WrackWeb - Jean Respen wrote:
Hello,
I've some troubles with spamassassin. Sometimes (and i don't know why
exactly) my server has high load (like 50 or 60 or more) so it's
impossible to connect via ssh, everything crash (apache, qmail, vpopmail,
etc...) and i have to go to the datacenter to reboot the server... and if
i'm able to connect via ssh (after having waited 40 minutes...) i can see
that perl or spamd are the processes which take the most part of the load
(perl due to qmailscanner, who invoke spamassassin). I'm under a debian
stable, i've installed spamassassin via package AND source... maybe it's
because of this
In my syslog i've this just before it crashed :
Jul 24 15:49:07 localhost spamd[16465]: spamd: result: . 1 -
BAYES_00,SPF_HELO_SOFTFAIL,SPF_SOFTFAIL
scantime=1169.8,size=4834,user=qscand,uid
=0,required_score=5.0,rhost=localhost.localdomain,raddr=127.0.0.1,rport=44964,mid=<[EMAIL
PROTECTED]>,bayes=0,autolearn=no
Jul 24 15:49:27 localhost imapd: DISCONNECTED, user=hebergement,
ip=[::ffff:83.79.3.207], headers=398, body=99132, time=62494
Jul 24 15:55:47 localhost spamd[29579]: prefork: child states:
BBBBBBBBBBBBBBBBBBBBBBBBBKBBBBBBBB
Jul 24 16:03:18 localhost spamd[11444]: spamd: copy_config timeout,
respawning child process after 1 messages at /usr/bin/spamd line 969.
Has anyone got an idea?
Unless you've got at least 2GB of RAM in the system, you've got way too
many spamd children running and are swap thrashing. You normally want
to budget at least 45MB of RAM per child in addition to whatever memory
other things on the system need.
So, you can start off by reducing the number of children to avoid load
related problems in the first place (note: spamd is failing due to
system load, the system load is probably due to swap thrashing and not
spamd). Then you can upgrade SpamAssassin to a newer version which has
a better chance of surviving, and not then further contributing to, such
a high load.
Daryl
