SA bayes not working / autolearn inactive?

[Glenn Steen]

On 23/08/06, Daniel Maher <daniel.maher at ubisoft.com> wrote:
>
>
> > -----Original Message-----
> > From: mailscanner-bounces at lists.mailscanner.info [mailto:mailscanner-
> > bounces at lists.mailscanner.info] On Behalf Of Green, Rodney
> > Sent: August 22, 2006 12:34 PM
> > To: MailScanner discussion
> > Subject: Re: SA bayes not working / autolearn inactive?
> >
> > I didn't see anything in your original posting saying that you populated
> > the bayes db to try it. Also, the lint output you posted says that it
> > does not meet the threshold requirement.
> >
>
> Thanks for the reply,
>
> I have since populated the bayes DB (via sa-learn backup/restore) from one of the functioning mail servers.  A lint check now shows this:
> ---
> [19182] dbg: bayes: DB journal sync: last sync: 1156345626
> [19182] dbg: bayes: corpus size: nspam = 5507601, nham = 1761804
> ---
>
> As before, my learning threshold config line is:
> ---
> bayes_auto_learn_threshold_spam         12
> ---
>
> Take, for example, the following mail which passed through a few moments ago:
> ---
> Aug 23 15:14:05 yosemite MailScanner[18668]: Message 593C069B0A.6BE83 from 222.35.236.81 (offjskcjacpiz at yahoo.co.kr) to ubisoft.com is spam, SpamAssassin (score=31.401, required 6, DATE_IN_FUTURE_96_XX 1.89, DATE_SPAMWARE_Y2K 1.82, FORGED_IMS_HTML 2.27, FORGED_IMS_TAGS 2.12, FORGED_MUA_IMS 1.20, FROM_ILLEGAL_CHARS 3.28, FROM_LOCAL_NOVOWEL 2.33, HTML_MESSAGE 0.50, HTML_MIME_NO_HTML_TAG 0.51, MIME_BOUND_DD_DIGITS 3.60, MIME_HTML_ONLY 0.00, MIME_HTML_ONLY_MULTI 0.00, MISSING_MIMEOLE 1.39, MPART_ALT_DIFF 0.14, MSGID_SPAM_CAPS 3.52, RCVD_NUMERIC_HELO 1.25, REPTO_QUOTE_IMS 0.00, SARE_RECV_IP_222032 2.22, SUBJ_ILLEGAL_CHARS 3.36, UNPARSEABLE_RELAY 0.00)
> ---
>
> Clearly, this email meets the threshold for autolearning; however, there is no autolearn string to be found, and the access times on the bayes database do /not/ change.
>
> Any further input is more than welcome.
>
I think you're "using the wrong database" (user) for the lint, or the
postfix user cannot write to the file(s). Do:
su - postfix -s /bin/bash
Then at that prompt, rerun the spamassassin lint. Any problems? Same
is very likely the case for AWL.
Fix is to see that the ownership/permissions are correct on the files
as such, as well as checking/changing the appropriate *Permissions
(Incoming and Quarantine, IIRC) are set correctly in
MailScanner.conf... And (probably best... mandatory if you use
MailWatch):
bayes_path /etc/MailScanner/bayes/bayes
bayes_file_mode 0770
... or similar in /etc/mail/spamassassin/mailscanner.cf (which is a
link to spam.assassin.prefs.conf).

-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se