[
https://issues.apache.org/jira/browse/QPID-3704?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13180249#comment-13180249
]
Keith Wall edited comment on QPID-3704 at 1/5/12 8:58 AM:
----------------------------------------------------------
Problem is a race condition between an incoming enqueue and an outbound
delivery. It is trigger by a particular data pattern. I think the problem lies
in OutOfOrderQueue#checkSubscriptionsNotAheadOfDelivery().
Imagine a set of messages with keys {3, 5, 4}. Message 3 has been sent to a
consumer, message 5 is in the process of being delivered, and message 4 is
being enqueued.
When the unlucky timing strikes, the thread processing the enqueue observes
_lastSeenEntry at 3, so does not update _releasedEntry. A moment later the
delivery thread sending Message 5 to the consumer calls #setLastSeenEntry
setting _lastSeenEntry to 5. This means message 4 becomes stuck and will
remain undelivered to the consumer until a message with key less than 4 arrives.
I think the _lastSeenEntry optimisation is unsafe and incoming deliveries
should always _releasedEntry backwards if necessary. I attach a patch. Any
thoughts?
was (Author: k-wall):
Problem is a race condition between an incoming enqueue and an outbound
delivery. It is trigger by a particular data pattern. I think the problem lies
in OutOfOrderQueue#checkSubscriptionsNotAheadOfDelivery().
Imagine a set of messages with keys {3, 5, 4}. Message 3 has been sent to a
consumer, message 5 is in the process of being delivered, and message 4 is
being enqueued.
When the unlucky timing strikes, the thread processing the enqueue observes
_lastSeenEntry at 3, so does not update _releasedEntry. A moment later the
delivery thread sending Message 5 to the consumer calls #setLastSeenEntry
setting _lastSeenEntry to 5. This means message 4 becomes stuck and will
remain undelivered to the consumer until a message with key less than 4 arrives.
I think the _lastSeenEntry optimisation is unsafe and incoming deliveries
should always move _releasedEntry backwards. I attach a patch. Any thoughts?
> SortedQueueTest.testTransactedSortedQueue failing occasionally on 0-9-1
> profile
> -------------------------------------------------------------------------------
>
> Key: QPID-3704
> URL: https://issues.apache.org/jira/browse/QPID-3704
> Project: Qpid
> Issue Type: Bug
> Components: Java Broker, Java Tests
> Reporter: Keith Wall
> Fix For: 0.15
>
> Attachments:
> 0001-QPID-3704-OutOfOrderQueue-checkSubscriptionsNotAhead.patch
>
>
> Test appears to have begun failing after the performance improvement work.
> See
> https://builds.apache.org/view/M-R/view/Qpid/job/Qpid-Java-Java-TestMatrix/188
> and
> https://builds.apache.org/view/M-R/view/Qpid/job/Qpid-Java-Java-TestMatrix/184
> {code}
> Incorrect number of messages received expected:<200> but was:<199>
> Stacktrace
> junit.framework.AssertionFailedError: Incorrect number of messages received
> expected:<200> but was:<199>
> at
> org.apache.qpid.server.queue.SortedQueueTest.runThroughSortedQueueForSessionMode(SortedQueueTest.java:153)
> at
> org.apache.qpid.server.queue.SortedQueueTest.testTransactedSortedQueue(SortedQueueTest.java:115)
> at
> org.apache.qpid.test.utils.QpidBrokerTestCase.runBare(QpidBrokerTestCase.java:240)
> at org.apache.qpid.test.utils.QpidTestCase.run(QpidTestCase.java:135)
> {code}
--
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
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
