Apache Tomcat crash on 6.1

[Kurt Miller]

Hi Martijn,

On Thursday 06 July 2006 9:27 am, Martijn Veening wrote:
> Hi Kurt, thanx,
> see below, for remarks:
> 
> On Thu, 2006-07-06 at 15:14, Kurt Miller wrote:
> 
> > Hi Martijn,
> > 
> > On Wednesday 05 July 2006 5:28 pm, Martijn Veening wrote:
> > > Hi Kurt, thnx for suggestions,
> > > 
> > > Indeed it seems that java process memory increases, although the reading
> > > is different from the tomcat manager reading. The size started at 450M
> > > (res 100M) and after 6 hours increased to 600M (res 160M). I can imagine
> > > that if that continues another 10 hours, the 1 Gb limit is reached
> > > causing crash.
> > > Tomcat however read lower memory-usage, but that will be memory within
> > > the JVM-reserved memory, which differs from the process-memory in
> > > FreeBSD (I presume).
> > 
> > Right, tomcat reported memory usage doesn't account for
> > memory the jvm uses in addition to the java object heap.
> > 
> > Hmmm, looking over the original bug report I don't see
> > -Xmx or -Xms arguments being used. Also, it could take
> > some time for the java process memory utilization to
> > stabilize. Even after it stabilizes there is some
> > fluctuation when full GC's occur.
> > 
> > For the sake of minimizing some of these issues, try
> > running tomcat with -Xmx250m -Xms250m so that all of
> > the java heap is allocated upfront and then look for
> > increasing java process memory usage after things
> > stabilize.
> 
> >From the beginning it runs with Xms200m and Xmx256m opts (added in
> catalina.sh script).
> It now runs for 16 hours and has added up to SIZE=891m and RES=581m
> (says 'top').
> It increases slowaly, but continually.

It does sound like you've found a leak. Now the hard part
is isolating it. It would be helpful to know if you see the
same behavior with the port version of the jdk. This would
reveal if the leak is new or not.

If you could try to isolate the leak to simple test program
that would be helpful. The last time I hunted down a leak in
the jdk I had a test program to work with. The process is
tedious - track each malloc and match it up with a free. 
Attempting to do that on tomcat with a full blown web app
would be impossible.

Regards,
-Kurt