According to Jay Pipes comments: "I do not think you need to put nova-scheduler as a single resource for pacemaker, no. The state that each scheduler process keeps is kept up to date with calls to ComputeNode.get_all() during each beginning of scheduler select_destinations() loop call.
So there is no stale data anywhere... and when the compute node aborts a resource claim, the resource request is rescheduled to another host. so no, I don't think the scheduler needs to be a single resource for pacemaker :)
for cinder-api, there is no state, therefore it should be active/active as well."
According to Jay Pipes comments: "I do not think you need to put nova-scheduler as a single resource for pacemaker, no. The state that each scheduler process keeps is kept up to date with calls to ComputeNode. get_all( ) during each beginning of scheduler select_ destinations( ) loop call.
So there is no stale data anywhere... and when the compute node aborts a resource claim, the resource request is rescheduled to another host. so no, I don't think the scheduler needs to be a single resource for pacemaker :)
for cinder-api, there is no state, therefore it should be active/active as well."