In Search of an Understandable Consensus Algorithm

Why this one

The unusual thing about this paper is that understandability was the design goal, stated up front, and the authors ran a user study to check they had hit it. That makes it the one consensus paper you can actually finish. Raft splits the problem into three pieces you can hold separately - leader election, log replication, and safety - and then insists that all writes flow through a leader so that the log is only ever appended to in one place. Read it when you reach distributed coordination, and read the figure with the state transitions rather than skimming to the end; the terms, the election timeout and the commit index are the whole algorithm, and every production system you will meet later (etcd, Consul, TiKV) is a direct implementation of that figure.