Re: Going from procmail based SA to spamass-milter produces more spam

[Ken Long]

On Fri, 2006-10-27 at 17:03 -0600, John E Hein wrote: 
> Ken Long wrote at 16:55 -0400 on Oct 27, 2006:
> My guess is that you are not using the right Bayes DB and AWL DB.

I've confirmed it IS using the Bayes DB, so that's not it.

> Bump up the debug on spamd (and maybe spamass-milter).
> Is it trying to lookup the Bayes db, etc., as the 'nobody' user?

I know it's doing the "nobody" thing for multiply addressed messages,
but that should be fine.  I even went in and manually trained "nobody"
with a couple thousand hams and spams to make sure that would not be the
problem.  In fact, that's how I know for sure the Bayes db's are right.
I had never done manual training before, so I took an account and
manually trained it and it got a lot better.  My boss's opinion is,
however, that we should not expect our users to have to learn how to
train the program and put a lot of work into it to just get back to
where they were before we made the change, and I can certainly
understand that position.  Going from procmail to milter should not have
been a regression.

> And take a look at 'spamassassin -t' scores for a particular
> spam that you received vs. the scores you see in the mail header.

Ok, I tried that, but I seem to be getting erratic results.  One message
was identical, but another had differences.

And here's an interesting twist.  I have spamass-milter set to reject
mail that matches my threshold.  I had spamassassin set to tag it prior
to going to the milter.  Just for kicks, I turned back on my procmail
filter so that now mail is first processed by spamass-milter and then
processed by spamc in my procmailrc file.  Messages that make it through
the milter are now being tagged as spam by SA, using the exact same
preferences and bayes db.  What could cause that???

I did see that there was a newer version of the milter, so I downloaded
0.3.1 and tried that, but it's still behaving the same.  

Any thoughts??

Thanks!

-Ken