activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From <Mats.Henrik...@sungard.com>
Subject JVM crash, out of permgen space?
Date Thu, 13 Aug 2009 19:53:01 GMT
Hi,

We had a JVM crash today running ActiveMQ 5.2 on a linux system, Java
HotSpot(TM) Server VM (1.5.0_17-b04 mixed mode).

It's an old JVM, so I have just upgraded it to the latest 1.6 update 16
version, just in case.

Looking at the JVM error file (hs_err_pidXXX.log) this is where it went
wrong: 

Current thread (0x080ff2c0):  JavaThread "CompilerThread1" daemon
[_thread_in_native, id=11524]

But we also have this in the memory summary:

compacting perm gen  total 24576K, used 24334K [0xed4f0000, 0xeecf0000,
0xf14f0000)
   the space 24576K,  99% used [0xed4f0000, 0xeecb3a58, 0xeecb3c00,
0xeecf0000)

Searching on the net this has been associated with running out of
permgen space (the 99% above), and upping the amount of permgen space
should help. However, if this is due to a leak then sure more permgen
space will buy more time, but eventually we'll always be hitting the
limit.

I've just abused my broker making thousands of connections on hundreds
of topics and queues and sending hundreds of thousands of messages with
both consumers and producers, basically trying to exercise it as much as
I could. With jconsole connected the non-heap memory is slowly ticking
up, mostly down to code cache (does this get GCed at some point?), but
the permgen is ever so slowly ticking up as well. I'm not a JVM memory
expert but once you are running and just doing the same thing over and
over shouldn't the permgen usage be static?

Has anybody else had problems with permgen, or strange JVM crashes? 


--
Mats Henrikson
Software Engineer
SunGard Kiodex

Think before you print




Mime
View raw message