NATIVE Heap issues with Introscope and WAS 7 fp17,19,21

Discussion created by WarrenPage on Aug 4, 2012
Latest reply on Aug 6, 2012 by kuchimanchisrikanth
Reposting my reply to other thread with heading so others may catch it.

There is a known NATIVE memory issue when using IBM Java SDK 1.5 SR9 and a Profiling tool. This is caused because of a JITing issue that means finalization methods are not being called to clean up resources.


The fix is to go to SR10 (WAS FP23) - which I havent done so cannot confirm.

The work around mentioned in the link DOES however work for me -

ie.. Add •

I havent seen any noticeable performance problem when doing this.

From the Introscope 9.1.1 Readme notes...

OOM error requires update to IBM JDK SR10 - 71618
Due to a third-party issue, if you are using the CA APM agent to monitor applications that run IBM's Java 1.6 and are using the preferred –javaagent switch, you will experience severe memory overhead. This is due to a native memory leak in the JDK. The applications eventually run out of memory. The memory leak occurs irrespective of the max heap size (-Xmx) setting. This issue also occurs when Autoprobe instrumentation is disabled (introscope.autoprobe.enable=false) while still using –javaagent. The issue is specific to object finalization in the IBM JVM. Details of this issue are available from IBM’s website at http://www-01.ibm.com/support/docview.wss?uid=swg1IZ99243.
As a temporary workaround, you can use the AutoProbe Connector for IBM and deploy it in the bootstrap classpath using the –Xbootclasspath switch. Note that this is a temporary workaround only and it is recommended that you obtain IBM’s official patch that resolves this issue.

Note I also have an issue with heap dumps on transaction traces which looks like it might be caused by some EJB2/3 incompatibilities. Still workiong on that one - might be a JDK issue but hoping an Introscope fix could stop it.