Bayes not being used

[Steve Thomas]

MailScanner 4.25-14
RedHat Linux 7.3
kernel 2.4.20-18.7
SA 2.70-cvs (1.222-2003-12-17-exp)

We've been running MS/SA for quite a while now, but without using bayes. Our system stores all its user info in LDAP and I don't want to use SQL prefs, so we've just been using the stock SA config with some extra rules added. It works decent, but not having bayes running is taking its toll. On my home server, bayes catches pretty much all the mail that normal SA tests miss, but here at work I'm still getting a thousand or so spams a day getting through to my users.

I created two shared folders for the users to put their spam/ham into and every hour, a cron job trains bayes with them and then deletes them. I make sure to use the prefs-file= option with sa-learn, pointing it to spam.assassin.prefs.conf. If I do a sa-learn --dump magic, I can see that over 2700 spams have been learned from and over 500 hams - more than enough for SA to start using bayes during processing.

If I run a message through spamassassin from the command line (in debug mode, again using the prefs-file option), I can see that it's using bayes, so SA is doing what it should be.

MS doesn't seem to use it, however. No BAYES_* tests are showing up in the headers of any messages. I've linted the SA config and it's fine, and I've restarted MS manually a couple of times.

Any ideas?

TIA,
Steve


--
"If a man does his best, what else is there?"
- General George S. Patton (1885-1945)