[ http://issues.apache.org/jira/browse/GERONIMO-1125?page=all ]
Kevan Miller updated GERONIMO-1125:
-----------------------------------
Attachment: InterceptorDestroy.patch
The attached patch breaks the chain of strong references that will keep
MultiParentClassLoaders alive. It also performs additional cleanup of
ManagedConnections maintained by Pooling Interceptors.
ConnectionInterceptor interface has a new destroy() method.
AbstractConnectionManager now implements GBeanLifeCycle. doStop() or doFail()
will cause destroy() to be called on the stack of ConnectionInterceptors.
destroy() will cause the IdleReleaser TimerTask (if one exists) to be
cancelled. Also, any Pooled ManagedConnections will be destroyed. If a
ManagedConnection is returned after a PoolInterceptor has been destroyed, the
ManagedConnection is destroyed during returnConnection() processing. Finally,
if getConnection() is called on a destroyed Pooling Interceptor, a
ResourceException will be thrown.
> AbstractSinglePoolConnectionInterceptor$IdleReleaser keeps
> MultiParentClassLoaders alive
> ----------------------------------------------------------------------------------------
>
> Key: GERONIMO-1125
> URL: http://issues.apache.org/jira/browse/GERONIMO-1125
> Project: Geronimo
> Type: Bug
> Components: connector
> Versions: 1.0-M5
> Environment: JDK 1.4/WinXP
> Reporter: Kevan Miller
> Assignee: Kevan Miller
> Fix For: 1.0
> Attachments: InterceptorDestroy.patch
>
> After a deploy/undeploy of DayTrader, the following chain of references
> (there are others which I'm investigating) is keeping a MultiClassLoader
> instance from being marked as available for GC.
> java.util.TaskQueue.queue --> java.util.TimerTask[128]
> java.util.TimerTask[5] -->
> org.apache.geronimo.connector.outbound.AbstractSinglePoolConnectionInterceptor$IdleReleaser
> IdleReleaser.this$0 -->
> org.apache.geronimo.connector.outbound.SinglePoolConnectionInterceptor
> SinglePoolConnectionInterceptor.next -->
> org.apache.geronimo.connector.outbound.XAResourceInsertionInterceptor
> XAResourceInsertionInterceptor.next -->
> org.apache.geronimo.connector.outbound.MCFConnectionInterceptor
> MCFConnectionInterceptor.stack -->
> org.apache.geronimo.connector.outbound.ConnectionTrackingInterceptor
> ConnectionTrackingInterceptor.next -->
> org.apache.geronimo.connector.outbound.TCCLInterceptor
> TCCLInterceptor.classLoader -->
> org.apache.geronimo.kernel.config.MultiParentClassLoader
> The default interval for the IdleReleaser TimerTask is 30 minutes. Plenty of
> time for us to run out of PermGen space. Currently the task is never
> cancelled. I'm working on cancelling the task. However, that's not
> sufficient. TimerTask.cancel() simply updates state. It doesn't remove the
> Task from the TimerQueue. So, the task lives until it expires (looks like
> this "feature" is fixed in 1.5). Easiest fix is to break the chain of
> references at the IdleReleaser task when the task is cancelled. This should
> be good enough. Alternative is to implement our own Timer -- which wouldn't
> be too hard... Or have multiple Timers and cancel the whole timer...
> I'm working on breaking the chain of references at IdleReleaser. Note that
> this means the IdleReleaser classloader will be kept alive until the
> TimerTask expires. However, the IdleReleaser classloader is a long-lived
> Geronimo class loader. So, this shouldn't be a problem, but it's not ideal,
> either...
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
