Bayes database management when using a milter

[Jeff Makey <je...@sdsc.edu>]

I am using SpamAssassin 2.63 with MIMEDefang 2.39 and Sendmail 8.12.9
on a Solaris 8 system to block spam during SMTP sessions.  I want to
have the milter use per-user Bayes databases while allowing the users
to train them with sa-learn, but I am finding it difficult to keep the
milter from writing to the databases in a manner that does not
generate error messages.  The same issue exists when simply using
spamc/spamd.

My ~/.spamassassin directory looks like this:

 drwxrwsr-x   2 jeff     defang       512 Jan 28 16:29 .
 drwxr-sr-x  10 jeff     staff       1024 Jan 27 19:06 ..
 -rw-------   1 defang   defang     17067 Jan 28 16:29 bayes_journal
 -rw-rw----   1 jeff     defang     90112 Jan 28 16:29 bayes_seen
 -rw-rw----   1 jeff     defang   1302528 Jan 28 16:29 bayes_toks

to allow the milter to work cleanly, but because of the journal file
permissions I can't use sa-learn properly.  My goal is to have
SpamAssassin (when running in the milter context) never attempt to
update the Bayes database, but still use it when checking messages.

It's pretty clear that letting the milter have write access to each
user's database is incompatible with allowing each user to do training
with sa-learn, but I am having difficulty tracking down all of the
places that want to update the Bayes database even when
bayes_auto_expire and bayes_auto_learn are both 0.  If anyone has
solved this problem I would love to know how it was done.

                          :: Jeff Makey
                             jeff@sdsc.edu

SA:RBL

["Gregory Sloop, Sloop Network & Computer Consulting" <ls...@sloop.net>]

Newbie question here...

(In fact, if someone will give me the right answers, I'll try to do the wiki
on this subject.)

I setup SA in the last day or so.

Initially, none of the spams were getting flagged by the RBL tests. I wanted
to do some testing to make sure the RBL system was functioning right, but
couldn't find any deterministic way to do so.

I did run spamassassin -D (with debug info) and could see it was looking up
the MX record, so I was pretty sure Net::DNS was installed and functioning
correctly.

Later in the day, the RBL test just "magically" started showing up on Spam I
was receiving.

Why? I have no idea, and I'm not going to waste any time trying to figure
out why.

The bigger question is: How can one test the RBL portions of SA
*deterministically* (as opposed to waiting for a likely spam to happen
along...)
(I'm doing 3-4 site wide installs soon, and a "compile and hope" strategy
for RBL doesn't tickle me fancy.)

*** Summary:
How would one notice or find errors from the RBL portion of the tests - it
seems the debug portion of the code isn't very clear about this...

(I assume this would also apply to DCC, Pyzor and Razor, though, I'm not
using these systems, and they seem too complex and hairy to offer much
additional benefit. I'm using SA, bayes, and RBL and am getting very good
results.)

Thanks for your time!

Oh, I almost forgot! Thanks SA crew! This program rocks! (I used to use SA
under v 2.31 but ran into some problems, and have just come back.)
BTW, is there somewhere I can "donate" to the SA project? (Perhaps I missed
it somewhere...)

Thanks again,
Greg

=================================================================
Sloop Network & Computer Consulting
Gregory Sloop, Principal
networkguru@sloop.net / www.sloop.net
PO Box 16990 Portland OR 97292
v. 503.251.0452
================================================================