Also Raft is susceptible to extra leader flapping than some other consensus algorithms. That’s not the consensus algorithm I would choose for leader election for a primary datastore where you value correctness and uptime.
I haven’t fully understood the full post yet but for the pure need of selecting a primary I would use a paxos variant (so many!) that aligns with those tradeoffs.
You don’t really need something that is trading off some things for throughput.
It’s important to remember, Raft was created to be easier to understand and easier to implement. It makes tradeoffs for understandability.
This is both good and bad. It means making a pretty good raft library is more accessible to more people.
But it means creating a really great Raft library can still suffer from less reliability than a really great Paxos library for example because of those tradeoffs.
