Update-anywhere replication of distributed systems

• US 7,765,186 B1
• Filed: 11/30/2005
• Issued: 07/27/2010
• Est. Priority Date: 04/13/2005
• Status: Active Grant

• Current Assignee: Progress Software Corporation
• Original Assignee: Progress Software Corporation
• Inventors: Hu, Carl, Rousseau, Benjamin A., Tompkins, Frederick A. Jr.
• Primary Examiner(s): Rones; Charles
• Assistant Examiner(s): Lewis; Alicia M
• Application Number:
  US11/292,055
• Time in Patent Office: 1,700 Days
• Field of Search: None
• US Class Current: 707/612
• CPC Class Codes: G06F 11/1425 by reconfiguration of node membership
• CPC Class Codes: G06F 16/273 Asynchronous replication or reconciliation
• CPC Class Codes: G06F 9/466 Transaction processing

Abstract

Transactional database replication techniques are disclosed. The techniques can be realized in many ways, such as in a replicator module that can be instantiated at each site in a distributed system. The techniques support a variety of data storage technologies such as traditional durable transactional databases as well as in-memory data storage systems, and a variety of data types such as relational, flat file, XML, and object-oriented data. Guaranteed levels of distributed transactional data consistency for both reads and writes are enabled for database applications. The techniques also support the ability of applications distributed across a network to perform updates concurrently, and the ability to replicate many standard database systems without needing to modify their intrinsic behavior or needing them to be aware of the replicator, and the ability to tolerate network and process failures without any change to the data consistency guarantees provided to the applications.

Claims

1. A replicator for replicating data among a plurality of processing sites each communicatively coupled to a network, the replicator comprising:

a fault tolerant distributed lock manager (FTDLM) for acquiring one or more distributed exclusive locks associated with a write transaction, the write transaction comprising a set of data changes; and

a total persisted order state machine (TPOSM) for creating and sending a change set to the plurality of processing sites, the change set comprising the data changes retrieved after the write transaction is committed and preserving change ordering in a total persisted order;

wherein the one or more distributed exclusive locks release after the change set is sent to the plurality of processing sites; and

wherein the plurality of processing sites reconcile the change set with a received change set and a received lock management message for changing ownership of the one or more distributed exclusive locks, so that the change sets are applied and locks are released in the total persisted order.

2. The replicator of claim 1 further comprising:

a replication coordination state machine (RCSM) for coordinating operation of replicator components including the TPOSM and the FTDLM to implement read, write, commit, and reconciliation rules.

3. The replicator of claim 1 further comprising:

a reconciler module for receiving data changes, and updating data in a storage with those data changes.

4. The replicator of claim 1 further comprising:

a change capture retriever (CCR) module for retrieving committed data changes from a storage so that those data changes can be sent by the TPOSM.

5. The replicator of claim 1 further comprising:

a router for multiplexing outputs from the FTDLM and the TPOSM into an extended virtual synchrony (EVS) transport.

19. The method of claim 1 wherein the reconciling is achieved using an external process that receives the delivered messages from a total order broadcast primitive.

6. A method for replicating data among a plurality of processing sites each communicatively coupled to a network, the method comprising:

acquiring one or more distributed exclusive locks associated with a write transaction, the write transaction comprising a set of data changes;

creating a change set that comprises the data changes retrieved after the write transaction is committed and preserves change ordering in a total persisted order;

sending the change set to the plurality of processing sites;

releasing the one or more distributed exclusive locks after the change set is sent to the plurality of processing sites; and

reconciling the change set with a received change set and a received lock management message for changing ownership of the one or more distributed exclusive locks, so that the change sets are applied and locks are released in the total persisted order.

7. The method of claim 6 further comprising:

coordinating operation of replication components to implement read, write, commit, and reconciliation rules.

8. The method of claim 6 further comprising:

receiving data changes; and

updating data in a storage with those data changes.

9. The method of claim 6 further comprising:

retrieving committed changes from a storage so that those data changes can be sent to the plurality of processing sites.

10. The method of claim 6 further comprising:

multiplexing outputs into an extended virtual synchrony (EVS) transport.

11. The method of claim 6 further comprising:

acquiring a distributed share lock associated with a read transaction; and

acquiring a local read lock associated with the read transaction.

12. The method of claim 6 further comprising:

acquiring a local write lock associated with the write transaction.

13. The method of claim 6 further comprising:

capturing the data changes in a database change log.

14. The method of claim 13 wherein the method further comprises:

retrieving the data changes associated with the write transaction from the database change log to form the change set that preserves change ordering; and

assigning an identifier to the change set, the identifier specifying a site that initiated the commit.

15. The method of claim 6 further comprising:

passing the change set to a total persisted order state machine (TPOSM), the TPOSM using an extended virtual synchrony (EVS) broadcast primitive.

16. The method of claim 11 wherein after the change set is sent to the plurality of processing sites, the distributed share lock and the one or more distributed exclusive locks associated with the transaction are released.

17. The method of claim 6 wherein sending the change set and identifier to the plurality of processing sites is achieved using a total order broadcast primitive.

18. The method of claim 6 wherein in response to the plurality of processing sites requesting the distributed lock, the method further comprises:

broadcasting messages to the plurality of processing sites using a total order broadcast primitive in order to transfer ownership of the distributed lock.

20. A machine-readable medium encoded with instructions, that when executed by a processor, cause the processor to carry out a process for replicating data among a plurality of processing sites each communicatively coupled to a network, the process comprising:

acquiring one or more distributed exclusive locks associated with a write transaction;

creating a change set that preserves change ordering in a total persisted order containing the data changes associated with the write transaction retrieved after the write transaction is committed;

sending the change set to the plurality of processing sites;

releasing the one or more distributed exclusive locks after the change set is sent to the plurality of processing sites; and

reconciling the change set with a received change set and a received lock management message for changing ownership of the one or more distributed exclusive locks, so that the change sets are applied and locks are released in the total persisted order.

21. The machine-readable medium of claim 20, the process further comprising:

coordinating operation of replication components to implement read, write, commit, and reconciliation rules.

22. The machine-readable medium of claim 20, the process further comprising:

receiving data changes; and

updating data in a storage with those data changes.

23. The machine-readable medium of claim 20, the process further comprising:

retrieving committed changes from a storage so that those data changes can be sent to the plurality of processing sites.

24. The machine-readable medium of claim 20, the process further comprising:

multiplexing outputs into an extended virtual synchrony (EVS) transport.

25. The machine-readable medium of claim 20, the process further comprising:

acquiring a distributed share lock associated with a read transaction; and

acquiring a local read lock associated with the read transaction.

26. The machine-readable medium of claim 20, the process further comprising:

acquiring a local write lock associated with the write transaction.

27. The machine-readable medium of claim 20, the process further comprising:

capturing the data changes in a database change log.

28. The machine-readable medium of claim 27 wherein after successfully committing the data changes, the process further comprises:

retrieving the data changes associated with the write transaction from the database change log to form a change set that preserves change ordering; and

assigning an identifier to the change set, the identifier specifying a site that initiated the commit.

29. The machine-readable medium of claim 20, the process further comprising:

passing the change set to a total persisted order state machine (TPOSM), the TPOSM using an extended virtual synchrony (EVS) broadcast primitive.

30. The machine-readable medium of claim 25 wherein after the change set is sent to the plurality of processing sites, all distributed share locks and distributed exclusive locks associated with the transaction are released.

31. The machine-readable medium of claim 20 wherein sending the change set to the plurality of processing sites is achieved using a total order broadcast primitive.

32. The machine-readable medium of claim 20 wherein in response to the plurality of processing sites requesting the distributed lock, the process further comprises:

broadcasting messages to the plurality of processing sites using a total order broadcast primitive in order to transfer ownership of the distributed lock.

33. The machine-readable medium of claim 20 wherein the reconciling is achieved using an external process that receives the delivered messages from a total order broadcast primitive.

34. The machine-readable medium of claim 20, the process further comprising the preliminary steps of:

requesting one or more distributed exclusive locks associated with the write transaction for locally stored data;

receiving data changes caused by another write transaction performed at another processing site on a replica of the locally stored data, the received data changes indicating differences between the locally stored data and the replica; and

guaranteeing that all the received data changes are applied to the locally stored data before any of the distributed exclusive locks associated with the write transaction are granted.