Re: [exim] Possible Spamassassin related feature request

On 22/06/11 00:10, Marc Perkel wrote:
> Just trying the idea here before I ask for a feature request.
>
> As we all know Exim has a spamassassin interface for testing email to
> see if it's spam. I'd like to see an interface added for learning email
> as spam/ham from exim. In my configuration the Exim server are separate
> from the SA servers and there is yet another common server for MySQL.
>
> Thoughts? Does this make sense?

This is already easy to do in a number of ways. Pipe transport, or
process a dropbox mailbox, etc. I believe I'm thinking a little
differently than you though. I required users that received spam or
misclassified ham to report it to an appropriate email address.

It requires protection of the report-spam@ and report-ham@ email
addresses though as if a spammer finds out that you have such and
address, they can flood it with spam (since it has to accept everything
regardless of its spam/ham) and ruin the training. When I did do this,
it was restricted to computers inside the network and required human
intervention ie. a person went through the list of emails reported and
double checked that no one had been an idiot. The forwarded emails were
not used to train, but copies of the original messages from the archived
flow. Subject, from and to extracted from the forwarded message were
used to locate the message in question.

I don't run that system any more since the bayes database ended up being
next to useless on that mail stream.

If you are deciding what you want to learn as ham/spam during the
delivery of emails, the dropbox & process method would suffice. You
would have the original emails, possibly in nice Maildir format (for
dropboxes for spam, I use file mode with the $message_exim_id as the
filename), and all the time in the world to send them to the SA servers.
Perhaps when the mail load is low enough that SA wont mind that the
bayes DB is locked and unavailable.