Spamass-milter vs Milter-Spamc

[Cowles, Steve]

Ow Mun Heng wrote:
> Guys,
> 
> 	I'm doing Spamass-Milter currently. Don't have much spam to train on
> though. I only have like 18 spam mails.
> 
> Just wondering, which is the better alternative? I got spamass-milter
> from dag's repo and seems like it's out of development vs Milter-Spamc
> which is in active development.
> 
> (I don't want to go into procmail/MIMEDefang/mailscanner etc configs.
> I at a time)
> 
> Spamass-milter seems to work OK except that I it doesn't really _yet_
> know how to differentiate spam & Ham
> 
> This is what I have in /etc/mail/spamassassin/local.cf
> 
> required_hits                   5
> rewrite_subject                 1
> subject_tag                     [N-SPAM]
> report_safe                     0
> use_bayes                       1
> bayes_path                      /var/run/.spamassassin
> 
> Note : I did not turn on additional stuffs. just the above. I'm
> hoping I don't have to and get the Bayes filter to do it.
> 
> (but somehow autolearn is off in the headers)
> $ ps awux | grep spam
> root      spamass-milter -p /var/run/spamass.sock -f -r 15 -u
> defaultuser spamd /usr/bin/spamd -d -c -a -m5 -u spamd -x
> 
> Am I doing everything right here?

Spamass-milter is a sendmail milter designed to call spamassassin after
sendmail receives the DATA portion of the e-mail, but prior to sendmail
selecting the mailer for final delivery. i.e. procmail, smtp, etc...

With that in mind, if inbound e-mail are being piped to spamassassin through
spamass-milter, then tagged per your SA configuration, then re-injected back
into sendmail and then delivered per the defined mailer for your domain,
then the problems you describe in your post have to do with your spamassasin
configuration, not spamass-milter.

BTW: If I remember correctly, it takes something like a 1000+ e-mails before
the bayesion filters start kicking in and SA starts properly tagging
ham/spam. Even then, I had to train bayes with sa-learn to deal with FP's.
In fact, I still do.

Steve Cowles