Details
Description
This morning we ran into a situation in which some 0.2.x region servers started up and joined a 0.18.1 cluster.
This 'sort of' worked in that the hrs could communicate with the master, but clients could not communicate with the 0.2 region servers because the api version changed (the master wouldn't have liked it if it had deployed root or meta there).
Suggestion is that we have a single api version that gets bumped when any of the interfaces changes.