On Wed, 11 Feb 2004, David F. Skoll wrote: > On Wed, 11 Feb 2004, Jonas Eckerman wrote: > > > With most soplutions using an embedded perl interpreter, the perl > > interpreter is never unlodaded. This means that if the perl > > interpreter itself has memory leaks, those can accumulate over time. > > Correct. But read the multiplexor source code to see how we defend against > this.
Well, it happened again. I was running embedded perl, 15 min slaves, 50 max slaves. At the time this happened, the max slaves limit got hit. There were 200+ sendmail processes running at this time. No single process was consuming any great amount of memory. Swap utilization was creeping up, but not high. The load average shot over 100, and the CPU free dropped from it's usual 80-90 to around 10. Other network applications started throwing errors at this time also: apache and imapproxyd are the ones I noticed. I killed all sendmail, mimedefang, apache, and imapproxyd processes, but nothing brought the machine back. I had to run over and use console because the ssh connections I had were almost unuseable. I'm somewhat suspicious the broadcom network interfaces / tg3 driver may be involved. I've started logging /proc/net/dev once a minute to see if the interfaces start throwing errors. I also started logging vmstat output every fifteen seconds. Anything else anyone would recommend logging? I'm now running without embedded perl, min slaves set to 25, and max set to 100. I'd like to be more scientific about exploring the boundary conditions, but I'm more interested in getting things stable... _________________________ Ron Peterson Network & Systems Manager Mount Holyoke College _______________________________________________ Visit http://www.mimedefang.org and http://www.canit.ca MIMEDefang mailing list [EMAIL PROTECTED] http://lists.roaringpenguin.com/mailman/listinfo/mimedefang
