Our redis_misc cluster has some important clients (ores, mediawiki lockmanager etc), but there are quite a few facts about it that we may need to rethink/investigate:

a) when a primary server fails, services must be failed over to the secondary one by one by humans
b) when we need to do maintenance work on a host, we have to find and ask service owners one by one whether it is OK to do it (eg reboot)
c) evaluate if we are properly utilising those machines