Java application monitoring tool?

[Leon A. Howorth]

I have a flaky (excuse the technical jargon) java-based application server software product that we are finding difficult to troubleshoot. There doesn't seem to be any pattern for when failures occur, and the application logs are not helping to determine cause of failure. The only thing we have determined is that if we have only ONE (java 1.2.2 process) for this application running on a DEDICATED HP-UX 11i machine it appears to be alot more stable - as opposed to being on a "shared" app server system with other similar as well as other vendor's processes. I would like to get a monitoring tool, specific to java apps, that would allow me to automatically capture more information when the process seems to "lock up". Right now we simply have the operators stop and start the process. Unfortunately I cannot simply dump the application.
Since the "lock-ups" often occur off-hours I would prefer something that either requires no intervention or very little operator intervention.
Any suggestions regarding a monitoring tool are welcome. Thanks.

[Sridhar Bhaskarla]

Hi,

The immediate thing that occured to me was using tusc on this process. However, since there is no pattern and you cannot reproduce the problem, then analyzing the tusc output will be of no use.

You can use Measureware to monitor the application. Define this process as an "Application" inside the measureware. Measureware can collect a variety of data corresponding to this application. Look at /var/opt/perf/reptall and search for APP and see what all it can offer.

If you want more details, then you can digout stats using the PROC metrics. However, Measure logs information about the processes only if they cross the thresholds as defined in /var/opt/perf/parm file. Look for "process" and you will find thresholds for CPU and disk. You can adjust the threshold for CPU to get this process logged. Having too less threshold will make your system spend resources in just monitoring.

There is an evaluation version available in the Application CDs. Try it and if it is useful, then you can purchase the product.

-Sri

[Charles Floss]

Try this url: http://hpux.cs.utah.edu/. They have a large number of utilities. One in particular was "tusc". It helped me determine why the JVM crashing without logging any problems. Running it with the PID of the JVM process allowed me to find out that the JVM was receiving a SIG11, and that it was in a BEA class when the event occured.