Non-disruptive data movement and node rebalancing in extreme OLTP environments

• US 9,075,858 B2
• Filed: 12/16/2010
• Issued: 07/07/2015
• Est. Priority Date: 12/16/2010
• Status: Active Grant

• Current Assignee: Sybase Incorporated (SAP SE)
• Original Assignee: Sybase Incorporated (SAP SE)
• Inventors: Andre, Mihnea, Cheng, Xun, Wang, Yanhong, Wildermuth, Elton Philip, Shang, Heping, Shepherd, Stephen C.
• Primary Examiner(s): Reyes, Mariela
• Assistant Examiner(s): VU, THONG H
• Application Number:
  US12/970,219
• Publication Number:
  US 20120158805A1
• Time in Patent Office: 1,664 Days
• Field of Search: 707/10, 707/101, 707/718, 707/822, 707/794, 707/695, 707/792, 707/764
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24552 Database cache management
• CPC Class Codes: G06F 16/278 Data partitioning, e.g. horizontal or vertical partitioning

Abstract

Systems and methods for non-disruptive cache startup, shutdown, data partition movement, and node rebalancing for a data fabric within a data grid in a database environment. The data grid is a clustered in-memory database (IMDB) cache comprising one or more data fabrics including multiple IMDB cache nodes. The method takes a time tag and moves data from source to destination nodes while allowing updates to data on the source nodes. The method propagates updates occurring after the time tag to destination nodes from source nodes and switches applications and ownership to the destination nodes. Another method slices data for a partitioned fabric and stores slices on nodes. Another method starts a data fabric by moving data from a grid backend to the nodes. In an embodiment, another method rebalances nodes in a non-disruptive manner. When a fabric is shutdown, the method saves data from the nodes to the grid backend.

Claims

1. A method, comprising:

moving a data partition containing a data page, in response to a request for moving the data partition, from a source node to a destination node while simultaneously allowing an application read and write access to the data partition on the source node, wherein the source node retains ownership of the data partition;

propagating data changes from the source node to the destination node, wherein the data changes were applied by the application to the data partition on the source node during the moving;

iteratively propagating additional data changes from the source node to the destination node until a number of remaining additional data changes is below a threshold value, wherein the additional data changes were applied by the application to the data partition on the source node during a previous propagation;

limiting the read and write access of the application to the source node containing the data partition at the conclusion of the iterative propagation;

instructing the application to perform future read and write operations on the destination node;

switching ownership of the data partition to the destination node; and

resuming the read and write access of the application to the data partition.

2. The method of claim 1, wherein the limiting further comprises:

allowing current application transactions against the source node to complete;

committing the allowed transactions on the source node;

blocking new application transactions targeting the source node; and

propagating data changes resulting from the committed transactions to the destination node.

3. The method of claim 1, further comprising:

taking a time tag before the moving by creating a marker in log records of a plurality of in-memory database (IMDB) system nodes in a data grid indicating the state of data at a point in time, wherein the data changes applied by the application during the moving were applied after a time indicated by the time tag; and

recording the marker.

4. The method of claim 1, wherein the threshold value is based on a resource utilization of the source node.

5. The method of claim 4, further comprising:

upon determining that an amount of additional data changes to the data partition to be propagated is greater than the threshold value, performing another iteration of propagation.

6. The method of claim 4, wherein the threshold value is a tunable value.

7. The method of claim 6, wherein the tunable value is a number of data manipulation language (DML) statements to be executed on the destination node.

8. The method of claim 1, further comprising:

determining if a current amount of additional changes to the data partition to be propagated is greater than or equal to a prior amount of additional changes in the previous propagation iteration;

upon determining that the current amount is less than the prior amount, performing another iteration of propagation; and

upon determining that the current amount is greater than or equal to the prior amount;

limiting the read and write access by throttling the access by the application to the source node, and

performing another iteration of propagation.

9. The method of claim 1, wherein the propagating and iteratively propagating are initiated as part of the response to the request for moving the data partition.

10. A non-transitory computer readable storage medium having instructions stored thereon that, in response to execution by a computing device, cause the computing device to perform operations for performing a non-disruptive cache shut down in a database environment, the operations comprising:

receiving a command to shut down a data fabric within a data grid in the database environment;

creating a time tag indicating the state of data at a point in time within an in-memory database (IMDB) system nodes;

moving a data partition containing a data page associated with the time tag, in response to the receiving, from the IMDB system node to a disk resident database (DRDB), while simultaneously allowing an application read and write access to the data partition on the IMDB system nodes, wherein the IMDB system node retains ownership of the data partition;

propagating data changes from the IMDB system node to the DRDB, wherein the data changes were applied by the application to the data partition on the IMDB system node during the moving;

iteratively propagating additional data changes from the IMDB system node to the DRDB until a sufficiently small number of additional data changes remain, wherein the additional data changes were applied by the application to the data partition on the IMDB system node during a previous propagation iteration, and wherein the sufficiently small number is based on a threshold value;

upon finishing the iterative propagation, limiting the read and write access of the application to the source node containing the data partition; and

shutting down the data fabric.

11. A computer program product comprising a non-transitory computer readable storage medium having computer program logic stored thereon that, in response to execution by a processor, causes the processor to perform a non-disruptive cache startup in a database environment by performing operations comprising:

receiving a command to start up a data fabric within a data grid in the database environment;

reading metadata within the data grid indicating a data partition associated with an IMDB system node;

moving the data partition containing a data page from a disk resident database (DRDB) to the IMDB system node, in response to the receiving, while simultaneously allowing an application read and write access to the data partition on the DRDB, wherein the DRDB retains ownership of the data partition;

propagating data changes from the DRDB to the IMDB system node, wherein the data changes were applied by the application to the data partition on the DRDB during the moving;

iteratively propagating additional data changes from the source node to the destination node until a number of remaining additional data changes is below a threshold value, wherein the additional data changes were applied by the application to the data partition on the source node during a previous propagation;

limiting the read and write access of the application to the DRDB containing the data partition on the conclusion of the iterative propagation;

instructing the application to perform future read and write operations on the IMDB system node;

switching ownership of the data page to IMDB system node; and

resuming the read and write access of the application to the data page.

12. The computer program product of claim 11, the operations further comprising, prior to the reading:

retrieving a time tag indicating the state of data at a point in time within a plurality of in-memory database (IMDB) system nodes organized as the data fabric, wherein the reading is based on the time tag.

13. The computer program product of claim 12, wherein the point in time is the time of a prior data fabric shutdown.

14. The computer program product of claim 11, wherein the propagating and iteratively propagating are initiated as part of the receipt of the command to start up the data fabric.

15. A system, comprising:

a data fabric comprising a first in-memory database (IMDB) system node and a second IMDB system node, the first IMDB system node and the second IMDB system node organized to store data according to a data granularity type;

a load determining module configured to determine a first load value for the first IMDB system node and a second load value for the second IMDB system node, wherein the load determining module is further configured to identify, based on the first and second load values, a more-loaded IMDB system node and a less-loaded IMDB system node;

a rebalancing module configured to, in response to a request to rebalance IMDB system nodes of the data fabric, move a data partition containing a data page from the more-loaded IMDB system node of the first IMDB system node and the second IMDB system node to the less-loaded IMDB system node to better distribute a data or processing load within the data fabric and retain data partition ownership with the more-loaded IMDB system node, wherein the rebalancing module is further configured to simultaneously allow an application read and write access to the data partition on the more loaded IMDB system node during the moving;

an update catch-up module configured to;

propagate data changes from the more-loaded IMDB system node to the less-loaded IMDB system node, wherein the data changes were applied by the application to the data partition on the more-loaded IMDB system node during the moving, and

iteratively propagate additional data changes from the source node to the destination node until a number of remaining additional data changes is below a threshold value, wherein the additional data changes were applied by the application to the data partition on the source node during a previous propagation; and

a switching module configured to;

limit the read and write access of the application to the more-loaded IMDB system node containing the data partition at the conclusion of the iterative propagation

instruct the applications to perform future read and write operations on the less-loaded IMDB system node;

switch ownership of the data partition to the less-loaded IMDB system node; and

resuming the read and write access of the application to the data partition.

16. The system of claim 15, wherein the first load value is determined as a relative load in relation to other IMDB system nodes in the data fabric.

17. The system of claim 15, wherein the first load value is determined as a relative processing load as compared to other IMDB system nodes in the data fabric.

18. The system of claim 15, wherein the first load value is determined as a relative data load as compared to other IMDB system nodes in the data fabric.

19. The system of claim 15, wherein the first load value is determined based on an absolute processing load on the first IMDB system node.

20. The system of claim 19, wherein the absolute processing load is measured as a percentage of central processing unit (CPU) capacity available on the first IMDB system node.

21. The system of claim 15, wherein the first load value is determined based on an absolute data storage load measured as a percentage of data storage capacity available on the first IMDB system node.

22. The system of claim 15, wherein the propagating and iteratively propagating are initiated as part of the response to the request to rebalance IMDB system nodes of the data fabric.