[Linux-cluster] corosync issue with two interface directives

[Florian Haas]

Ben, I'm afraid you're completely missing the distinction between
internal cluster communications (the "interface" definitions in
corosync.conf), and the clients' communications with networked cluster
resources.

On Mon, Feb 6, 2012 at 5:34 PM, Ben Shepherd <bshepherd at voxeo.com> wrote:
> Basically traffic of both types comes in from BOTH networks.
> We send the traffic to the VIP's on each network.
> These VIPS will be held by the Active server.
>
> Traffic will go to Server 1 on both Network1 and Network2.

When you say Network1 and Network2, does that mean two network
interfaces connected to two distinct subnets?

> If we lose either the interface to Network1 or the interface to Network2
> we need to fail over the VIP's to the other server.

That's what connectivity monitoring is for, which is a cluster
service. Corosync doesn't concern itself with that; Pacemaker will
manage it. The ocf:pacemaker:ping resource agent was designed for that
purpose.

> We cannot keep the VIP on the active server if 1 of the networks is not
> working as an entire service will go down.
>
> Yes I would prefer a single ring with 2 interfaces...that fails over if
> either interfaces reports a problem.

No you don't; you always want your cluster to communicate over as many
rings as possible. You want your cluster resource manager to fail over
if there is a problem on the upstream network.

I hope this helps. Try to think of cluster communications and cluster
resource management as two distinct layers in the stack.

Cheers,
Florian

-- 
Need help with High Availability?
http://www.hastexo.com/now