Comment 5 for bug 1406052

Here's a slight simplification: One approach is to install two nodes with the 4 agents (Metadata, DHCP, OVS and L3). Start the agents on only one node. Then if an external system (Say: Pacemaker) detects an issue with the first node, you configure the 2nd node with the host of the first, and start all of those agents. When they start and sync with the controller, they identify themselves using the old host, and will all of the already schedules resources (Routers, DHCP networks, ports).