Comment 11 for bug 1422186

This issue was confirmed back as for given case at least running resources must migrate from affected node. So, the issue is correct from the cluster management perspective.

But the fix is not possible without built-in Pacemaker STONITH (fencing). Note that such issue might not affect corosync service health, and pacemaker might not be able to detect the node failure in order to trigger a STONITH. So, some external monitoring actions could be required additionally to "help" stonith-ng daemon with fencing decision.