[Linux-HA] about the resource stickiness

[Andrew Beekhof]

On 10/5/06, Andreas Kurz <akurz at sms.at> wrote:
> Luis Carlos Peinado Bravo wrote:
> > Hello, I have read the web page
> > http://wiki.linux-ha.org/v2/faq/forced_failover and some post about the
> > resource stickiness, resource failure stickiness, forced failover... but
> > I couldn't find the way to get success.
> >
> >
> >
> > As you can see in my cib file I have 2 nodes and 4 resources in a group.
> > The resource_stickiness=500, resource_failure_stickiness=-100, the score
> > in the rsc_location for node iusslae02=1500 and for iusslae03=1000. When
> > I force failover the resources change the node after 60 failures!!! All
> > this configuration was made using the GUI in the 2.0.7 and I think that
> > maybe it isn't complete at all.
> >
> >
> >
> > Could you tell the formula I have to use to get appropriate values for
> > the different stickiness and scores? Could you tell me how are they
> > combined to get a given numbers of failures? Do I have to define the
> > resource stickiness for each resource? I'm very confused about this
> > values.
> >
>
> Hello!
>
> I've also had some problems to find out the correct formula, but Andrew
> gave me some hints for using heartbeat 2.0.7 .
>
> Calculated with your scores:
>
> node iusslae02, score 1500 (all ressources running):
> 4*(1500*2) + 4*500 = 14000
>
> node iusslae03, score 1000:
> 4*(1000*2) = 8000
>
> If I understood right, the Problem in Version 2.0.7 is the
> multiplication of the scores in location constraints, that produces much
> higher scores than expected. In your case factor 2.

correct

worth noting, this multiplication has since been fixed/removed for the
next release


> The difference when running on the prefered node 14000-8000=6000, thus
> you need a 60*(-100)=-6000 score from failing resource to get an equal
> weight for both nodes.
>
> So try e.g a score of 200 and 100 for your nodes and increase your
> resource_failure_stickiness to -500 and the failover should hapen after
> the 5th failure of a ressource.
>
> Regards,
> Andi
> _______________________________________________
> 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
>