Yep, now it's working. Debian exim/sa-exim/spamassassin solution lets SA run as user Debian-exim. During some investigations, I found out that bayes wasn't running at all (due to unspecified db directory). At that time, I must have run spamassassin manually under root, thus creating one of the files in the db dir with root ownership.
Thanks again, Carsten. Sietse van Zanen wrote: > > Check if the user you are running spamassassin under has r/w access to the > files in /var/spool/exim4/.spamassassin/bayes > > It is also best if you run spamd with option '-u user' if you have a side > wide bayes. Otherwise all your local users + the user nobody need r/w > access. > > -Sietse > > > > > From: Carsten > Sent: Mon 18-Sep-06 16:55 > To: users@spamassassin.apache.org > Subject: Autolearn doesn't work > > > Hi, > > I have autlearn enabled and configured for auto-expire: > > use_bayes 1 > bayes_path /var/spool/exim4/.spamassassin/bayes > bayes_auto_learn 1 > bayes_auto_learn_threshold_spam 5.0 > bayes_auto_learn_threshold_nonspam -1.5 > bayes_min_ham_num 100 > bayes_min_spam_num 100 > bayes_auto_expire 1 > bayes_journal_max_size 204800 > bayes_expiry_max_db_size 300000 > > The following happens: > > 1. I receive a spam. Score is above threshold, but header says: > auto-learn=unavailable: > > X-Spam-Status: Yes, score=14.8 required=3.0 tests=FORGED_IMS_TAGS, > FORGED_MUA_IMS,HELO_DYNAMIC_IPADDR2,HTML_IMAGE_ONLY_04, > HTML_IMAGE_RATIO_02,HTML_MESSAGE,RCVD_IN_XBL,SPF_HELO_SOFTFAIL, > UNPARSEABLE_RELAY autolearn=unavailable version=3.1.4 > > 2. So I do spamassassin -D -t and check, but it says bayes db is ok and > autolearn spam: > > [15669] dbg: learn: auto-learn: currently using scoreset 3, recomputing > score based on scoreset 1 > [15669] dbg: learn: auto-learn: message score: 20.987, computed score for > autolearn: 14.766 > [15669] dbg: learn: auto-learn? ham=-1.5, spam=5, body-points=8.238, > head-points=8.378, learned-points=3 > [15669] dbg: learn: auto-learn? yes, spam (14.766 > 5) > [15669] dbg: learn: initializing learner > [15669] dbg: learn: learning spam > > 3. Although I would now expect that it has learnt this spam, a manual > call > of sa-learn --spam on that mail > reports it has learnt it. Don't know, whether it helps, but here a > sa-learn -dump magic: > > data:/var/spool/sa-exim/SAdevnull/new# sa-learn --dump magic > 0.000 0 3 0 non-token data: bayes db version > 0.000 0 1540 0 non-token data: nspam > 0.000 0 13874 0 non-token data: nham > 0.000 0 122142 0 non-token data: ntokens > 0.000 0 1155499496 0 non-token data: oldest atime > 0.000 0 1158589607 0 non-token data: newest atime > 0.000 0 1158590802 0 non-token data: last journal sync > atime > 0.000 0 1158342736 0 non-token data: last expiry atime > 0.000 0 2764800 0 non-token data: last expire atime > delta > 0.000 0 31397 0 non-token data: last expire > reduction count > > Thanks, > Carsten. > -- > View this message in context: > http://www.nabble.com/Autolearn-doesn%27t-work-tf2291791.html#a6365367 > Sent from the SpamAssassin - Users mailing list archive at Nabble.com. > > -- View this message in context: http://www.nabble.com/Autolearn-doesn%27t-work-tf2291791.html#a6384116 Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
