Mark Adams wrote:
>
> Currently this isn't enabled, I've also disabled the URIDNSBL and
> Hashcash checking options incase this was the issue - but this has not
> helped the situation
So you don't have Bayes enabled at all? Just recently I had to fight a
spamd going CPU nuts on an Exim/SA machine that has been working for
years (it's RedHat 7.3, woo!), and as others pointed out it was the
bayes auto_expire that started...I don't know, just going really wrong.
(Tokens Gone Wild!)
Anecdotal evidence from googling is that spamd has issues auto-expiring
(but I couldn't determine the *exact* reason for this, maybe a bug?)
that can be solved by using a bayes journal and sa-learn.
1) in your local.cf, set:
bayes_auto_expire 0
bayes_learn_to_journal 1
bayes_journal_max_size 0
...and restart spamd.
2) nightly sync and expire your bayes using "sa-learn --sync" and
"sa-learn --force-expire", like at 2am.
Check your databases -- i.e. with BerkeleyDB run "db_verify" on the
auto_whitelist and bayes_*. It's weird how these things can get corrupt
all on their own. Also, my auto_whitelist was huuuuge, like 80meg; I
downloaded 'trim_whitelist' and cleaned out years of cruft, bringing it
down to 8meg or so which I'm sure helped. You can grab it here:
http://www.deepnet.cx/~kdeugau/spamtools/
Given all the problems I've seen, every time SA breaks it always has
something to do with the db files that need fixed, it's never Exim or
the actual spam-scanning itself.
hope this helps,
-te
--
Troy Engel | Systems Engineer
Fluid Inc. |
http://www.fluid.com
