[jboss-jira] [JBoss JIRA] Commented: (JBMESSAGING-1889) Duplicate messages are produced during cluster failover when a cluster node is shutdown

[Yong Hao Gao (JIRA)]

    [ https://issues.jboss.org/browse/JBMESSAGING-1889?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12616199#comment-12616199 ] 

Yong Hao Gao commented on JBMESSAGING-1889:
-------------------------------------------

The problem is that the ConnectionConsumer uses an internal non-transactional session to receive messages and then the messages were dispatched to server sessions, which can be a transactional session. However the commit of transaction is done through the internal session. During failover, JBM will recover messages that not acked/committed, but as the acks list goes with the internal session, which is non tx one, so JBM won't get the correct ACK list. The result is that the message that has been delivered won't get correctly processed at the server side, therefore the message will be delivered again.


> Duplicate messages are produced during cluster failover when a cluster node is shutdown
> ---------------------------------------------------------------------------------------
>
>                 Key: JBMESSAGING-1889
>                 URL: https://issues.jboss.org/browse/JBMESSAGING-1889
>             Project: JBoss Messaging
>          Issue Type: Bug
>          Components: Messaging Core
>    Affects Versions: 1.4.8.SP1
>         Environment: JBoss EAP 5.1.1
>            Reporter: Tom Ross
>            Assignee: Yong Hao Gao
>         Attachments: DistributedQueueExample.java
>
>
> The scenarion describes a set-up consisting of JBoss cluster and a stand alone JBoss instance hosting a MDB. 
> During the test the master node of the JBoss cluster is shutdown using CTRL/C or shutdown.sh script.
> The following is a description of the events that hava happened that lead to duplicate messages.
> - Message is delivered to a MDB from node-1 of the cluster to a MDB in a transaction. The transaction prepare has failed because the server peer of node-1 is shutting down.
> - Message is rolled back and received again in another transaction. This happens at the client side only. Then the transaction is committed on node-2 as the client has failed over to node-2 of the cluster.
> - Meanwhile the fact that node-1 is shutting down causes server side failover to happen and the message is failed over to node-2.
> - node-2 redelivers the message to the MDB in a separate transaction and committs.
> So the message is received twice.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira