Related to the latency question, I just watched the Jane Street video (very nice!) and he mentioned that they use operator-initiated failover and he didn't know of anyone using a consensus based approach because it adds an extra hop. Does your Raft-based failover solution do automatic failover?