I'm a little bit confused.
Looks like in the beginning you had problems with
ConcurrentLinkedQueue#Node leak, right?
Then, after applying the patch you see
LinkedBlockingQueue#Node leak, which is mostly caused by
sun.nio.ch.SelectorImpl.select()
The problem is that patch doesn't change the type of queue in
sun.nio.ch.SelectorImpl, so the same LinkedBlockingQueue#Nodes had to
be created even without the patch. Can you pls. check that?
Thanks.
WBR,
Alexey.
On Mar 30, 2010, at 15:23 , glassfish_at_javadesktop.org wrote:
> According to the heapdump that's the stack the nodes are created.
> Attached an analysis done by eclipse memory analizer.
>
> We don't think that the comet causes the problems, because most of
> the nodes contain requests to css files, standard http request and
> so on.
>
> Suddenly, the server got out of memory, so we've got a 2 gb
> heapdump. We'll start analyzing it and post again.
> [Message sent by forum member 'kimschneider']
>
> http://forums.java.net/jive/thread.jspa?messageID=394475
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe_at_glassfish.dev.java.net
> For additional commands, e-mail: users-help_at_glassfish.dev.java.net
>