[ale] Spamassassin fun?

[Jason Day]

On Thu, Oct 23, 2003 at 11:30:52AM -0400, Robert L. Harris wrote:
> I'm running spamassassin at home for myself, my wife and a few friends.
> I'm the only one reporting to the baysean filter (sa-learn --spam
> through mutt).  I'm running spamd in daemon mode.

Actually, spamassassin has autolearn enabled by default, which will
automatically send high-scoring emails to the bayes learner.  Unless you
disabled this in the config file, your whole family *might* be learning
some messages as spam.

Having said that, depending on your setup, you might not be reporting
*any* messages to the bayes filter.  If you are running spamd as root
(which isn't recommended), then there is no problem (configuration-wise;
there is a problem security-wise :) ).  If you are running spamd as
non-root though, you have to jump through quite a few hoops to configure
everything.  I posted a somewhat long message about how I got it working
a while back.

The problem with the non-root spamd is that the spamd daemon cannot
change its euid to the uid of the user running spamc, so it needs to
keep its own separate config files, including the bayes files.  But when
you run sa-learn, you use *your* configuration and bayes files.

> Does anyone know, if I report a spam, will the daemon know it for
> everyone or just for my user?  If the latter is there an easy way to
> have my learned spam applied to my whole site?

Again, unless you've done some configuration magic, learning a spam will
only learn it for you.  It is possible to use a site-wide bayes filter,
but it is not recommended, because it won't work nearly as well.  The
reason it won't work as well is because the bayes learner just gathers
statistics, and the bayes filter applies those statistics.  Changing the
To: header also changes the statistics, and will ultimately make the
filter less effective.  Also, different people tend to get different
types of mail.  For instance, you might get email that is in general
more technical than your wife's.  A bayes database that is just for you
will be able to adapt to that, but if you and your wife share the same
database then the different types of email will "confuse" the bayes
engine.

I hope I've explained this enough to make some kind of sense.  Feel free
to email me for more information.  Also the spamassassin FAQ is very
helpful, as is the sa-talk mailing list.

HTH,
Jason
-- 
Jason Day                                       jasonday at
http://jasonday.home.att.net                    worldnet dot att dot net
 
"Of course I'm paranoid, everyone is trying to kill me."
    -- Weyoun-6, Star Trek: Deep Space 9