[Linux-HA] rsc_location problem

[Andrew Beekhof]

On 11/9/06, Zachár Balázs <zachar at direkt-kfki.hu> wrote:
> Yes, I see it but if I right, it isn't help me.
>
> If the A node can't ping the ping host (-110 score) I would like to
> migrate the resource to B node, so the resource_stickiness must allow
> the migration.
> But, If the A node can ping again the ping host (+110 score) the scores
> are the same as the start scores so It will migrate back the resource,
> and because the resource_stickiness allow the previous migration it
> allow this migration too.

assuming:
* no rsc_location constraints,
* the pingd score for nodeA is X
* the pingd score for nodeA is Y

if X == Y, then even a resource_stickiness of 1 will leave it on the same node.
if X < Y, then as long as resource_stickiness < (Y-X), the resource
will always run on Y
if X > Y, then as long as resource_stickiness < (X-Y), the resource
will always run on X

So also assuming a pingd multiplier of 100, then resource_stickiness
should be less than 100 to do what you want.

In practice, .7 had some non-intuitive multiplications going on... so
I'd suggest a resource_stickiness of about 10 to be on the safe side.

>
>
> Sorry about my poor English.
>
> Regards,
> Balázs
>
> Andrew Beekhof írta:
> > On 11/9/06, Zachár Balázs <zachar at direkt-kfki.hu> wrote:
> >> Helo @list!
> >>
> >> I have a 2 nodes active/passive cluster with one resource group which
> >> works fine.
> >>
> >> I have a question for the rsc_location.
> >> The problem:
> >>
> >> I have a rule about add 100 score to node wich uname = cl-cluster-a.
> >>
> >> And I use pingd with one pinghost and 110 multiplier...
> >>
> >> So,
> >> 1. If I start the heartbeat both node, my resource group starts on
> >> cl-cluster-a. (Scores: A: 210 B: 110)                        It is OK!
> >> 2. I unplug the network cable from A node the cluster failovers.
> >> (Scores: A:100 B:110)                                               It
> >> is OK!
> >> 3. I plug back the cable, the cluster migrate back the resource group.
> >> (Scores: A: 210 B: 110)                                     This is what
> >> I don't want to!!!!!!!!!!!!!!
> >>
> >> If a node have a pingd fail, I don't want to migrate back the resource
> >> there.
> >>
> >> I tried the resource_failure_stickiness with -120 value but it didn't
> >> help for me. (I think, because CIB is only use this value on a resource
> >> fail.)
> >>
> >>
> >> Can I do this with scores or I must use stonith?
> >
> > have you looked at the resource_stickiness? (its the opposite of
> > resource_failure_stickiness in that its a measure of how much you want
> > the resource to stay where it is)
> > _______________________________________________
> > Linux-HA mailing list
> > Linux-HA at lists.linux-ha.org
> > http://lists.linux-ha.org/mailman/listinfo/linux-ha
> > See also: http://linux-ha.org/ReportingProblems
> >
> >
> >
>
>
> _______________________________________________
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>