local.cf MySql bayes settings

[Ed Kasky <ed...@esson.net>]

I have been toying with converting my bayes db to MySql.  I am curious as 
to what most folks use for local.cf settings.  Anything that might help 
performance, etc??

Ed Kasky
~~~~~~~~~
Randomly Generated Quote (44 of 477):
"Discourage litigation. Persuade your neighbors to compromise
whenever you can. As a peacemaker the lawyer has superior
opportunity of being a good man. There will still be business
enough."   --Abraham Lincoln

Re: AWL Question

[Matt Kettler <mk...@evi-inc.com>]

First, the AWL isn't a whitelist. Period. It's a score averager.

http://wiki.apache.org/spamassassin/AutoWhitelist



> I have a particular address whitelisted via "spamassassin
> --add-addr-to-whitelist=user@address.com"

No, that command does not whitelist a sender.

It adds ONE message scored at -100 to the sender's averages. It's intended to
correct minor imbalances in the AWL system, and is not an effective method of
whitelisting email. Any single message score winds up lost in the surf of
averages over time.

If that sender has sent 100 emails, this command only gives them a -1.0
advantage, for example.


> I thought AWL assigns a score of -100;

It does not.

If you want to assign a score of -100 to a particular sender you must use
whitelist_from or whitelist_from_rcvd.

Note that these commands have nothing at all to do with the AWL.

Re: AWL Question

[Kris Deugau <kd...@vianet.ca>]

Richard Ozer wrote:
> I have a particular address whitelisted via
> "spamassassin --add-addr-to-whitelist=user@address.com"
> 
> It works for a while, but then comes back reported as spam after a
> week or two.

This is "normal" due to the way the AWL works.

> I recieve mail from this user, tagged with AWL, as well as BAYES40.
> However there are a couple of spammy rules fired that seem to
> overwhelm the negative scores from AWL and BAYES40.

Feed the message to sa-learn as ham.  Test it again;  it should now show
up with BAYES_00.  Do this often enough, and mail from this person
should always show up as BAYES_00.

Unfortunately this isn't always enough to get some people's mail
through.  :(

If someone's mail is hitting high-scoring rules, you can also add
entries in your user_prefs or global config to drop the score on those
rules.  I've had to do that once or twice.

> I thought AWL assigns a score of -100; or will "blacklist" entries
> continue to build up until the whitelist score is negated?

Not exactly.  Your initial call to add the address to your AWL set the
AWL score to -100...  but the AWL is a long-term score averager, and so
the AWL's effect will reduce over time.

The first message that came in from this person after you did
--add-addr-to-whitelist would have had ~ -50 for the AWL score (assuming
the message scored close to 0 without).  I'm not certain on the exact
change, but the next message would see somewhere between -25 and -33. 
Over time, the AWL adjustment will tend towards 0, only showing up on
mails that are either much more spammy or much less spammy that "usual"
for that person.

If this is someone that regularly sends you spammy mail, feed their mail
to Bayes as ham, and add a whitelist_from_rcvd entry for them.  (Only
use whitelist_from if they regularly send email from a wide variety of
places.)

> I thought I understood whitelisting until this situation arose...

"Regular" whitelisting uses whitelist_from and whitelist_from_rcvd in
either your ~/.spamassassin/user_prefs file or your main SA config files
(usually /etc/mail/spamassassin/*.cf).  They assign static negative
scores that don't change the way the AWL scores do.

See http://wiki.apache.org/spamassassin/AutoWhitelist for more detailed
information on the AWL.

-kgd
-- 
Get your mouse off of there!  You don't know where that email has been!

AWL Question

[Richard Ozer <ro...@ois-online.com>]

I have a particular address whitelisted via 
"spamassassin --add-addr-to-whitelist=user@address.com"

It works for a while, but then comes back reported as spam after a week or 
two.

I recieve mail from this user, tagged with AWL, as well as BAYES40.  However 
there are a couple of spammy rules fired that seem to overwhelm the negative 
scores from AWL and BAYES40.

I thought AWL assigns a score of -100; or will "blacklist" entries continue 
to build up until the whitelist score is negated?

I thought I understood whitelisting until this situation arose...

RO