Is this in the server or the testsuite jvm?
If it is on the server, we need to check that it is not leaking classloaders
which will stop those objects in the perm generaton getting garbage collected.
Also, previously a similar problem was caused by the AOP tests turning on the flag
to transform all classes and not turning it off again.
This is on the server.
My trying now is to set
<start-jboss conf="all" jvmargs="-Xmx256m"/>
If it doesn't work I will try setting some -XX:PermSomething (I forgot the parameter names now... I will need to google it) in the JVM options.
And my guess is the leak on the classloads also. I just wanted to have this working in cruisecontrol.
I ran the full testsuite yesterday prior to your compatibility matrix addition and it ran without any memory issues. At what point is the OME occurring, and is this with or without your new compatibility tests?
The cruisecontrol for jboss-4.0 testsuite:
Is broken for about a month. Try to look at test results.
If you run cruisecontrol's, HEAD/tools/etc/cruisecontrol/scripts/do_test.sh you will have the PermSize problem. And this only occur in the QA environment.
I don't know exactly why do_test.sh which aparently doesn't do anything special would change PermSize behavior, but I know if I execute this script instead of calling ant directly I would have the error.
My test doesn't have anything to do with that. We are having the PermSize OME without activate the matrix testsuite, just our regular testsuite, as I didn't configure cruisecontrol for that yet. (It will require to create some directories at QA box - dev05).
We (Ryan and I) were running cruisecontrol using the PermSize option at the startserver. We have used both the default JVM (1.4.2) and changed it to 1.5_01, just to see what happens.
When using JVM 1.4.2 with PermSize option, the Server's JVM hangs. it hangs so badly that even a kill -3 was ignored by that JVM. We don't have any core dumped.
When using JVM 1.5_01 with PermSize option, the testsuite passed ok:
We are now removing the PermSize, and running it again with JVM 1.5, but it looks like we have some issue going on here.
I don't want to make any guesses now, but I will try to look for ClassLoading issues regarding deployments and undeployments happening on the testsuite.