[plug] Uninterupted service network

[Hook]

Grahame Bowland wrote:

On Saturday 12 October 2002 11:54, Jeff Williams wrote:
> I was wanting to setup up a network where:
>
> All of the servers (www, mail, sql) and their backups are machines on a
> 192.168.0.* network.
> All of the DNS entries for the servers point to the outside address of
> the router for the 192.168.0.* network, and the router
> uses NAT to send the data to the appropriate server.
> In the event of a server down the router routes data to the backup
> server, util the master comes back up again.
>
> Now, I don't have a problem with setting up the above, but I what
> happens if the router dies? I was wanting any failures to be covered
> without manual intervention, but I can't think how you can go to a
> backup router. Is there a solution, or am I heading in the wrong
direction?

Use two Ciscos with HSRP running and the load balancing code.

Or, use two Linux boxen running heartbeat for the external IP address.
Heartbeat can run over serial, UDP or multicast and I've used it
successfully. Route the traffic to the heartbeat address and put loadsharing
stuff into both the linux routers.

One gotcha with heartbeat: edit /etc/ha.d/resources/if-up or whatever and
make
sure it does "ifconfig eth0:0 0" rather than "ifconfig eth0:0 down", or your
eth0 will keep mysteriously vanishing :) This is fixed in the newer Debian
packages though.


Or you could download a package called 'keepalive.d" which, when coupled
with the LVS kernel patches effectively runs HSRP on 2 or more Linux servers
allowing failover.  It's *not* load-balancing though.

The Hooker