Wolfgang Jeltsch wrote: > Am Dienstag, 8. August 2006 23:04 schrieb Bowie Bailey: > > Wolfgang Jeltsch wrote: > > > > > > I use Courier MTA. Courier MTA replaces certain mailformed mails > > > with mails which contain some explaination and the original mail > > > as an attachment. The attachment of the mail you're just reading > > > contains such a mail produced by Courier MTA. > > > > > > Do those modifications done by Courier MTA confuse SpamAssassin's > > > spam detection algorithm? Does SpamAssassin look at attachments > > > at all? If yes, are they taken as seriously as message bodies? > > > And what about training the bayesian filter? Should I feed such > > > Courier-MTA-modified mails to sa-learn or should I better not do > > > this? > > > > Yes, those modified emails will confuse SA. They will also confuse > > your users. The best option is to tell Courier to leave the emails > > alone. > > > > In your /etc/courier/bofh file, add this line: > > > > opt BOFHBADMIME=accept > > Thanks for this tip. I didn't know that it's possible to stop > Courier MTA rewriting those mails.
There are lots of options to control Courier's behavior. Unfortunately, the documentation is a bit piecemeal and the config options are scattered among several different files, so it can be a bit challenging to figure out how to do things. If you join the Courier mailing list, Sam (or someone else) will usually answer questions on how to do things. 'man courier' will tell you about most of the config files and their options. > > I use Courier as well and SA works great for me. The main thing you > > will want to do is start up the spamd daemon and use spamc instead > > of spamassassin in maildrop. > > I decided against this but have forgotten why I did so. Maybe > because of security issues. Since my server serves very few users, I > see no resource problems in using spamassassin instead of > spamc/spamd. But could using spamc/spamd resolve the locking problem > I described? There shouldn't be many security issues with running spamd. The main daemon runs as root, but it spawns off unprivileged children to deal with the connections. The main advantage is speed and resource usage. The Perl language is very powerful, but loading the interpreter (and all of the spam rules) is slow. Spamd lets you load all of that at startup and speeds up spam processing considerably. If spamassassin works for you, there's no real reason to change, but if you start running into resource problems or slowdowns, this is the first change I'd make. As an example, a 7k message from my server took 3.7 seconds to scan with spamassassin and only 1.1 seconds with spamc. I don't know if spamd would help your locking problem. Usually that is caused by multiple processes trying to open the same Bayes DB. You may want to watch the server and see if it tends to have multiple spamassassin processes running from time to time. If this is the case, then switching to spamd might help since it would cause the processes to finish faster. In general, a low-volume server can deal with running spamassassin on every message, but as soon as your volume increases, or you start having to deal with mail coming in large batches, you should switch to spamc/spamd. > > If you have any specific questions about interfacing SA with > > Courier, I'll be glad to help out. > > Thanks a lot! > > Best wishes, > Wolfgang -- Bowie
