Which JDK version are you using, I have had problems with Sun 1.4.0 crashing that went away with 1.4.1.
Haven't seen it first hand on Linux and we run RedHat 7.2 with the 2.4.18 kernel, using an IBM SDK.
I've only seen Tomcat crash under Tru64. I assume you don't have any problems such as the JSP compiled classes being cleaned up by Linux.
What JDK is it? Apparently some people have experienced a few problems remedied by things such as setting -Xrs. I think that was for 1.3.x JDK releases.
It is also possible to crash Tomcat with leakage due to the Java compiler in the VM. I'm not sure that it applied to the 4.0.x releases but it did in the 4.1.x releases. The Sun JDK 1.4.1_02 and _03 releases were supposed to deal with this issue. We still use Jikes as it is faster and the Tomcat developers recommended it as it is said not to cause the memory leakage problem. I doubt it is your issue as a crash time of 2 hours is pretty small - although you do not state the load conditions.
If all else fails, you may want to put it under observation using something like jmemprof.
Hope it helps.
Thanks for the swift replies!
We are using the Sun jdk1.4.0_02, but I will try upgrading the jdk to 1.4.1 next week hopefully.
One other detail I didn't mention in the first post is that we have SLSBs on this machine that lookup SLSBs on another machine over the internet and call methods on them (as they have access to our Oracle db).
The second machine is also running RedHat 7.2 and JBoss3.0.4 / Tomcat 4.0.6 and has never crashed (it only has EJBs, no JSPs/Servlets/HTML).
A crack-pot theory we have is that the communication between the 2 servers is over the internet, which does have a tendency to go down sporadically where we are, and that maybe this disconnection in the communication causes JBoss to crash or shutdown somehow.
Of course the other explanation could be that since the second machine does not handle web requests at all, it is tomcat that is bringing down jboss on the first machine.
With regards to load, there is little load on the machine, maybe 100 concurrent users at most, usually much less. Also there does not seem to be greater load when the machine crashes after 2 hours than when it stays up for 6 weeks.
Upgrading may fix your problem. Thinking about it, it may pay to try -Xrs in the mean time since your JVM may be receiving OS signals that it interprets as a kill or terminate instruction, perhaps due to a quirk in 1.4.0.
My problem with 1.4.0 had nothing to do with signals from the operating system.
The JVM was attempting access a class loader that had since been removed, when this happened the JVM core dumped.
The crashes occur seemingly randomly, it might stay up for 6 weeks before crashing, or sometimes only 2 hours.
I am now having this exact same problem with JBoss 3.2.3. Did you ever resolve this problem? What was the resolution?
Nathan, which JDK are you using?