At 01:16 PM 7/9/03 +1200, Simon Byrnand wrote:
> hi to all, > > I'm still having an unresolve problem with my Email server having > spamassassin. My server load increases when it scans less than a hundred > of > emails in transit. I'm running spamassassin daemon. > > Any idea why? Please help us.
To know the answer to that we'd actually have to know something about your system ;-)
Actualy Simon, if you backtrack, he's already provided most of the info you asked for back in a 6/20 thread you replied to :)
Sorry, my temporary memory buffer isn't that long, it gets flushed out every couple of days ;-) 20 days is ancient history on this mailing list...
Still, starting a new message thread more than 20 days later shouldn't require people to go digging back in the archives for the pertinent information....
To summarize the information he's provided:
Hardware: PII 350MHz 128MB (no specs on disk, but I'd assume a low-end IDE disk).
MTA: sendmail 8.11.6
Integration tool: Some commercial content filter called XAMIME (www.xamime.com) which calls a shell script as a pre-filter:
spamc -x -t 30 < $mailpack > $mailpack.tmp.
SA version: 2.54
Questions:
you stated the load went up after "only 100 messages".. 100 messages in what timeframe? 1 hour? 1 second?
Suggestions:
1) really whatever is calling spamc should be using some kind of mechanism to limit the number of spamc's it calls. If xamime is calling out 20 spamc's at a time, you're hosed. Make sure it calls them in a serial fashion, or at least not more than 5 at a time in parallel.
This is the key point. I actually limit the concurrency to 30 but thats on a nice Beefy P2.4Ghz with 1GB of ram, and it handles it fine. Only a lower spec machine, say <1Ghz and/or only 128MB of ram I'd limit the concurancy to between 5 and 10.
2) Do not use -t 30, unless you have disabled RBLs or modify your RBL timeout to be less than 30. If you use -x and -t 30, any RBL outage will cause SEVERE problems. Either increase to -t 40 or add the following to your local.cf:
I wouldn't recommend using -t at all...
rbl_timeout 10
Recommended in any case...
3) on a low-spec machine I'd consider disabling bayes by adding the following to your local.cf
use_bayes 0
You might also consider disabling RBL's entirely if you can't prevent "landslides" as per 1.
skip_rbl_checks 1
Certainly adding both of these options as a test to see if the problem goes away is worthwhile.
I don't think the rbl checks are a big problem, unless the timeout is set too long....
Regards, Simon
------------------------------------------------------- This SF.Net email sponsored by: Parasoft Error proof Web apps, automate testing & more. Download & eval WebKing and get a free book. www.parasoft.com/bulletproofapps _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
