SingletonService.elect() currently checks whether there is only a single node in the cluster view and returns that node as the master rather then delegating to the election policy. While the policy obviously cannot return any other node it could have returned null instead, and not delegating prevents that from happening.

This means that it is impossible to implement an election policy that requires a quorum, since single-node cluster caused by a split will elect itself as master rather than letting the election policy return null to indicate it should have no HA singleton services running.