On 12/28/2010 06:46 PM, Dejan Muhamedagic wrote: > > 40 order constraints? A big cluster. >
We have currently 40 VM's (XEN) on it. I can't put them in a group since they have to run independently and not necessarily on the same node(s). To make it worse I also have location constraints and additional constraints for ping(d). > Yes, the GUI should work. You can also use the crm shell to edit > constraints. It should look sth like this: > > order o inf: lvm ( rsc1 rsc2 ... rscn ) > > providing that rsc1 ... rscn don't depend on each other. > >> And is this feature now safe to use? > > Should be. But the number of resources you want to put into the > set is unusually large. Try this first on a test cluster. It is > also possible to use several constraints: > > order o1 inf: lvm ( rsc1 rsc2 ... rscm ) > order o2 inf: lvm ( rscm+1 rscm+2 ... ) > ... > I did it now similar to the documentation, which is pretty sleek since in the GUI I have only one entry now :) <rsc_order id="ordered-start"> <resource_set id="pre-VM" sequential="true"> <resource_ref id="clvm"/> <resource_ref id="lvm"/> </resource_set> <resource_set id="VMs" sequential="false"> <resource_ref id="vm1"/> <resource_ref id="vm2"/> </resource_set> </rsc_order> And it looks good so far. But we have quite some problems with the cluster. A complete (friendly - i.g. I manually put one node to standby) failover from one node to another takes a lot of time because all the VMs have to be live-migrated. Now it seems that pacemaker only migrates one resource at a time and I remember Andrew telling me that I should add kind=Serialize to my resource_set. So I would just write: <resource_set id="VMs" sequential="false" kind="serialize"> correct? Bye, Tobi _______________________________________________ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems