The clients are configured to look up the master servers using DNS SRV records.
Expected behaviour - if one master server times out or doesn't respond at all then the client should fall back to trying to talk to the same server using IPv4.
If there is still no response of the server times out then use the next SRV record (IPv6 first then IPv4).
Actual behaviour - the client stubbornly refuses to retry the current request.
Given that our clients are spread out geographically in literally hundreds of locations around the world, it is reasonable to expect that one master or the other will be unreachable from time-to-time.