Handle better failing over non-running instances