Re: Bayesian filtering not kicking in, but it's trained.
RinkWorks wrote: There must be a way to have spamd run in a way that it looks at each individual user's .spamassassin directory instead of the mail daemon user. I'd think that would be a common thing. But I can't figure out how to set it up that way. Whether you can do that depends on how you're calling SA. If you really want per-user preferences, SA should be called as one of the last steps in your mail processing. It sounds like you've got SA called as an MTA content filter from Exim, rather than from procmail or maildrop (or whatever else you might be using) just before delivery. You **MAY** be able to modify your existing call to SA enough to do what you want, but content filtering in the MTA is prone to conflicts between what different recipients want done. My own per-user setups call SA from individual .procmailrc files, at which time there is only one recipient for the message, and it's clear who that recipient is. -kgd
Re: Bayesian filtering not kicking in, but it's trained.
RinkWorks wrote: So it's a mystery, I guess, but case closed. But thank you very much for giving this matter your attention. I was wrong -- the case is still open. But I found out why Bayes wasn't working and then kicked in. Basically, I discovered that Spam Assassin wasn't paying attention to the whitelist_from statements in my user_prefs file. So I wondered if it was using a different .spamassassin directory somewhere. Sure enough, there's a /var/spool/exim4/.spamassassin directory. The reason why Bayesian filtering wasn't working, then suddenly kicked in, is because *THAT* director's bayesian filtering database hadn't gotten enough hams and spams yet, but eventually it autolearned enough of both to kick in. That directory is owned by the Debian-exim4 user, which is the user that owns the exim4 daemon process. However, the spamd processes are running as root. There must be a way to have spamd run in a way that it looks at each individual user's .spamassassin directory instead of the mail daemon user. I'd think that would be a common thing. But I can't figure out how to set it up that way. Anybody know? Just to reiterate from before, when /etc/init.d/spamassassin start runs, I get a process that looks like this: root 25165 0.0 1.3 32176 28780 ?SNs Sep07 0:05 /usr/sbin/spamd --create-prefs --max-children 5 --helper-home-dir -d --pidfile=/var/run/spamd.pid Thanks in advance. -- View this message in context: http://www.nabble.com/Bayesian-filtering-not-kicking-in%2C-but-it%27s-trained.-tf4384676.html#a12571789 Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayesian filtering not kicking in, but it's trained.
On Sat, 2007-09-08 at 10:25 -0700, RinkWorks wrote: Basically, I discovered that Spam Assassin wasn't paying attention to the whitelist_from statements in my user_prefs file. So I wondered if it was using a different .spamassassin directory somewhere. Sure enough, there's a /var/spool/exim4/.spamassassin directory. The reason why Bayesian filtering wasn't working, then suddenly kicked in, is because *THAT* director's bayesian filtering database hadn't gotten enough hams and spams yet, but eventually it autolearned enough of both to kick in. That directory is owned by the Debian-exim4 user, which is the user that owns the exim4 daemon process. However, the spamd processes are running as root. Sounds like the default installation of OS X Server. For that the fix is deleting one of the directories and putting in a link to the other one. Crude, but it works. -- Jerry Durand, Durand Interstellar, Inc. Los Gatos, California, USA, www.interstellar.com tel: +1.408.356.3886, USA: 866-356-3886, Skype: jerrydurand
Re: Bayesian filtering not kicking in, but it's trained.
On 05.09.07 08:28, RinkWorks wrote: I'm trying to run Spam Assassin 3.1.7 as root on a Linux machine (Debian Etch, Perl 5.8.8), with individual user Bayes databases. Everything seems to be working except that I'm getting no BAYES_* scores for anything. So, when reading mail for the 'ss1' user (which is me), I see lots of SpamAssassin headers but no BAYES scores. However, ~ss1/.spamassassin is populated with bayes_seen and bayes_toks (no bayes_journal), and I am able to run sa-learn as the 'ss1' user and see these files being updated with the new data. If I run spamassassin -D --lint as the 'ss1' user, grepping for bayes, I get this: [32082] dbg: config: read file /usr/share/spamassassin/23_bayes.cf [32082] dbg: bayes: tie-ing to DB file R/O /home/ss1/.spamassassin/bayes_toks [32082] dbg: bayes: tie-ing to DB file R/O /home/ss1/.spamassassin/bayes_seen [32082] dbg: bayes: found bayes db version 3 [32082] dbg: bayes: DB journal sync: last sync: 0 [32082] dbg: bayes: DB journal sync: last sync: 0 [32082] dbg: bayes: corpus size: nspam = 2655, nham = 786 [32082] dbg: bayes: score = 0.168968394084945 [32082] dbg: bayes: DB journal sync: last sync: 0 [32082] dbg: bayes: untie-ing [32082] dbg: bayes: untie-ing db_toks [32082] dbg: bayes: untie-ing db_seen This tells that spamassassin did check the bayes database and spam probability of scanned message is 0.168968394084945 which should be matched by BAYES_20. don't you have turned bayes filtering off somewhere? use_bayes_rules 0? As far as autolearn goes, some emails are autolearn=ham but the rest are autolearn=no -- I don't see that I'm getting anything being autolearned as spam, but maybe I haven't gotten anything recently that scored high enough for that. No idea if the data on the autolearned hams is actually making it to the right bayes database. Do you reject mails with score over some value? The defailt value for spam learning is 10, if you reject that mail, you'll never autolearn spam... -- Matus UHLAR - fantomas, [EMAIL PROTECTED] ; http://www.fantomas.sk/ Warning: I wish NOT to receive e-mail advertising to this address. Varovanie: na tuto adresu chcem NEDOSTAVAT akukolvek reklamnu postu. Fighting for peace is like fucking for virginity...
Re: Bayesian filtering not kicking in, but it's trained.
Matus UHLAR - fantomas wrote: don't you have turned bayes filtering off somewhere? use_bayes_rules 0? No. That's what's so confusing. But there's an update now. Apparently at some point yesterday, BAYES tests just suddenly started showing up. I wasn't doing anything at the time; it just suddenly started kicking in. That doesn't make a whole lot of sense to me unless I had *just* autolearned enough spams and hams for Bayesian filtering to take hold. But as I say, I was hundreds of hams and thousands of spams over the minimum long before that. So it's a mystery, I guess, but case closed. But thank you very much for giving this matter your attention. -- View this message in context: http://www.nabble.com/Bayesian-filtering-not-kicking-in%2C-but-it%27s-trained.-tf4384676.html#a12523761 Sent from the SpamAssassin - Users mailing list archive at Nabble.com.
Re: Bayesian filtering not kicking in, but it's trained.
Hi, RinkWorks wrote: Matus UHLAR - fantomas wrote: don't you have turned bayes filtering off somewhere? use_bayes_rules 0? No. That's what's so confusing. But there's an update now. Apparently at some point yesterday, BAYES tests just suddenly started showing up. I wasn't doing anything at the time; it just suddenly started kicking in. That doesn't make a whole lot of sense to me unless I had *just* autolearned enough spams and hams for Bayesian filtering to take hold. But as I say, I was hundreds of hams and thousands of spams over the minimum long before that. So it's a mystery, I guess, but case closed. But thank you very much for giving this matter your attention. To me this sounds like the Bayes database you are looking at when you check the number of learnt messages is not the same one used when scanning emails. Are you running the checks on the Bayes database as the same user that SA runs as normally? -- Anthony Peacock CHIME, Royal Free University College Medical School WWW:http://www.chime.ucl.ac.uk/~rmhiajp/ A CAT scan should take less time than a PET scan. For a CAT scan, they're only looking for one thing, whereas a PET scan could result in a lot of things.- Carl Princi, 2002/07/19
Re: Bayesian filtering not kicking in, but it's trained.
RinkWorks wrote: I'm trying to run Spam Assassin 3.1.7 as root Let me stop you right there. You cannot run spamd as root. It drops privs, and runs as user nobody. /usr/sbin/spamd --create-prefs --max-children 5 --helper-home-dir -d --pidfile=/var/run/spamd.pid It would be best to create a spamd user, and start with this: /usr/sbin/spamd --create-prefs --username=spamd --max-children 5 --helper-home-dir -d --pidfile=/var/run/spamd.pid You can specify a bayes_path in your config, and run sa-learn as root if you'd like. -=Aubrey=-
Re: Bayesian filtering not kicking in, but it's trained.
maillist wrote: RinkWorks wrote: I'm trying to run Spam Assassin 3.1.7 as root Let me stop you right there. You cannot run spamd as root. It drops privs, and runs as user nobody. Not quite correct... spamd will drop privs to nobody *for that call* if spamc is run by root without -u non-root-username. Otherwise, per-user configs using system users don't work, because if spamd doesn't run as root, it can't fork and drop priviledges (or whatever the exact process is; IIRC it changed a while back) to the calling user. I'm happily running 3.1.9's spamd as root, calling spamc from individual .procmailrc files on several systems. (I have had to switch to calling spamassassin for root's mail filtering, however.) -kgd
Re: Bayesian filtering not kicking in, but it's trained.
On 05.09.07 08:28, RinkWorks wrote: Subject: Bayesian filtering not kicking in, but it's trained. is it trained with enough of spams and hams? -- Matus UHLAR - fantomas, [EMAIL PROTECTED] ; http://www.fantomas.sk/ Warning: I wish NOT to receive e-mail advertising to this address. Varovanie: na tuto adresu chcem NEDOSTAVAT akukolvek reklamnu postu. WinError #98652: Operation completed successfully.
Re: Bayesian filtering not kicking in, but it's trained.
Matus UHLAR - fantomas wrote: is it trained with enough of spams and hams? Yes. I've got the defaults of 200 hams and 200 spams required, and as you can see from the -D output, I've got 2655 spams and 786 hams that it currently knows about in the ss1 user's bayes data files. -- View this message in context: http://www.nabble.com/Bayesian-filtering-not-kicking-in%2C-but-it%27s-trained.-tf4384676.html#a12510206 Sent from the SpamAssassin - Users mailing list archive at Nabble.com.