[jboss-jira] [JBoss JIRA] (WFWIP-28) [Artemis 2.x upgrade] Unexptected crash of server in SOAK test

[Miroslav Novak (JIRA)]

    [ https://issues.jboss.org/browse/WFWIP-28?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13617987#comment-13617987 ] 

Miroslav Novak commented on WFWIP-28:
-------------------------------------

[~fnigro] After reducing Xmx to 1GB for both of the servers, SOAK test finally passed.

??HEAP size (-xmx) + OFF-HEAP/native allocations (they could be == -xmx, but it isn't counted by the JVM as heap allocation) + JVM footprint (~100MB would be enough).??
I agree with this formula. afaik Oracle JDK allocates OFF-HEAP memory == -Xmx (heap memory). Based on this we're still in limits but it's suspicious that it was not hit by Artemis 1.5.

[~mtoth]
>From customer perspective consuming more resources after upgrade of EAP 7.1(Artemis 1,5)->EAP 7.2(Artemis 2.x) is not a good thing. Do you have a test where you check native memory allocation for Artemis 2.x and how often it's cleaned by GC cycles?



> [Artemis 2.x upgrade] Unexptected crash of server in SOAK test
> --------------------------------------------------------------
>
>                 Key: WFWIP-28
>                 URL: https://issues.jboss.org/browse/WFWIP-28
>             Project: WildFly WIP
>          Issue Type: Bug
>          Components: Artemis
>            Reporter: Miroslav Novak
>            Assignee: Francesco Nigro
>            Priority: Blocker
>              Labels: feature-branch-blocker
>         Attachments: sosreport-rvaisdebug.asd-20180807101658.tar.xz
>
>
> After ~13 hours there is unexpected crash of one server in SOAK test. There is no error/warning in the logs. 
> Test Scenario:
> * Start 2 servers 
> * Client sends messages to input queue. Messages then go through:
> * One server to another through MDB reading and sending them from remote container through resource adapter
> * Messages are forwarded from one server to another over JMS bridge and back over Core bridge
> * Messages have JMSReplyTo defined with a temporary queue, that is filled with responses for the client
> * Messages are read from the destination with stateless EJB and sent back to clients
> * Client reads the messages after the pass through all the soak modules.
> Pass Criteria: In the last step receiver consumes all messages sent by producer.
> Actual Result:
> After ~13 hours 1st server suddenly crashes. There is no error/warning in server logs.
> Issue was hit with Artemis 2.5.0 with https://github.com/jmesnil/wildfly/tree/WFLY-9407_upgrade_artemis_2.4.0_with_prefix (commit 51dd8102f103ccb0470a3cfc8713d3f9bdb1b65d)



--
This message was sent by Atlassian JIRA
(v7.5.0#75005)