Avoiding a two-step creation of the config map when using DNS names

[brownleej]

What would you like to be added/changed?

When we are first creating a cluster, we currently create the config map with a placeholder cluster file, then create the pods, and then update the cluster file with the initial coordinators. This slows down the cluster creation, because we have to wait for the cluster file to propagate. When using DNS names in the cluster file, we may be able to choose the initial coordinators before creating the pods, because the DNS names will be predictable. We should try this to see if we can create the cluster without having to update the config map a second time, and ideally with only a single pass through the reconciliation loop.