sa-learn expiry problem

[Charles Farinella <cf...@appropriatesolutions.com>]

I'm getting this message when I run sa-learn -D:

"Can't use estimation method for expiry, something fishy, calculating
optimal atime delta (first pass)"

It seems to have just started this week.  It displays this message for
quite some time, then goes on normally.  

Does anyone know what would cause this, and how I can correct it? 
SpamAssassin version 2.63.

-- 
Charles Farinella 
Appropriate Solutions, Inc. (www.AppropriateSolutions.com)
cfarinella@AppropriateSolutions.com
603.924.6079

Re: sa-learn expiry problem

[Charles Farinella <cf...@appropriatesolutions.com>]

On Thu, 2005-10-13 at 10:46, Theo Van Dinter wrote:

> You can read about the expiry algorithm in the sa-learn docs. 
...snip...
> Hope this helps. :)

Yes, your explanation is very helpful, thank you.  As to the docs, it
would be better to type 'man sa-learn' on the mail server, than to do it
on my local machine where SpamAssassin isn't installed.  Duh!  :-)

Thanks again.

--charlie

-- 
Charles Farinella 
Appropriate Solutions, Inc. (www.AppropriateSolutions.com)
cfarinella@AppropriateSolutions.com
603.924.6079

Re: sa-learn expiry problem

[Theo Van Dinter <fe...@apache.org>]

On Thu, Oct 13, 2005 at 10:11:53AM -0400, Charles Farinella wrote:
> "Can't use estimation method for expiry, something fishy, calculating
> optimal atime delta (first pass)"
> 
> It seems to have just started this week.  It displays this message for
> quite some time, then goes on normally.  
> 
> Does anyone know what would cause this, and how I can correct it? 
> SpamAssassin version 2.63.

You can read about the expiry algorithm in the sa-learn docs.  The short
version is that expiry assumes the number of tokens to expire is roughly the
same from run to run.  It calculates how many tokens it needs to expire now,
and compares the situation to the previous run.  If they're similar, it'll
estimate the atime delta (tokens older than now-delta get expired) and just
start expiring.  If they're not similar (you get the above message) the new
run has to go through all the tokens to figure out the delta, then go through
and do the expiry, so it takes a lot longer.

As for correcting it ...  You can't.  If the runs are similar, they're
similar -- if they're not, they're not.  There's nothing really tunable
in that respect.

Hope this helps. :)

-- 
Randomly Generated Tagline:
Fifth Law of Applied Terror:
 	If you are given an open-book exam, you will forget your book.
 
 Corollary:
 	If you are given a take-home exam, you will forget where you live.