[jboss-jira] [JBoss JIRA] (WFLY-10531) Wildfly leaks ActiveMQ connections

[Jan-Willem Gmelig Meyling (Jira)]

    [ https://issues.jboss.org/browse/WFLY-10531?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13642247#comment-13642247 ] 

Jan-Willem Gmelig Meyling commented on WFLY-10531:
--------------------------------------------------

Hi [~jondruse], I have been trying to construct a reproducer myself as well. Apparently this is quite an edge case, because even with some background knowledge about the situation I am trying to create, I have failed to reproduce it so far. I know at least one "broken" use case of JMSContext must exist in my code base, because I am suffering from the issue, but I've failed to isolate it so far. From looking at the code  the fix must be sufficient, but I just can't imagine why the situation would occur in the first place. I am also wondering whether we aren't creating new instances of WFLY-9501.

My current belief is that we might incorrectly think that the inTX() check is to blame, and that perhaps the issue is that the resource is added under a conflicting name, henceforth thinks a clean up listener is already added and just ignores the new JMSContext?  I am thinking for example about sub transactions (i.e. an EJB method entering an EJB TX within a CDI TX or an EJB entering a method with transaction NEW mode). Do they share resources with the "parent" tx?

Anyway, I'd would like to create a reproducer test case so we can gather some confidence in whether this implementation solves both WFLY-10531 and WFLY-9501 in all cases :) I'll invest some time in it too, see what I can come up with.

> Wildfly leaks ActiveMQ connections
> ----------------------------------
>
>                 Key: WFLY-10531
>                 URL: https://issues.jboss.org/browse/WFLY-10531
>             Project: WildFly
>          Issue Type: Bug
>    Affects Versions: 13.0.0.Final
>         Environment: openjdk 8 / openjdk 9, Linux
>            Reporter: Marcel Šebek
>            Assignee: Jeff Mesnil
>            Priority: Major
>         Attachments: WFLY-10531-ear-1.0.ear, WFLY10531.zip
>
>
> After upgrading our application from wildfly 12 to 13, the app started to crash after a while (hours, days, depending on circumstances). It crashes on
> IJ000453: Unable to get managed connection for java:/JmsXA
> and other errors (it simply cannot perform all the jobs it contains). I found that when shutting down the server which has been running for a while, I can see a bunch of these messages in the log:
> WARN  [org.jboss.jca.core.connectionmanager.pool.strategy.PoolByCri] (ServerService Thread Pool -- 117) [:::] IJ000615: Destroying active connection in pool: ActiveMQConnectionDefinition (org.apache.activemq.artemis.ra.ActiveMQRAManagedConnection at 2f37f69)
> Bascially, the longer the server was running, more of these messages are shown. I cannot find a way how to reproduce the issue. When the server runs for short time but with some load, no connection is leaked (or just one, rarely). On the other side, it leaks connections even without any particularly high load (just a few requests and @Schedule jobs) when running for longer time.
> It may also be a bug in our application, which just happen to have more serious impact with the new wildfly version.



--
This message was sent by Atlassian Jira
(v7.12.1#712002)