CEO <-> ETCD is spending an large chunk of CPU time on member listing and health checking (+ usual TLS handshakes).

With the addition of more controllers we're calling quorum checks more often, which in turn cause unnecessary load from multiple goroutines to get the exact same result.

This could be improved by a cache that refreshes either in the background or on demand after a certain staleness threshold is exceeded. 

AC:

• reduced member listing RPC / health check calls to etcd