[jboss-dev-forums] [Design of POJO Server] - Re: LogConfigurationException from commons-logging

[bstansberry@jboss.com]

Hmm, thought I'd replied before, but don't see the reply.

Anyway, pretty sure you were on the right track.  Log from last night's run showed a similar problem, but root cause was an ISE from BaseClassLoader when trying to use it after it had been undeployed -- i.e. a CL leak to a JGroups thread pool thread.

I think this started intermittently popping up when I set "loopback=true" on the channel configs to correct the FLUSH issues you were seeing.  All your fault. ;-)  With loopback enabled, app threads with the app's TCCL not only push messages down the stack, they also handle carrying the message back up and passing it off to the multiplexer thread pool. If the pool allocates a new thread to handle that message, the app's TCCL leaks to the pool.

I replaced the mulitplexer with the shared transport this evening; I think that should take care of this. Posted on jgroups-dev to check for other similar scenarios: 
http://www.nabble.com/Classloader-leaks-in-thread-pools-to16642502.html

I'm working to get it so the standard JEE distributed caches are lazy initialized. That means some user deployment's TCCL is going to be in effect when cache and associated channel are created.  Your post makes me realize I need to be sure to set the TCCL to the default domain loader before creating any cache or channel.

Thanks. :)

View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4143613#4143613

Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&p=4143613