[Linux-cluster] power controller is interal/loss of pwer prevents failover: any ideas

[Lon Hohberger]

On Mon, 2006-09-11 at 17:31 +0200, Riaan van Niekerk wrote:
> Lon Hohberger wrote:
> > On Fri, 2006-09-08 at 11:16 -0700, Rick Rodgers wrote:
> >> Does anyone know of a good solution to providing good failover 
> >> for somthing like a Dell 1850? The issue here is that the power 
> >> souce plug in the back provides power for both the internal power
> >> controller
> >> and the node itself. So if you pull the cord it will not failover
> >> because
> >> it can not Stonith the failed node (power controller is down also).
> > 
> > Generally, you can't handle this without external fencing.
> > 
> > https://www.redhat.com/archives/linux-cluster/2006-September/msg00026.html
> > 
> > -- Lon
> > 
> 
> Lon - having reread that previous posting of yours, and esp the last 
> paragraph:
> 
> +++
> (c) ... If a host does a "graceful shutdown" when
> you fence it via IPMI, you need to disable ACPI on that host (e.g. boot
> with acpi=off).  The server should turn off immediately (or within 4-5
> seconds, like when holding an ATX power button in to force a machine
> off).
> ++++
> 
> Just so I am absolutely sure about this: Is the above the only scenario 
> when would have to disable ACPI? e.g. a graceful shutdown is easy to 
> spot. If I don't see one in the logs, that means I can leave ACPI on?

Basically, yes.

If you want to be sure, watch the machine's console while you perform a
power off using the integrated power management.  If the machine shuts
off immediately (while ACPI is enabled) then leaving it enabled should
not cause any problems with the cluster.

Note: Setting acpid to do /sbin/poweroff or its likeness does not count
as an "instant off"...  Don't cheat :)

-- Lon