[Linux-HA] Question regarding the stonith resources

[Andrew Beekhof]

On 10/31/05, Peter Kruse <pk at q-leap.com> wrote:
> Hello,
>
> Andrew Beekhof wrote:
> > On 10/31/05, Peter Kruse <pk at q-leap.com> wrote:
> >
> >>>
> >>I still don't understand the reasoning behind this.  Why do you
> >>request a node to shoot itself at all?
> >
> >
> > a resource might have failed to stop.
>
> sorry, didn't make it clear, the question shoud read:
> Why do you request a node to shoot _itself_ at all?
> shouldn't a node always stonith another node?

because it achieves the right result and because we can

it just happens that the node in charge of detecting the problem was
also the one with the problem.

> That's what the name (shoot THE OTHER node ...)
> suggests, no?
>
> >>How do you
> >>know that it suceeded?  I mean, a node cannot declare itself
> >>dead.  How can you know that a stonith operation failed?
> >
> >
> > it doesn't really matter - if it succeeds then the node will reboot itself.
> > the only time this will happen is if the node is in a cluster
> > partition of size 1,
> > so there isnt really anyone else to notify anyway.
> >
>
> hm, that is a one-node-cluster?  are you saying that
> in a two-node cluster it's always one node fencing the other?

no - i'm saying that all fencing directives come from the DC.
who actually caries out the directive is dependent on how you
configured the stonithd (via the placement and configuration of
fencing resources)

>
> Also, this sounds like you don't check the success of the
> stonith operation, although I see in the log messages:
>
> info: Succeeded to STONITH the node ha-test-1: optype=1. whodoit: ha-test-2
>
> Are you ignoring the result?
>

		} else if (ST_OK == stonithd_node_fence( st_op )) {

and

	switch(op->op_result) {
		case STONITH_SUCCEEDED:

So the CRM is definitely checking the result.  However when the DC
shoots itself, it is by definition not there to check the result
(unless of course the op fails).