[infinispan-dev] AS clustered classloader leak hunt and web thread local storage keeping reference to a org.infinispan.context.impl.LocalTxInvocationContext ...

[Scott Marlow]

I'm seeing the following:

1.  InvocationContextContainerImpl.createInvocationContext creates a new
LocalTxInvocationContext (lets call it #3011) and sets thread local
(icTL) to it.

2.  We InvocationContextContainerImpl.suspend() LocalTxInvocationContext
#3011 which clears the thread local reference.

3.  We InvocationContextContainerImpl.resume() LocalTxInvocationContext
#3011 which sets the thread local reference.

4.  We InvocationContextContainerImpl.suspend() LocalTxInvocationContext
#3011 which clears the thread local reference.

5.  We InvocationContextContainerImpl.resume() LocalTxInvocationContext
#3011 which sets the thread local reference.

I don't see another suspend call, so LocalTxInvocationContext #3011 is
still referenced by the thread local InvocationContextContainerImpl.icTL

I'll create a ISPN jira for this. :)


Scott

On Mon, 2010-11-15 at 13:12 -0500, Scott Marlow wrote:
> I'm getting a little closer, should be closer sooner :)
> 
> The LocalTxInvocationContext that got left in thread local storage was
> instantiated from here http://pastie.org/1300157
> 
> That should bring me closer to the cause.  
> 
> On Mon, 2010-11-15 at 10:29 -0500, Scott Marlow wrote:
> > Hi,
> > 
> > I'm trying to chase down some memory leaks for
> > https://jira.jboss.org/browse/JBAS-8613 (turned into a catch all jira
> > for a few different AS clustering failures).
> > 
> > I captured a picture of a classloader leak that is attached to
> > JBAS-8613.  
> > 
> > The http thread threadLocals has a
> > org.infinispan.context.impl.LocalTxInvocationContext that is keeping the
> > classloader in memory (via what it references which is shown in the
> > abovce link or more directly
> > https://jira.jboss.org/secure/attachment/12338702/Eclipse+Memory
> > +Analyzer+Tool.png)
> > 
> > I'm looking at a different problem at the moment but would like to fix
> > this leak soon.
> > 
> > Why would org.infinispan.context.impl.LocalTxInvocationContext still be
> > referenced in thread local storage?  Any suggestions on specific classes
> > that should be looked at?
> > 
> > Thanks,  
> > Scott
> > 
> > _______________________________________________
> > infinispan-dev mailing list
> > infinispan-dev at lists.jboss.org
> > https://lists.jboss.org/mailman/listinfo/infinispan-dev
> 
> 
> _______________________________________________
> infinispan-dev mailing list
> infinispan-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/infinispan-dev