On 15/05/2013, at 6:44 PM, Ferenc Wagner <[email protected]> wrote:

> Andrew Beekhof <[email protected]> writes:
> 
>> On 15/05/2013, at 5:23 PM, Ferenc Wagner <[email protected]> wrote:
>> 
>>> If a resource fails to stop, the node hosting it is fenced immediately.
>>> Isn't it possible to move the other resources off the node before it's
>>> fenced?
>> 
>> Sometimes, but you're in trouble if another resource can't stop until
>> the failed one has (which doesn't occur until the node is fenced).
>> So there can easily be a dependancy loop there.
> 
> Sure, I'd fully expect the dependencies of the failed resource to fail
> to stop as well.  But trying to migrate resources away would hardly
> hurt,

Think about what you just said one line earlier.
You're extending the service outage every time a service fails to stop.

Additionally, there are resources we can't even try to stop - because the stop 
action depends on the fencing operation which depends on the stop operation.
Its a circular dependancy with no termination.

> apart from possibly wasting some time on unsuccessful migration
> attempts before the fencing.
> 
> Based on your reply I sense there's no such feature in Pacemaker.  Maybe
> it could be considered then for future development, as an option.
> -- 
> Regards,
> Feri.
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems

_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to