On Wednesday 30 July 2008, daniel peess wrote:
> hello andreas,
>
> On Wed, Jul 30, 2008 at 11:04:10AM +0200, Andreas Kurz wrote:
> > Ok .. I see. Try to set the 'default-resource-stickiness' to a
> > positive value and give each of your groups a different 'priority'.
> > That should do the trick.
>
> setting the 'default-resource-stickiness' to a positive value now prevents
> the restart behavior when a node returns, thanks.
> but this is only half of a workaround for the problem below.
>
> this doesn't help if you crash/standby/stop a node.
> resources that were running on this node are pushing away
> other resources, although both of their scores are equal/unset.
> if other free nodes are available the failing resource should start there,
> and if none are available shouldn't start at all.
>
> instead heartbeat restarts resources depending on the colocation
> constraints, although those should just distribute the resources across all
> nodes.
>
> again, all groups shall be treated equal if they have the same score,
> the failure of one shall never affect other ones.
> IMHO this is a bug.

hmm ... I've done some ptests (v2.1.3 ) and one way to prevent "resource 
shuffling" was to set 'symmetric-cluster' to 'false', stickiness to INFINTIY 
and  to define a complete set of location constraints for each group/node 
combination with the same positive value. 

Don't know if this behavior is a bug or a feature ... only Andrew could I 
assume. Maybe it's worth to try a recent pacemaker version?

Regards,
Andreas

>
> greetings,
> daniel
_______________________________________________
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