Has anyone else seen such behavior?
See it all the time. The root cause is dangling object references that pin the parent class loader and all the classes it loaded on startup. For non-trivial apps that is typically a lot of data. Since class definitions are placed in the PermGen slot of the JVM, it quickly fills up on repeated Webapp deploys. Increasing PermGen size only delays the inevitable.
This comes up occasionally but it's genuinely a difficult problem. It is feasible to track down all the places where we might leak resources (e.g. thread locals) in CAS, but the problem is fairly intractable when considering the myriad libraries on which CAS depends. The Spring Framework and Hibernate are two huge dependencies; there are many others. It's easier to simply deploy CAS in a Tomcat instance by itself and restart the container as part of redeployment. I believe most CAS deployers do this; we certainly do.
M -- You are currently subscribed to [email protected] as: [email protected] To unsubscribe, change settings or access archives, see http://www.ja-sig.org/wiki/display/JSG/cas-user
