Re: [qmailtoaster] Training spamassassin with Maildir/Qmail

[Jake Vickers]

Jared Markell wrote: Thanks Jake - your method works great. It was what I was trying to do, but wasn't sure how to get it to work with the /new /cur Maildir layout. So I did the whole "for" loop idea and sent it message by message to sa-learn then deleted them. So slow...   Anyways, the only thing I changed on your script is added "--no-sync" to the sa-learn --spam lines. The way it's setup on your page, it resyncs as it learns, then you resync (again) at the end, so I think that's double the work? no-sync allows it to skip the syncing 'til the end.   /usr/bin/sa-learn --spam --no-sync /home/vpop .... ld.com/spam/Maildir/new/*   Hope this tweaks performance a bit!   Also, I toyed with the idea of having the catch-all on our domain to send all that mail to a user named "Spam"'s email inbox, which then gets sent to sa-learn. This way, spammers who are guessing usernames are actually doing us a favor! I haven't implemented this yet - I wanted to get other people's ideas/comments on this first. Anything ? Glad it helped out. I had considered creating a spam box like you mention, but after doing some research decided to hold off on this. The consensus is mixed on the subject; some say it will taint your DB, others say it will not. I've seen other implementations of mail servers that create a "[EMAIL PROTECTED]" email address and have their users all send it to this, but I have not had time to install this on a test machine to see if they do it a different way or not.