Pressing Questions

[codger <li...@pmbx.net>]

I've a some questions that I haven't felt were answered by the 
documentation I've read.

1. I'm concerned about the AutowhiteList function. It seems to work in 
the favor of spammers as I check all the mail marked as spam.

2. Do most people turn AutowhiteList off? I've heard the basic 
description of AutowhiteListing but still I can't understand the 
necessity?

3. Also, if I have spam that has a really negative autowhitelist  
score, how is the autowhitelisting affected when I run sa-learn  as I 
do routinely as my spam corpus grows?

4. Also should spamassassin NOT be called via spamd or CGPSA during an 
sa-learn session, or can that safely be done while spam is being 
processed by spamassassin?

Answers greatly appreciated!

Codger

Re: Pressing Questions

[Matt Kettler <mk...@comcast.net>]

At 01:37 PM 2/7/04 -0500, codger wrote:
>I've a some questions that I haven't felt were answered by the 
>documentation I've read.
>
>1. I'm concerned about the AutowhiteList function. It seems to work in the 
>favor of spammers as I check all the mail marked as spam.

First, before thinking of the AWL as working "in favor" of anything, I'd 
suggest reading this one:

http://wiki.spamassassin.org/w/AwlWrongWay

If you're having problems with mis-marked spam, then yes, you've got a 
problem. If it's deducting 5 points from a 20+ point spam, that's very normal.


>2. Do most people turn AutowhiteList off? I've heard the basic description 
>of AutowhiteListing but still I can't understand the necessity?

I have it disabled.

I would not regard it as "necessary" but it is certainly useful in many 
cases. It's a score averager, and as such, it's going to "smooth out the 
bumps" in scores.


>3. Also, if I have spam that has a really negative autowhitelist
>score, how is the autowhitelisting affected when I run sa-learn  as I do 
>routinely as my spam corpus grows?

The AWL is not impacted by sa-learn at all. However, I'd suggest seeing #1 
again.


>4. Also should spamassassin NOT be called via spamd or CGPSA during an 
>sa-learn session, or can that safely be done while spam is being processed 
>by spamassassin?

It's safe to call SA via any method you like during sa-learn runs. The 
bayes DB has lockfiles to prevent more than one process from writing it at 
a time, and most runs of SA don't need to write the bayes db at all.

However, if any other spamassassin process wants to do any opportunistic 
bayes updates such as autolearning, or expiry, they will fail to get the 
lock and the bayes update will be skipped. The same thing also happens when 
there's multiple calls to spamassassin at a given moment and more than one 
decides to do learning. This is quite normal and doesn't really cause any 
problems (and it's better than the alternative of blocking up the mail queue).





>Answers greatly appreciated!
>
>Codger