[Linux-HA] Pacemaker 1.0 + heartbeat 2.99 / Problem with failover scores ? (Dejan Muhamedagic)
dejanmm at fastmail.fm
Tue Jul 7 10:06:03 MDT 2009
On Tue, Jul 07, 2009 at 01:30:22PM +0200, Alain.Moulle wrote:
> Thanks Dejan, you're right , there was something in the underlying
> that prevent the failover from the node with score 50.
> (By the way, the command ptest -Ls returns : ptest: option `-Ls' is
Funny, it works here. I guess that ptest -L -s should do in any
> Anyway, my problem now is that if something fails on the expected
> failover node,
> the resource will be failovered on one of the remaining nodes (I have 4
> nodes) and
> I don't want this. So I thought about adding two constraints with
> for the both remaining nodes, so for one resource I would have :
> 1 location constraint score="100" for node1
> 1 location constraint score="50" for node2
> 1 location constraint score="-INFINITY" for node3
> 1 location constraint score="-INFINITY" for node4
> but as I have 32 resources dispatched on the 4 nodes, that makes a lot
> of constraints
> and Pacemaker or Heartbeat-v2 seems to have a limit max for
> the number of constraints, doesn't it ??
Nope, it doesn't, but that'll certainly make your configuration
rather big. Perhaps you can try with the asymmetric cluster (set
symmetric-cluster to false). In that case you should get by with
only the first two location constraints. But note that you will
need location constraints for all rsc-node combinations when you
want to allow the resource to run.
> Thanks for your help.
> > Did you check the scores with ptest -Ls? Perhaps some of the
> > resources failed on their prefered nodes earlier. Other
> > constraints, if there are any, may also have influenced the CRM.
> > If that's fine, then it should work the way you expect. And if it
> > doesn't please file a bugzilla with hb_report for the incident.
> > Thanks,
> > Dejan
> Linux-HA mailing list
> Linux-HA at lists.linux-ha.org
> See also: http://linux-ha.org/ReportingProblems
More information about the Linux-HA