Jim Conner wrote:
> Im running spamassassin on a 700mHz machine.  When there are large
> amounts of email

What kind of email volume?  How many accounts?

How much memory is in this system?

I'm right on the borderline of stable continuous processing on a
PII/450/512M handling ~80-85K messages/week, SA called as spamc from
individual .procmailrc files, along with a full AV scan in a milter. 
I've also had to add a microapp to deliberately tempfail mail delivery
globally if the system load goes over a configurable level (currently
3.3).

> I get some serious resource issues.  The following is a
> snippet of my machine state.  I have no idea what is causing this
> problem.  Let me know what else I can provide to help everyone out in
> understanding what is going on better.

OK, based on your process list, you've managed to *completely* serialize
SpamAssassin calls from procmail (you have many, MANY idle procmail
processes, with matching sendmail processes, but only ONE spamassassin
process), and you're calling spamassassin instead of using spamd/spamc.

Deliberately serializing SA calls only helps if you're not seeing
processing delays, and if you can tempfail messages waiting for delivery
rather than letting the procmail and sendmail processes hang around
occupying memory.

Calling spamassassin instead of spamc is Bad on a high-load mail server,
and you should probably start by replacing all calls to spamassassin
with spamc (after making sure that spamd is running).  There is a VERY
significant overhead in starting the Perl interpreter and parsing the
global configuration for each call to spamassassin, where spamd/spamc
only incurs that overhead once when the spamd process starts.

> A few questions...what is going on with the tmp files in my home root
> directory???  Is that usual?

I've never seen any such files on any system I've adminned;  any way to
trace what's creating them and manipulating them?

>  Also, why is it taking so long for
> procmail to finish its steps?

Calling spamassassin vs spamc is probably part of the problem.

What's in your /etc/procmailrc, and your ~/.procmailrc, and a "typical"
~/.procmailrc for other accounts on the system?

It also looks a little like you might have some unusual configuration;
your mail log looks like you're not calling procmail as a local delivery
agent.

-kgd
-- 
"Sendmail administration is not black magic.  There are legitimate
technical reasons why it requires the sacrificing of a live chicken."
   - Unknown

Reply via email to