[
https://issues.apache.org/jira/browse/AMQ-3872?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gary Tully resolved AMQ-3872.
-----------------------------
Resolution: Fixed
Assignee: Gary Tully
fix in http://svn.apache.org/viewvc?rev=1345202&view=rev
JDBC 2PC is now two JDBC ops, an insert/update in prepare and an update for the
outcome.
An additional xid column is needed in both the acks and messages table. A non
null value for the xid means that the update is pending the outcome of the xa
transaction matching that xid.
Recovery scans for non null xid and recreates the appropriate state pending the
outcome.
Pending transactions are visible via jmx so that they can be forced.
> Implement "exactly once" delivery with JDBC and XA in the event of a failure
> post prepare
> -----------------------------------------------------------------------------------------
>
> Key: AMQ-3872
> URL: https://issues.apache.org/jira/browse/AMQ-3872
> Project: ActiveMQ
> Issue Type: Bug
> Components: Message Store
> Affects Versions: 5.6.0
> Environment: jdbc persistence adapter, XA
> Reporter: Gary Tully
> Assignee: Gary Tully
> Labels: JDBC, MessageStore, XA, recovery
> Fix For: 5.7.0
>
>
> JDBC store does not honor the prepare phase so a broker failure after prepare
> always results in a rollback which is in violation of the contract.
> In the cases where XA is required, the store needs to do its best to respect
> a prepare outcome.
> Having the TM use last resource commit one phase (or the one phase commit
> optimisation) avoids this issue but is not always possible.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
