[horde] Run Horde on a cluster?

[Patrick Boutilier]

Kevin Konowalec wrote:
> I've currently got 12 front ends to horde.  The way I accomplish the 
> load balancing/round-robining is to put 2 boxes running openBSD in 
> front  of the cluster.  The boxes are set up to use CARP to listen for 
> the "webmail" address and then using pf it routes the incoming 
> connection requests to one of the 12 servers.  I use 2 openBSD boxes 
> because they dynamically keep state between them so if one fails the 
> other automatically takes over and the users are none the wiser.  It's 
> pretty slick.  The only problem I've encountered is that the CARP/pf 
> combination can't tell if the machines they are routing to are alive 
> (which can cause problems if they're sending users to a dead box).  I 
> got around this by keeping all the machines in the cluster in a table 
> and every minute I check to see which ones are alive and pop the ones 
> that are down out of the table.  If users are connected when those 
> machines go down PF will route them to the next machine when the one 
> they were on is popped out of the rotation.
> 
> In terms of states I use memcached... that way it doesn't matter which 
> machine a user is on - and it doesn't matter if in the middle of the 
> session they get reassigned to a different server... they never know the 
> difference.

You must be using VFS for sending of attachments? Otherwise they will 
notice a difference when they send a zero byte attachment. :-)


> 
> Kevin
<snip>
> --Horde mailing list - Join the hunt: http://horde.org/bounties/#horde
> Frequently Asked Questions: http://horde.org/faq/
> To unsubscribe, mail: horde-unsubscribe at lists.horde.org