Hello,
On Mittwoch, 15. September 2004, Stephen Gran wrote:
> On Wed, Sep 15, 2004 at 11:01:54PM +0200, V. T. Mueller, Continum said:
> > Can fallback_hosts be combined with a different interface option
> > in the same transport?
> I think you do want two routers, but a single transport. I'm assuming
> you have something like a LAN behind machines that also have public
> addressing, and what you want to do is send to another machine on the
> LAN when the route to the internet is down. Maybe I'm misreading you,
> though. If this is the case, use ignore_targtet_hosts for your LAN
> addresses, and put them in a LAN router, and then use the LAN address
> for the fallback host. Or was that misunderstood, and a waste of
> advice? On rereading your message, it seems like you may also be doing
> bonding or something.
It´s different and will probably look a bit weird. All hosts live in
a somewhat complicated dynamic routing setup, which is wanted and
desired by design for 'higher reasons'.
Well, I have to think about a simple real world szenario... Ok...
Imagine three networks. Two are official, the third is private. One of
the former two uses VLAN and hosts in it can´t communicate directly.
So a host in the first net could use the third net to reach a host that
sits on an official IP without VLAN constraints, which in turn is able
to reach any host in the other official network (and has no path to the
private net in this case).
The option with two routers does work, I was just thinking about the
other attempt being the more cleaner approach. I admit that this is
of cosmetical nature rather than being a problem. Nevertheless, it's
the purpose of test designs to deal with design questions on that level
so that real problems are prevented in production use later. One effect
would be on statistics, depending on whether they need to be implemented
in routers or in transports - to give just one example.
Having the original problem going round and round my brain, I think it
would actually be sufficient if exim would "forget" about the interface
setting for the transport once the fallback_hosts option jumps into
action. Will have to read the code tomorrow if that can be built in
and how exim reacts upon EHOSTUNREACH. Too late now...
Kind regards,
vt
--
V. T. Mueller
Continum AG
Wentzinger Strasse 7a
79106 Freiburg i. Br.
http://www.continum.net
Tel.: +49 761 479409 70
Fax.: +49 761 479409 33
Mail: v.t.mueller@???
