On 27/06/17 15:15 -0400, Dan Ragle wrote: > All that said, and for anyone interested, here's the recipe I tried that > appears to work well. After setting it up this way, I was able to > standby/unstandby each of the nodes in turn with the clones consistently > re-splitting after each unstandby (and the RealN resources all remaining on > node1 if it was available and node otherwise): > > # pcs property set placement-strategy=balanced > > # pcs node utilization node1-pcs weight=250 > # pcs node utilization node2-pcs weight=250 > > # pcs resource create Real1 ocf:pacemaker:Dummy > # pcs resource create Real2 ocf:pacemaker:Dummy > # pcs resource create Real3 ocf:pacemaker:Dummy > # pcs resource create Real4 ocf:pacemaker:Dummy > # pcs resource utilization Real1 weight=1 > # pcs resource utilization Real2 weight=1 > # pcs resource utilization Real3 weight=1 > # pcs resource utilization Real4 weight=1 > # pcs constraint location Real1 prefers node1-pcs=INFINITY > # pcs constraint location Real2 prefers node1-pcs=INFINITY > # pcs constraint location Real3 prefers node1-pcs=INFINITY > # pcs constraint location Real4 prefers node1-pcs=INFINITY > > # pcs resource create Dummy1 ocf:pacemaker:Dummy meta priority=-1 > # pcs resource create Dummy2 ocf:pacemaker:Dummy meta priority=-1 > # pcs constraint location Dummy1 prefers node2-pcs=-INFINITY > # pcs constraint location Dummy2 prefers node1-pcs=-INFINITY > # pcs resource utilization Dummy1 weight=100 > # pcs resource utilization Dummy2 weight=100 > > # pcs resource create ClusterIP ocf:heartbeat:IPaddr2 ip=1.2.3.4 nic=bond0 > cidr_netmask=24 meta resource-stickiness=0 > # pcs resource utilization ClusterIP weight=100 > # pcs resource clone ClusterIP clone-max=2 clone-node-max=2 > globally-unique=true
You may find "pcs config export pcs-commands" helpful -- it will expose the existing cluster configuration as a set of commands to reinstate it (hopefully). Actually, if you want just the CIB subset (like above), it is currently not exposed through pcs, you'd use "clufter cib2pcscmd -s". Of course, it would not express the conceptual boundaries as in the human-crafted case. It would follow the preferred accumulate-and-push approach though, which is not prone to keeping cluster in the needless reevaluations of what should run where (amongst others), unlike the stream of one-off changes to the course of cluster's run. -- Jan (Poki)
pgpr7GGrZlJrU.pgp
Description: PGP signature
_______________________________________________ Users mailing list: [email protected] http://lists.clusterlabs.org/mailman/listinfo/users Project Home: http://www.clusterlabs.org Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf Bugs: http://bugs.clusterlabs.org
