[Linux-cluster] Few queries about fence working

[Digimer]

On 01/26/2012 08:29 AM, emmanuel segura wrote:
> I think your problem it's in your rhedhat cluster fencing, because in my
> current job i use SAN and we have the some problem, the only workaround
> it's fence delay in redhat cluster fencing agent

It's not related to Red Hat so much as the fence device. Both nodes
begin the process of fencing their peer before either node actually
dies. With ACPI enabled, the call to power down a node begins an orderly
shutdown. This allows the node's own fence call to fire off a shutdown
of the peer node before it dies (~4 seconds between initial shut down
call and node death).

Disabling ACPI should help in that the node should instantly power down
when the first call to shut down comes in (moral equivalent to pressing
and holding the power button). With ACPI, you have to hold the power
button for 4 seconds before power off. Without ACPI, the node should
instantly power off.

-- 
Digimer
E-Mail:              digimer at alteeve.com
Papers and Projects: https://alteeve.com