[jboss-jira] [JBoss JIRA] Commented: (JBMESSAGING-1159) Deadlocks with ClusterRoundRobinDistributor

[Tim Fox (JIRA)]

    [ http://jira.jboss.com/jira/browse/JBMESSAGING-1159?page=comments#action_12388639 ] 
            
Tim Fox commented on JBMESSAGING-1159:
--------------------------------------

> Yes, we are using it with caution, almost exclusively for in-container sending.

I can see from your stack that you're using Spring DefaultMessageListener container. I would be *very* careful so you don't lose messages (see http://www.jboss.org/index.html?module=bb&op=viewtopic&t=123433 - not the part about SQL)

> So far I have had quicker turnaround with the user forums and direct JIRA posts. 

Well.. sometimes you may be lucky :) But I wouldn't bet on it. If we have other higher priority stuff to do, the forums might go for days/weeks without us answering them.

Regarding the issue at hand:

I can see why the deadlock is occuring.

We have a scheduled delivery timing out and delivering a message.

This tries to obtain locks on the ClusterRoundRobinDistributor and Channel in that order

Concurrently you have a non scheduled delivery occurring. This is attempting to obtain locks on Channel and ClusterRoundRobinDistributor in that order.

Hence the deadlock.

I notice you are using a snapshot - so instability is expected.

The fix for this should be fairly straightforward. I'll make sure it goes into the SP2 release which should happen shortly.

> Deadlocks with ClusterRoundRobinDistributor
> -------------------------------------------
>
>                 Key: JBMESSAGING-1159
>                 URL: http://jira.jboss.com/jira/browse/JBMESSAGING-1159
>             Project: JBoss Messaging
>          Issue Type: Bug
>          Components: Messaging Core Distributed Support
>    Affects Versions: Stable branch
>         Environment: Platform AS 4.2.0.GA_CP01, JBM snapshot_stable 12-Nov-2007, Oracle 10g R2, RedHat 4, Sun Java 1.5.0_10-b03
>            Reporter: Travis Brown
>         Assigned To: Tim Fox
>         Attachments: threaddumps.zip
>
>
> During volume testing with 2 nodes I now see deadlocks in the thread dumps (platform AS 4.2.0.GA_CP01, JBM snapshot_stable 12-Nov-2007). My message processing comes to a halt, but I doubt it's fully related to these 6 threads (dealing with OutOfMemory too). Full thread dumps will be attached.

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://jira.jboss.com/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira