Spamassassin cache loop

[Knut Bjornstad]

We have a problem with our MailScanner installations - a small part of
the traffic get stuck in the Spamassassin cache and loops back to
the hold queue for reprossessing. This happens to ordinary mail, I am 
not sure if any of them is spam. The loop is not infinite thou, most of
the cases escape after a few hundred iterations. But there are som mails
that has been stuck for months.

We have quite a lot of mail going through our MailScanner filters, so I
have not given this attention before the last days. But the load on the
boxes (we have several running in parallell), kept increasing very
slowly. I was not able to reset the cache in any way. My knowledge of
SQLite and sql bases in general are very limited, but at last I managed
to reset the base on one of the boxes by overwriting it with an empty base! 
This had no effect, the looping continued. At last I stopped it by removing the 
offending mails from the hold queue. Then the load (as reported by
uptime) on the boxes fell immediately.

Here is some obeservations I have made:
The log gets lines like this:
"May  8 00:08:04 xxxx MailScanner[882]: SpamAssassin cache hit for
message A3C3B310132.D0C33" where the last part of the session-id varies
as the mail is put back in the hold queue

I have found no particular type of mail that gets caught - but all of them have
a few MIME components.

When I remove a mail, and then put it back in the hold queue, without
any modification, it got sent immidiately!

Our installation:
postfix postfix-2.2.1 and 2.1.5 on Suse Linux (the behaviour occurs on
both versions)
MailScanner 4.50.15
spamassassin 3.0.4
ClamAV 0.88.1

I wonder if this problem can be rectified by upgrading MailScanner, or
by changing the cache parameters (but the config comment says you
shouldn't ordinarily do this)

I would appreciate any comment.
-- 
--Knut Bjornstad -- IKTDriftstjenester, ErgoGroup AS ---Oslo, Norway-------
--kbjo at interpost.no -- t:47 23 14 53 36 -- mob: 901 15 917 --