9 Replies Latest reply on Feb 25, 2009 6:26 PM by pmuir

Memory not released after undeploy

dallen6 Feb 25, 2009 1:12 PM

When using the Web Beans (JSR-299) TCK test harness, many individual WARs and EARs are deployed individually, used, and then subsequently undeployed before the next is deployed. As each deployment is loaded, the amount of memory the JVM uses for JBoss 5.0.0.GA or 5.0.1.GA increases significantly. In the case of this TCK today, the memory usage rises quickly from 360MB after initial startup to over 1.3GB. This is likely to double in the coming weeks as new tests are added.

The log does show that each of these deployments is eventually being undeployed. For instance, before the next deployment, a DEBUG line like the following appears:

2009-02-25 00:30:32,015 DEBUG [org.jboss.deployers.plugins.deployers.DeployersImpl] (WorkerThread#0[127.0.0.1:40319]) Fully Undeployed vfszip:/home/dallen/JBossServer/jboss-5.0.1.GA/server/default/deploy/org.jboss.jsr299.tck.integration.context.passivating.PassivatingContextTest.war/

Is there a known problem with not cleaning up after something is undeployed (vfs)? Any way to force the memory to be released?

Thanks.

1. Re: Memory not released after undeploy

pmuir Feb 25, 2009 1:56 PM (in response to dallen6)

On my box (Mac OSX) the artifacts do undeploy when we call deploymentManager.undeploy() so I don't think it is a problem with the artifacts remaining in the deploy dir to long (you can also verify this by watching the content of the deploy dir and see how many artifacts are there at any one time.

If people want to try to reproduce, it's quite easy as the TCK run is automated.
Actions
2. Re: Memory not released after undeploy

dallen6 Feb 25, 2009 1:56 PM (in response to dallen6)

A few other details left out:

The OS is Ubuntu Linux 8.04LTS with kernel version 2.6.24-23.

The JVM is the Sun Java SDK 1.6.0_07-b06.
Actions
3. Re: Memory not released after undeploy

pmuir Feb 25, 2009 1:58 PM (in response to dallen6)

See http://seamframework.org/WebBeans/JSR299TCKPortingPackage#H-RIPortingPackage for instructions to run the TCK.
Actions
4. Re: Memory not released after undeploy

pmuir Feb 25, 2009 2:04 PM (in response to dallen6)

NB looks like a heapspace problem - the JVM runs out (with 512m allocated).
Actions
5. Re: Memory not released after undeploy

starksm64 Feb 25, 2009 2:18 PM (in response to dallen6)

We do have some memory leak testing in the jbossas testsuite, but it just takes a little issue to pull in huge objects graphs that are not getting cleaned up across redeployments. I'll create a JBAS issue to look into the tck behavior.
Actions
6. Re: Memory not released after undeploy

brian.stansberry Feb 25, 2009 2:26 PM (in response to dallen6)

Just an FYI, the testing Scott's referring to is described here:

https://www.jboss.org/community/docs/DOC-9294

As noted on the wiki, that testing covers a pretty limited number of basic code paths.
Actions
7. Re: Memory not released after undeploy

starksm64 Feb 25, 2009 2:26 PM (in response to dallen6)

Its https://jira.jboss.org/jira/browse/JBAS-6545
Actions
8. Re: Memory not released after undeploy

brian.stansberry Feb 25, 2009 2:30 PM (in response to dallen6)

Also FYI there is this known issue that seems to affect the Sun VM:

https://jira.jboss.org/jira/browse/JBAS-5959
Actions
9. Re: Memory not released after undeploy

pmuir Feb 25, 2009 6:26 PM (in response to dallen6)

I'll also double check that I'm not causing the leak (doubtful I think).

In the meantime I will put in a "restart jboss every 50 tests" to test deployment impl for the profile service.

Thanks for the help.
Actions

Go to original post