You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by "Frank M. Cook" <fc...@acsplus.com> on 2005/07/28 00:02:23 UTC

autolearn

I posted a message the other day asking why my spamd might backlog periodically and someone asked me if I could see from the log what was happening.  It started again today and I see something.  the last entry in the log when the jam occurred said

    autolearn=unavailable

when I stopped checking, cleared the spool, and restarted checking the log began to show

    autolearn=no

should I just set something in local.cf to turn off autolearning more completely so it knows not to try?  

would I be better off doing whatever is required to create a real autolearning system?  am I right in thinking that requires a database?

Frank M. Cook

Re: autolearn

Posted by Kai Schaetzl <ma...@conactive.com>.
Frank M. Cook wrote on Wed, 27 Jul 2005 22:34:02 -0500:

> is there a file I can check to see if it has become large?  better yet is 
> there some kind of routine to run to do the purging?

run sa-learn --force-expire

and see what it says. Be patient if it seems to hang.

Kai

-- 
Kai Schätzl, Berlin, Germany
Get your web at Conactive Internet Services: http://www.conactive.com
IE-Center: http://ie5.de & http://msie.winware.org




Re[4]: autolearn

Posted by Robert Menschel <Ro...@Menschel.net>.
Hello Frank,

Thursday, July 28, 2005, 2:09:52 PM, you wrote:

FMC> it's getting more complicated.  we didn't find any bayes
FMC> files at all. now I'm seeing some entries in the log saying
FMC> autolearn=no and others saying =unavailable.  they almost
FMC> alternate although perhaps the difference is a function of the
FMC> scan result.

autolearn=no guarantees you have bayes files somewhere.  Check your
local.cf files and user_prefs files and see if a special location or
name is specified.  Turn on debugging (should be a -D parameter to
your spamd system), and the debugging output should indicate what
Bayes files are being used.

Bob Menschel




Re: autolearn

Posted by "Frank M. Cook" <fc...@acsplus.com>.
<It doesn't mean disabled. The other one (unavailable) is more serious, >

we ran a session with -D and saw messages saying the DB wasn't installed.  it is.  we thought maybe the problem was an access list and added the local IP to the list in case the DB was being blocked but that didn't seem to help.  we've turned Bayes off for now and now we're getting autolearn=disabled.  we'll run this way probably over the weekend to confirm that Bayes is the cause of the blockage.

Frank M. Cook
Association Computer Services, Inc.
http://www.acsplus.com

Re: autolearn

Posted by Kai Schaetzl <ma...@conactive.com>.
Frank M. Cook wrote on Thu, 28 Jul 2005 16:09:52 -0500:

> now I'm seeing some entries in the log 
> saying autolearn=no

Frank, this just means that the score isn't in the range to autolearn it. 
It doesn't mean disabled. The other one (unavailable) is more serious, 
though ...

Kai

-- 
Kai Schätzl, Berlin, Germany
Get your web at Conactive Internet Services: http://www.conactive.com
IE-Center: http://ie5.de & http://msie.winware.org




Re: Re[2]: autolearn

Posted by "Frank M. Cook" <fc...@acsplus.com>.
it's getting more complicated.  we didn't find any bayes files at all. now I'm seeing some entries in the log saying autolearn=no and others saying =unavailable.  they almost alternate although perhaps the difference is a function of the scan result.

the bottom line problem remains that spamd has stopped keeping up.  I used to have my sending routine set to try spamd as many as three times before giving up.  last night I changed it to only two tries.  I think I'll change it now so if it doesn't get a valid return the first time it just gives up.


Frank M. Cook
Association Computer Services, Inc.
http://www.acsplus.com

Re[2]: autolearn

Posted by Robert Menschel <Ro...@Menschel.net>.
Hello Frank,

Wednesday, July 27, 2005, 8:34:02 PM, you wrote:

>> The Bayes system will occasionally determine that it needs to expire
>> old entries, to keep the database files within reasonable size, and
>> that expiration can take a long while.

FMC> is there a file I can check to see if it has become large?  better yet is
FMC> there some kind of routine to run to do the purging?  I could shut down
FMC> message checking to give an expiration routine time to do its thing.

The files are the bayes_* files, by default located in the user's
$HOME/.spamassassin directory (whatever user is doing the email check
at the time).

bayes_journal_max_size (default: 102400)
bayes_expiry_max_db_size (default: 150000)
bayes_auto_expire (default: 1)
bayes_learn_to_journal (default: 0)
are the local.cf or user_prefs parameters that affect bayes
expiration. See
http://spamassassin.apache.org/full/3.0.x/dist/doc/Mail_SpamAssassin_Conf.html#learning_options
for the 3.0.x application of these parameters.

See
http://spamassassin.apache.org/full/3.0.x/dist/doc/sa-learn.html#expiration
for a discussion of expiration.
> sa-learn --force-expire
will force an expiration run. So a script which will
> stop email filtering
> sa-learn --force-expire
> restart email filtering
will help if this is indeed your problem.

Bob Menschel





Re: autolearn

Posted by Loren Wilton <lw...@earthlink.net>.
> > The Bayes system will occasionally determine that it needs to expire
> > old entries, to keep the database files within reasonable size, and
> > that expiration can take a long while.
>
> is there a file I can check to see if it has become large?

Yes. I don't know quite how to do it, so someone else will chime in, I hope.

> better yet is
> there some kind of routine to run to do the purging?  I could shut down
> message checking to give an expiration routine time to do its thing.

You can turn off bayes_auto_expire in user_prefs, and then use a cron job to
run an expire every midnight or so.

        Loren


Re: autolearn

Posted by "Frank M. Cook" <fc...@acsplus.com>.
> The Bayes system will occasionally determine that it needs to expire
> old entries, to keep the database files within reasonable size, and
> that expiration can take a long while.

is there a file I can check to see if it has become large?  better yet is 
there some kind of routine to run to do the purging?  I could shut down 
message checking to give an expiration routine time to do its thing.

Frank M. Cook
Association Computer Services, Inc.
http://www.acsplus.com 


Re: autolearn

Posted by Robert Menschel <Ro...@Menschel.net>.
Hello Frank,

Wednesday, July 27, 2005, 3:02:23 PM, you wrote:

FMC> I posted a message the other day asking why my spamd might
FMC> backlog periodically and someone asked me if I could see from the
FMC> log what was happening.  It started again today and I see
FMC> something.  the last entry in the log when the jam occurred said
FMC>  
FMC>     autolearn=unavailable
FMC>  
FMC> when I stopped checking, cleared the spool, and restarted
FMC> checking the log began to show
FMC>  
FMC>     autolearn=no
FMC>  
FMC> should I just set something in local.cf to turn off
FMC> autolearning more completely so it knows not to try? 

Actually, that suggests your problem may be with an auto-expire
against the Bayes database, rather than auto-learn.

The Bayes system will occasionally determine that it needs to expire
old entries, to keep the database files within reasonable size, and
that expiration can take a long while.

I've not had the problem, so I can't suggest good solutions from
experience, but others here on the list can...

Bob Menschel