Here is the summary I posted a few months back: ---
Well, we mostly figured out what the problem is and how to work around it. Sun's best guess is we are hitting this bug from 2003 [1], which causes Felix to not be garbage collectible occasionally despite the lack of a gc root. The problem was exasperated by the fact that our system bundle activators weren't releasing their services on shutdown, causing the whole of JIRA to be dragged into the memory leak black hole. We fixed that issue with lots of service deregistrations and nulls (I guess sometimes final isn't the best option), so while we are still leaking a few Felix instances, our total memory leak over 200 restarts is 7mb. For the record, this bug is only with Sun's JVM 1.5 and 1.6 in server mode. Client mode, in a 32-bit JVM, is fine as is IBM's JVM. Thanks for the help, Don [1] http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=4957990 On Mon, Sep 14, 2009 at 4:58 PM, Felix Meschberger <[email protected]> wrote: > Hi, > > Don Brown schrieb: >> On Wed, Sep 9, 2009 at 10:33 PM, Karl Pauls <[email protected]> wrote: >>> framework is being shutdown. Additionally, you have more then one >>> framework inside the JVM. >> >> While this may technically be true, it is because the Java 1.6 server >> VM has a bug where it won't properly let go of Felix, so the instances >> start adding up. Apparently this bug has been around since 2004. > > That sentence alerts me ! > > Would you be able to tell me more about this issue ? Thanks alot. > > Regards > Felix > > >
