force-expire broke bayes? [Re: Bayes database size.]

[Jeff A. Earickson]

Are you running 4.26.8, maybe with Julian's patch to SA.pm from
a couple of days ago?  If so, then you *do not* want to do force-expire
via a cron job.  MS handles this internally in 4.26.8.

Jeff Earickson

On Thu, 5 Feb 2004, Eric Dantan Rzewnicki wrote:

> Date: Thu, 5 Feb 2004 13:01:00 -0500
> From: Eric Dantan Rzewnicki <rzewnickie at RFA.ORG>
> Reply-To: MailScanner mailing list <MAILSCANNER at JISCMAIL.AC.UK>
> To: MAILSCANNER at JISCMAIL.AC.UK
> Subject: force-expire broke bayes? [Re: Bayes database size.]
>
> I added --force-expire to my nightly 3am sa-learn cronjob. (previously
> done with only --rebuild). Since 3am no messages have bayes scores at
> all and it looks like autolearning is not working.
>
> Could I have done something stupid?
>
> -Eric Rz.
>
> On Thu, Feb 05, 2004 at 11:20:25AM -0500, Matt Kettler wrote:
> > At 08:40 AM 2/5/2004, Carles Xavier Munyoz Bald? wrote:
> > >My question is: will the learning process stop when there is enougth
> > >information on the database or will it continuosly learn new spam and ham
> > >messages ?
> >
> > It will keep learning.
> >
> > >That it is, is there any limit in the number of spam and ham messages
> > >learned
> > >by the Bayes database ?
> >
> > Yes, read man Mail::SpamAssassin::Conf if you want to try to change it.
> >
> >
> > >If there is no limit, will my database continuosly increase its size until
> > >I
> > >run out of disk space ?
> >
> > No it won't, as long as the expiry process can successfully run now and
> > again. The expiry pushes old tokens out of the bayes database if it's over
> > the size limits.
> >
> > If you're using an older version of MailScanner on a busy server, you may
> > need to run sa-learn --force-expire in your crontab. Newer versions of
> > MailScanner manage bayes expiry automatically.
> >
> > (SA will try to "opportunistically" run expiry as it scans mail, but on a
> > busy server, with multiple MailScanner children, it's unlikely to be
> > successful, as it can only succeed in locking the bayes database when only
> > one message is being SA'ed at the time it tries. Same rules of opportunism
> > apply to autolearning. It only happens if it can be done without waiting
> > for a lock.)
>