Re: [exim] Backup email server

Página Principal
Apagar esta mensagem
Responder a esta mensagem
Autor: Ian Eiloart
Data:  
Para: Jeroen van Aart, exim-users
Assunto: Re: [exim] Backup email server
Please don't hi-jack threads.

--On 31 July 2007 17:55:22 -0700 Jeroen van Aart <kroshka@???> wrote:

> Hello,
>
> I was wondering if anyone can point me to information how to set up a
> backup email server. In such a way that if the main server is down one
> can switch to the backup with no to minimal loss of email (connectivity).


If you're only bothered about incoming email, then set up a second server, 
identical to the first, except:
    a) do call-forwards to the first server to verify that the recipient 
exists.
    b) deliver local domains to the first server.
    c) give it a lower priority MX record - or equal priority if you want 
to load balance.


However, this won't help local mail submitters, and that's arguably more
important since MTAs usually handle retries better than MUAs.

None of the email clients (MUAs) that I've tested will try a second IP
address when they can't submit email. For higher availability of Message
Submission function, you need to arrange IP failover. Debian might support
some mechanism for that, but you might also look at a combination of Spread
and Wackamole, which I've found easier to configure, and more reliable,
than Apple's failover solution on OSX boxes - particularly when you get
beyond two hosts.

    <http://spread.org/> (last time I checked, you need version 3.x for 
wackamole. I'm using 3.17.03 Built 15/October/2004


    <http://www.backhand.org/wackamole/> Note that wackamole has a 
mechanism for running perl scripts when it detects loss or gain of an IP 
address - I use that to start/stop  IP address specific exim configurations 
(using "-DIPADDRESS=139.184.14.94", for example to persuade a new daemon to 
listen on that IP address), but the script running mechanism is a bit 
buggy. I'm using Version 2.1.2 Released 2004-09-07, but there's a later 
version Version 2.1.3 which supposedly fixes those bugs (but might break my 
workarounds :)



> I am thinking of a solution where if one server goes down (say an hd
> crash) the other server is used automatically. You could create a dns
> entry to randomly use one or the other IP, and mirror both systems. The
> central problem I believe is how to keep data like mailboxes in sync
> between 2 or more servers. If you'd use an NFS mounted server you create
> NFS related problems and an additional problem, of the NFS server
> possibly going down.
>
> (apologies for sending this mistakingly to a single person)
>
> Thank you,
> Jeroen




--
Ian Eiloart
IT Services, University of Sussex
x3148