Re: [exim] Cluster

Top Page
Delete this message
Reply to this message
Author: Ian Eiloart
Date:  
To: Mike Cardwell, exim-users
Subject: Re: [exim] Cluster


--On 13 August 2009 20:49:07 +0100 Mike Cardwell
<exim-users@???> wrote:

> pch0317 wrote:
>
>> Is exim capable to work in high-availability or load-balancing cluster?
>
> Yes.
>
>> If yes, where I can read about this. Please, give me a hint.
>
> 1.) Install Exim on each node in cluster
> 2.) Done.
>
> Any sort of clustering or load balancing would be done outside of Exim.
> high-availability is "built in" to SMTP by the ability to set multiple
> MX records, and the ability to retry.
>


Which is fine for MX servers. Not so good for MSA. MUAs (user clients)
don't use MX records, and don't retry even if the smtp server address is a
round robin in the DNS. At least, none of the servers that I've tried.

We have a cluster of four OSX Server servers with Exim. With three, Apple's
IP failover was just about managable. With four we switched to using
Wackamole on Spread. That works quite nicely, but occasionally a failover
throws up faults with ARP caches, leaving some of the IP addresses
unavailable.

As a result, we've recently put the MSA servers behind a CoyotePoint
equaliser. We publish a single IP address for MSA, and the equaliser does
load balancing and high availability. It regularly checks to see that it
can get an SMTP greeting from each of the servers - if not, then that
server doesn't get any traffic till its fixed.

Of course, the equaliser becomes a single point of failure... And so it
goes...

The equaliser doesn't fail as often as the ARP cache corruptions occur.

--
Ian Eiloart
IT Services, University of Sussex
01273-873148 x3148
For new support requests, see http://www.sussex.ac.uk/its/help/