on Tue, Feb 07, 2006 at 01:29:58PM +0100, Joost Kraaijeveld (J.Kraaijeveld@Askesis.nl) wrote:
> Hi,
>
> I have two stock Debian 3.1 (Sarge) machines that seem to show a
> different behavior of spamassassin.
>
> The first machine has the following files in the bayes_path:
>
> -rw------- 1 amavis amavis 344064 2006-02-07 13:06 auto-whitelist
> -rw------- 1 amavis amavis 45056 2006-01-23 08:19 bayes_seen
> -rw------- 1 amavis amavis 659456 2006-01-23 08:19 bayes_toks
>
>
> The second machine has the following files in the bayes_path:
>
> -rw------- 1 amavis amavis 12288 Feb 7 12:54 auto-whitelist
> -rw------- 1 amavis amavis 10056 Feb 7 12:54 bayes_journal
> -rw------- 1 amavis amavis 81920 Feb 7 12:54 bayes_seen
> -rw------- 1 amavis amavis 1302528 Feb 7 12:54 bayes_toks
>
> The difference is that in the first directory 1 file (bayes_journal) is
> missing (compared to the second director) and that the files
> "bayes_seen" and "bayes_toks" are never updated.
Not familiar with specifics, but are these the same versions of
spamassassin?
'man sa-learn' suggests that you can sync (and remove) journal and
database files by running 'sa-learn --sync'. This would suggest you
don't necessarially have a problem here.
> I suspect that on the first machine spamassassin is not auto-learning.
> Is there a way to check or see if spamassassin is actualy using the
> bayes features?
Well, you can check your spam scoring summaries to see whether or not
any spam is being caught on Bayes rules.
You can also dump a given users' token database with:
sa-learn --dump
... which will list contents. Certainly _appears_ that there's
something in these files.
Note too that SA's bayesian filter may be a tad funky in multiuser
systems, another area I haven't entirely kept up with.
Peace.
--
Karsten M. Self <kmself@ix.netcom.com> http://kmself.home.netcom.com/
What Part of "Gestalt" don't you understand?
Play it once, Sam, for old time's sake.
- Casablanca
Attachment:
signature.asc
Description: Digital signature