3 / 7 Consensus: Raft, Paxos & Leader Election

Wrong Tool for Traffic

A team proposes storing per-request session state, roughly 100k writes per second, in etcd because they want strong consistency. Why is this the wrong storage choice?

A

ZooKeeper would absorb the load instead, because ZAB commits on the leader without waiting for followers

B

Every write must reach a quorum and be durably logged before committing, so throughput is capped by that round trip

C

etcd offers only eventual consistency, so the strong consistency the team is asking for is not actually on offer here

D

Scaling the cluster out to a few dozen members would carry it, since each added member contributes write capacity