On 4/6/2015 10:08 PM, Asai wrote: > Greetings, > > We've been using Amavis for a number of years, but it seems to not > be doing what we need it to be doing regarding spam filtering. > e.g. I can't seem to get it to learn bayes data on a per user > basis. We have our spam filters turned up so high ( kill level 3 > ) for some users it just seems like we're doing something wrong. > I have deleted the Bayes data before and let it rebuild, but it > doesn't seem to make sense that we've got all this crazy spam, > some of which gets caught and some doesn't even though it's the > exact same spam (or nearly the same). Is Amavis to blame here? > Does it get in the way of Spamassassin running as it should, or is > it more just configuration problems on our part? > > Thanks.
Amavis is normally configured for site-wide bayes as the "vscan" or "amavis" user, not per-user. If you're training bayes per-user, but SA is running with a site-wide bayes, you'll get poor results. This is operator error, not the fault of Amavis or SpamAssassin. You must train the proper bayes database. -- Noel Jones
