ClusterLabs
fence_compute: Fix disabling force_down on node when action is on
When the action is on, the goal is to disable force_down for the nova-compute service on the node.
However, we were only doing that if the nova-compute service was up; which is impossible if it's forced to be down... So just always disable force_down, and then, if it's up, do more things.
~~Hrm, it's probably not correct as that would re-enable the nova-compute too early. We probably need to do that from the NovaCompute OCF agent, and add a condition here (like evacuate attribute unset or set to no).~~
Oh, actually it's probably correct because we use fabric_fencing. This still means the "on" action should be made from the NovaCompute OCF agent, though.
Rebased, cc'ing @krcmarik since he did 5964f95ab698a1f5cf50af3654350eebb961eb61 not too long ago, but I feel it's not fully correct.
@vuntz I don't follow the logic here, but this code has changed loads since then anyway, so I'm not sure if your concern is already addressed. Please can you check?
