subset of persistent data items that were assigned to that recovery node.

3. The method of claim 1, wherein the failure is a media failure of the particular storage device.

4. The method of claim 1, wherein the failure is a failure of a node that has exclusive ownership of a set of persistent data items that are stored on the particular storage device, and wherein performing parallel recovery further comprises assigning, to each of two or more recovery nodes, exclusive ownership of a subset of the persistent data items that were exclusively owned by the failed node.

5. The method of claim 4, wherein:

the two or more recovery nodes include a first recovery node and a second recovery node; and

a least a portion of a recovery operation performed by the first recovery node on the subset of data exclusively assigned to the first recovery node is performed in parallel with at least a portion of a recovery operation performed by the second recovery node on the subset of data exclusively assigned to the second recovery node.

6. The method of claim 4, further comprising computer-implemented steps of:

organizing, into a plurality of buckets, a plurality of persistent data items that are stored on the particular storage device; and

establishing a mapping between the plurality of buckets and a plurality of nodes, wherein each node has exclusive ownership of the data items that belong to all buckets that map to that node; and

determining which data items need to be recovered based on said mapping.

7. The method of claim 6, further comprising computer-implemented steps of:

performing a first pass on said mapping to determine which buckets have data items that need to be recovered;

performing a second pass on said mapping to perform recovery on the data items that need to be recovered; and

after performing the first pass and before completing the second pass, making available for access the data items that belong to all buckets that do not have to be recovered.

8. The method of claim 4, wherein each recovery node of the two or more recovery nodes performs a recovery operation based on recovery logs, associated with the failed node, on the particular storage device.

9. The method of claim 8, further comprising the computer-implemented step of a recovery coordinator scanning recovery logs associated with the failed node and distributing recovery records to the two or more recovery nodes.

10. The method of claim 8, wherein each of the two or more recovery nodes scans recovery logs associated with the failed node.

11. The method of claim 1, wherein performing parallel recovery comprises:

each of the two or more non-failed nodes applying undo records to blocks; and

tracking which undo records have been applied.

12. The method of claim 1, further comprising the computer-implemented step of, prior to the failure, the failed node storing, within redo records that are generated by the failed node, bucket numbers that indicate buckets to which the data items associated with the redo records belong.

13. The method of claim 1, wherein performing parallel recovery comprises:

a recovery coordinator determining that a first set of one or more tasks required for recovery of said failed node should be performed serially, and that a second set of one or more tasks required for recovery of said failed node should be performed in parallel; and

performing the first set of one or more tasks serially; and

using said two or more non-failed nodes to perform said second set of one or more tasks in parallel.

14. The method of claim 13, wherein the step of determining that a second set of one or more tasks required for recovery of said failed node should be performed in parallel is performed based, at least in part, on the size of one or more objects that need to be recovered.

15. The method of claim 13, wherein:

ownership of data items involved in said second set of one or more tasks is passed from the recovery coordinator to the two or more non-failed nodes to allow said two or more non-failed nodes to perform said second set of one or more tasks; and

after performance of said second set of one or more tasks and before completion of the recovery of said failed node, ownership of data items involved in said second set of one or more tasks is passed back to said recovery coordinator from said two or more non-failed nodes.

16. A non-volatile or volatile computer-readable non-transitory medium carrying one or more sequences of instructions which, when executed by one or more processors, causes the one or more processors to perform steps comprising:

determining that a particular node has failed, wherein the particular node is in a shared-nothing database system in which at least some database data, on which the particular node performed operations, is stored on a particular storage device; and

in response to determining that the failure has occurred, performing parallel recovery, by two or more non-failed nodes that have physical access to the particular storage device, of at least some of the database data on which the particular node performed operations.

17. The non-volatile or volatile computer-readable storage medium of claim 16, wherein the steps further comprise:

maintaining a plurality of persistent data items on the particular storage device;

wherein the particular storage device is accessible to a plurality of nodes;

wherein the persistent data items including a particular data item stored at a particular location on said particular storage device;

assigning exclusive ownership of each of the persistent data items to one of the nodes;

wherein a particular node of said plurality of nodes is assigned exclusive ownership of said particular data item;

wherein when any node wants an operation to be performed that involves said particular data item, the node that desires the operation to be performed ships the operation to the particular node for the particular node to perform the operation on the particular data item as said particular data item is exclusively owned by said particular node;

wherein performing parallel recovery comprises;

(a) assigning, to each of two or more recovery nodes, exclusive ownership of a subset of a set of persistent data items that were involved in the failure, and (b) each recovery node of the two or more recovery nodes performing a recovery operation on the subset of persistent data items that were assigned to that recovery node.

18. The non-volatile or volatile computer-readable storage medium of claim 16, wherein the failure is a media failure of the particular storage device.

19. The non-volatile or volatile computer-readable storage medium of claim 16, wherein the failure is a failure of a node that has exclusive ownership of a set of persistent data items that are stored on the particular storage device, and wherein performing parallel recovery further comprises assigning, to each of two or more recovery nodes, exclusive ownership of a subset of the persistent data items that were exclusively owned by the failed node.

20. The non-volatile or volatile computer-readable storage medium of claim 19, wherein:

the two or more recovery nodes include a first recovery node and a second recovery node; and

a least a portion of a recovery operation performed by the first recovery node on the subset of data exclusively assigned to the first recovery node is performed in parallel with at least a portion of a recovery operation performed by the second recovery node on the subset of data exclusively assigned to the second recovery node.

21. The non-volatile or volatile computer-readable storage medium of claim 19, wherein the steps further comprise:

organizing, into a plurality of buckets, a plurality of persistent data items that are stored on the particular storage device; and

establishing a mapping between the plurality of buckets and a plurality of nodes, wherein each node has exclusive ownership of the data items that belong to all buckets that map to that node; and

determining which data items need to be recovered based on said mapping.

22. The non-volatile or volatile computer-readable storage medium of claim 21, wherein the steps further comprise:

performing a first pass on said mapping to determine which buckets have data items that need to be recovered;

performing a second pass on said mapping to perform recovery on the data items that need to be recovered; and

after performing the first pass and before completing the second pass, making available for access the data items that belong to all buckets that do not have to be recovered.

23. The non-volatile or volatile computer-readable storage medium of claim 19, wherein each recovery node of the two or more recovery nodes performs a recovery operation based on recovery logs, associated with the failed node, on the particular storage device.

24. The non-volatile or volatile computer-readable storage medium of claim 23, wherein the steps further comprise the step of a recovery coordinator scanning recovery logs associated with the failed node and distributing recovery records to the two or more recovery nodes.

25. The non-volatile or volatile computer-readable storage medium of claim 23, wherein each of the two or more recovery nodes scans recovery logs associated with the failed node.

26. The non-volatile or volatile computer-readable storage medium of claim 16, wherein performing parallel recovery comprises:

each of the two or more non-failed nodes applying undo records to blocks; and

tracking which undo records have been applied.

27. The non-volatile or volatile computer-readable storage medium of claim 16, wherein the steps further comprise the step of, prior to the failure, the failed node storing, within redo records that are generated by the failed node, bucket numbers that indicate buckets to which the data items associated with the redo records belong.

28. The non-volatile or volatile computer-readable non-transitory storage medium of claim 16, wherein performing parallel recovery comprises:

a recovery coordinator determining that a first set of one or more tasks required for recovery of said failed node should be performed serially, and that a second set of one or more tasks required for recovery of said failed node should be performed in parallel; and

performing the first set of one or more tasks serially; and

using said two or more non-failed nodes to perform said second set of one or more tasks in parallel.

29. The non-volatile or volatile computer-readable non-transitory storage medium of claim 28, wherein the step of determining that a second set of one or more tasks required for recovery of said failed node should be performed in parallel is performed based, at least in part, on the size of one or more objects that need to be recovered.

30. The non-volatile or volatile computer-readable non-transitory storage medium of claim 28, wherein:

ownership of data items involved in said second set of one or more tasks is passed from the recovery coordinator to the two or more non-failed nodes to allow said two or more non-failed nodes to perform said second set of one or more tasks; and

after performance of said second set of one or more tasks and before completion of the recovery of said failed node, ownership of data items involved in said second set of one or more tasks is passed back to said recovery coordinator from said two or more non-failed nodes.