[ClusterLabs] Tracking Node Health

[Kerry, Ron (ERT/PIR - HPC Storage Software)]

We are contemplating implementing a "health" resource based on the 
example shown with the ocf:pacemaker:HealthCPU resource agent. This one 
will track the availability of some filesystems that are not part of the 
HA configuration. The idea is to run other HA resources only on nodes 
that have access to those filesystems. Building the new resource agent 
seems fairly straight forward. However I am unclear how using a "health" 
resource which itself must be running on all cluster nodes to set a 
#health node attribute can be effective. As soon as the #health node 
attribute is set to red, the "health" resource itself can no longer run 
on that node. So how useful can a node-health-strategy set to 
migrate-on-red or only-green or even progressive be? It seems the only 
usable choice is a strategy of "custom" with explicit rules defined just 
for the other cluster resources.

Am I missing something? It certainly would be easier to use 
node-health-strategy=migrate-on-red than to define a bunch of custom 
rules for resources.

Reference: 
http://clusterlabs.org/pacemaker/doc/en-US/Pacemaker/1.1/html-single/Pacemaker_Explained/index.html#s-node-health

-- 

Ron Kerry
Hewlett Packard Enterprise