|
On Fri, Feb 07, 2003 at 03:14:09PM -0600, Nate Carlson wrote:
> Not sure if this question belongs here or on the heartbeat list, let me
> know if I should go there. :)
>
> Like I mentioned in my last message, I've got a working LVS+Heartbeat load
> balancing solution set up. If I shut down heartbeat on the main box, all
> the services fail over to the second box properly. Ditto if I yank the
> heartbeat channels (ethernet+serial). Problem is, if I pull the external
> ethernet interface on the active balancer, the services don't fail over to
> the other one. This makes sense, as the boxes still see each other over
> the serial heartbeat. But since the external interface is down, the world
> can't reach the box.
>
> The only solution I've thought of for this would be to write a script that
> checks if it can ping an upstream IP, and if it can't, kill heartbeat.
> (And restart it when it's able to ping the upstream again). With this
> running on both boxes, I think I'd get the results I want.
>
> Has anyone ever tried something like this before? What have other people
> done for this case?
In more recent versions of heartbeat this is addressed using
the ping_node directive and ip_fail helper daemon. I would
suggest giving this a go. It works my monitoring access
to some external IP - say that of the switch or gateway -
and using that to influence decisions on weather to perform
takeover or not.
--
Horms
|
