On Fri, Apr 26, 2002 at 07:32:25AM +0200, Birger Toedtmann wrote:
> Mmmhhh.  Isn't this a very dangerous thing to do to the cyrus system?
> Not the idea as such but their implcations:  I imagine a slow/broken DNS
> and 100+ incoming mails/min.  I imagine a system with thousands of half-
> hanging cyrus processes that try to do some name/ip/rbl lookups.  

Not sure. Is this fundamentally different from the LMTP client being
sluggish?  Does Cyrus deal with that well?

I suppose this is something I could test. Set up an IP address that
blackholes DNS, run one of those mail benchmark things. I need the code
actually written first, but...well, trying to get that. ;)

> Seems to me it would be better to have the MTA mark the mails with some
> additional header on conditions described above by you.  You publicise
> these headers to your users and they may filter on the header content,
> not sieve-DNS-lookups.....

To summarize my objections:

- it's not possible to automatically determine what's available for this,
  for like a wizard to set up UCE stuff. the Sieve extension should make that
  possible.

- users don't have as much control. I.e., they can't say "I want to use
  some.random.rbl.domain" unless the admin has configured for it.

> Alright the slow-DNS-problem applies to MTAs as well - but they are _built_
> for this purpose, cyrus is not, is it?

Hmm. What's different about them to make that true?

Reply via email to