race condition between rollback() and the dispatcher thread in the java client
------------------------------------------------------------------------------
Key: QPID-573
URL: https://issues.apache.org/jira/browse/QPID-573
Project: Qpid
Issue Type: Bug
Components: Java Client
Affects Versions: M2
Reporter: Rafael H. Schloming
SVN change 572751 fixes a rather subtle race condition in the java client
implementation of rollback() that is responsible for one of the intermittent
failure modes of TransactedTest.
Prior to this change it was possible for the cleanup code in rollback() to
occasionally miss a message because the dispatcher thread in AMQSession holds
the message for a short time after extracting it from _queue and prior to
dispatching it to a consumer. If the rollback code runs during this period, the
prefetched message being held in the dispatcher thread is erroneously delivered
despite being rolled back.
I suspect this race condition also occurs on the M2 branch as well, so it may
be worth merging this fix to the branch. Placing a Thread.sleep(500) as the
first line in the dispatcher thread and running TransactedTest should
consistently reproduce the issue if it exists.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
