Distributed Systems

Leader election

Picking exactly one node as the coordinator.


In plain terms

Paxos, Raft, ZooKeeper, etcd. Must tolerate failures and partitions without electing two leaders.

Origin

Lynch et al. formalised leader-election in distributed systems in the 1980s. Practical implementations come from Paxos (1989), ZooKeeper Atomic Broadcast (2010), and Raft (2014).

Where it shows up in production
  • etcd Raft-based leader. Leader election triggers on heartbeat timeout (150ms default).
  • Kubernetes controllers Use a Lease object in etcd as a distributed lock for active-active-with-leader patterns.
  • ZooKeeper Ephemeral sequential znodes — the lowest-numbered live node is the leader.
On Semicolony
Sources & further reading
Found this useful?