[infinispan-issues] [JBoss JIRA] (ISPN-1444) Infinispan fails to shutdown gracefully

[James Mackenzie (JIRA)]

    [ https://issues.jboss.org/browse/ISPN-1444?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13150265#comment-13150265 ] 

James Mackenzie commented on ISPN-1444:
---------------------------------------

I had the same issue and came across this page on a Google Search.

The fix for any others that stumble across this:
* If you unregister the shutdown hooks via DONT_REGISTER, you're responsible for stopping Infinispan caches yourself
* If you don't, and you're using a Clustered Configuration, your Application Server process never gracefully terminate due to non-daemon JGroups theads (like those above) still running
* You need to call CacheManager.stop() yourself (perhaps via a ServletContextListener)

If you'd like more details, I wrote up the fix over here:
http://www.jamesfmackenzie.com/2016/01/18/infinispan-shutdown-fix/

Thanks!

Jimmy

> Infinispan fails to shutdown gracefully
> ---------------------------------------
>
>                 Key: ISPN-1444
>                 URL: https://issues.jboss.org/browse/ISPN-1444
>             Project: Infinispan
>          Issue Type: Bug
>    Affects Versions: 4.2.1.FINAL, 5.0.1.FINAL
>            Reporter: Luc Boudreau
>            Assignee: Manik Surtani
>            Priority: Blocker
>
> We have embeded Infinispan into our project, but when used, we cannot gracefully shutdown the JVM anymore. 
> There are a few exceptions thrown by late-access to the classloader from log4j, but these errors are easy to work around. Tomcat blocks any class loading after an application has been marked as shutting down. I can load them in the classloader in my application at runtime and circumvent those issues.
> The main problem is the fact that the threads are just hanging there. They are not marked as daemon threads, so the JVM doesn't shut them off automatically, if required. The culprit threads are:
>  - OOB-1
>  - OOB-2
>  - multicast receiver
>  - unicast-receiver
>  - TransferQueueBuilder
> Some of these threads might be related to JGroups. Please advise, and I will create a separate ticket in their bug tracker if needed.
> I know Infinispan registers a shutdown hook in order to cleanly shut down, but it looks quite unreliable and causes a lot of problems for us.



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)