On Tue, 6 Nov 2001, Odhiambo Washington wrote:
> The only person who accesses mail locally is me. The rest use POP3/IMAP
> access.
So you can switch the physical machine in case of the other's hardware
failing. Look: You have, say, two machines. A being your primary MX and
B being your secondary MX. Each of those has a primary IP that is not
subject to change. Additionally, there are two IPs - one for your POP3
and one for your imap server. These two IPs are logical interfaces that,
in case of one machine going down, will be set up on the other machine.
> | If the latter is the case, you have the chance to switch over the IP
> | and/or the hostname of your POP/IMAP server to the fallback MX. Users
> | then can still access new email until the primary host gets available
> | again.
>
> Hmm, I don't understand how that would happen when no deliveries are done
> on the fallback MX. It doesn't have users accounts/mailboxes on it. In fact
> it's my hope that it will just hold the mail in the queue until the main
> server comes up, no?
Well, you might opt for that, but it would have nothing to do with
HA. In fact, a delivery failure may in some cases even be better than
making people believe their mail has been received and noticed. Can you
guarantee n hours to get a failed host running again?
> | The setup starts to become interesting as soon if you implement a third
> | machine to hold copies for later synchronization.
> Someone is doing this already?? I have 3 servers and if that is a way with
> some HOWTO then I'd be willing to give it a try.
First, you can also run this with two hosts. You can make exim sending
copies of mail to your secondary MX during normal operation. In that case
users will have access to all their mail in case servers have to switch
over. It just requires a script to delete mails older n days.
If you get this running, you'll notice that a failover results in a
certain amount of additional work after services can be switched back to
the primary host and users start to ask for their received and sent mails
from the last ours or days. At that point, and this is what I meant above,
you implement a third host that all the time receives mails. On that host
C, there is no need for usual local deliveries. Mails are just needed for
keeping a record of mailflow and having copies available in order to
synchronize A and B. Sent mails, if stored on the server side, still have
to be synchronized between A and B.
> | I´m sorry. "HA" stands for "High Availability". Generally achieved by a
> You have some pointers on the web that I can look at???
This is going to be off-topic. I recommend you ask on a list that covers
your operating system. HA is a very wide area, and certainly nothing you
set up one weekend. I have done a very special configuration on HP-UX and
FreeBSD. The frustrating part is that the HP part is never failing... ;)
On the other hand, mail availability is a very hot topic and exim in my
opinion comes with functionality that makes it superior to other MTAs in
that field. Maybe someone should set up a couple of html pages on this
one, maybe even under exim.org..?
regards,
Volker
--
V. T. Mueller UCC Freiburg, Germany vtmue (at) uni-freiburg.de
"problems are just opportunities in work cloth"
