Currently switches are managed by whatever controller node responds first when they connect. There should be a mechanism to allow the OVSDB plug-in to load-balance switches among the controller nodes. (It's probably worth analysing the impact of shard leadership too.)