The OCF spec allows agents to support a recover action to accomplish the same thing as a stop followed by a start, if it is advantageous to do so. Pacemaker could scheduler a recover action if a stop and start are scheduled on the same node with no ordered actions between. (The existing code to replace a stop and start with live migration could be a good starting point.)

If a recover action fails, Pacemaker should schedule a stop and start (much like if a live migration fails).