Rolf Habing <r.j.habing@???> probably said:
> The problem I encounter is when people on server 2 want to mail server 1
> people. The obvious thing to do is probably a table of which user is on
> what server, and route accordingly. However, I'd like to get away from
> large lists as much as possible and am looking for a listless solution.
> (Each server hosts +10,000 users.)
>
> Introducing an unknown transport on server 2 back to server 1 would
> create a mail loop.
The problem with doing this without some kind of lookup is that if
server 1 goes down, server 2 doesn't get any mail.
The traditional way I solved this problam was to make the domain that
was being shared a virtual domain, split the MX between machines (or
preferably between a pair of relay hosts) so that if one machine
goes down everything else can still get mail.
Building the user lists can be automated and distributed fairly easily.
If you really don't want to build users lists (which would be the
fastest and most efficient way, IMO) you could implement some kind of
lookup (perhaps LDAP) on each server to give a response over the
network to the other machine if a user exists on the other machine,
and call that lookup as part of the decision on where to send or
rewrite that address.
I've always found splitting a local domain delivery more hassle than
it was worth, and it's ended up virtual and delivery on each machine
as machineN.dom.ain pretty quickly - easier, more reliable and scales
better IMO.
P.
--
pir pir@??? pir@???
