I've had some recent cases where the pen process has jumped to 95% CPU. I'm
using the latest release (0.12.0) on Solaris 8 with the -T switch to limit
tracking of clients to a little over 3 hours (11000 seconds). I don't have
enough information to pinpoint the timing but something triggers pen to stop
answering penctl requests and stop doing its job of connecting users to a web
server.
Killing the pen process and letting its watchdog restart it solves the problem.
Since it was happening pretty regularly about noon each day I removed the -T
switch. If I go a couple of days without incident I will put it back in again as
a test to determine if this is where the problem resides.
More detail: the web servers serve the Horde/IMP webmail product. The IMAP
server resides on another system altogether. The web servers don't ever seem to
be overloading their systems remaining 97% idle according to the sar average.
The IMAP server has been experiencing high Disk I/O rates at the time it starts
refreshing it's processes (old daemons die off, new ones are spawned). CPU and
memory utilization on the IMAP server remain at reasonable levels. When the IMAP
server starts having high Disk I/O then the the user perceives webmail to be
slow. It is generally about this time that pen will stop doing it's job and it's
CPU usage soars.
Charlie Reitsma
Systems Engineer
This archive was generated by hypermail 2.1.2 : Thu Nov 06 2003 - 17:12:58 CET