Just to followup, we have found a few things that were causing this leak, two that were particular to our setup, but the third seems to be a Tomcat problem (4.1.20 with Jasper2):
1) log4j was eating up a lot of memory and there was a slow leak. Since it wasn't strictly required, we've stopped using it and the largest leak stopped. 2) we are using jdbcpool from http://www.bitmechanic.com/projects/jdbcpool/ (it is the only connection pool we could find that can be instantiated programmatically from within a context without having to define a pool in advance via JDNI -- we give each context it's own database and therefore it's own pool) which doesn't seem to have a clean way to stop the pool manager thread when a context is stopped/reloaded. We've worked around this, however the memory leak remains and is due to context reloads / stops-starts 3) there seems to be a leak caused by reloading or stopping/starting a context (we have an automatic httpunit test that builds a jar file periodically and makes sure it is working in a context). We don't see the memory leak unless one or more JSPs are compiled before the context is reloaded or stopped/started. Is there some particular section of the code we should be examining to track this further? Adi > On Tue, 25 Feb 2003 22:08:41 -0600, Glenn Nielsen <[EMAIL PROTECTED]> said: > Aditya wrote: Glenn, several months ago you had posted a URL to a > document (at kinetic.more.net if I remember correctly) where you > talked about having to restart your production Tomcat(s) every 4 > weeks or so due to Heap exhaustion. Is that still the case? If so > what causes the heap exhaustion? >> > I think that part of the heap problem for me was the recent bug in > Jasper which I fixed where a number of resources such as Node trees > from a JSP page compile were not dereferenced between compiles. > This was fixed in Jasper before the 4.1.20 release. > We've looked high and low, with JProbe etc, and we still can't find > where the "leak" is. We're having to restart a Tomcat (4.1.20) with > -Xms and -Xmx both set to 256M every 4 days or so. >> > Does the increase in memory usage correlate with an increased number > of connectors due to a spike in request volume? > Perhaps you should try increasing the heap size. > Regards, > Glenn --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
