spamd support module

[Steve Freegard]

Ken A wrote:
> Wow. Nice work Steve! spamd already allows ~/.spamassassin configs, and 
> mysql backend for per user rules. How would you implement per user rules 
> scores differently, or why?

Well - the current code wouldn't work with the ~/.spamassassin configs 
currently as it always sends spamd whatever user is configured as 
$spamd_user in the .pm file.

For this to work properly in MailScanner it needs something like the 
following pseudocode:

if (number of recipients > 1) then
  if (number of recipient domains > 1) then
   spamd user = $spamd_user
  else
   spamd user = domain.com
  end if
else
  spamd user = user at domain.com
end if

This way - the code would send spamd a 'User:' header of the recipient 
e-mail address (if a single recipient) or the e-mail domain e.g. fsl.com 
if there are multiple recipients but all in a single domain otherwise it 
will send a default user if there are multiple recipients in mixed domains.

For this to work you'd need to configure spamd to use virtual config 
directories.

Training also becomes a bit tricky as the same process has to be run to 
work out which user to send to sa-learn.

I would probably suggest that per-user configs aren't really a good idea 
unless you really, really need them for a specific purpose.  It's 
probably best to only go down to the per-domain level by default.

The other way to do it would be to have some sort of 'rules' file to 
work out which user should be sent to spamd depending on the recipient 
of recipient domain etc.

Cheers,
Steve.