Why it matters
Leader election is invisible when it works and catastrophic when it doesn't. Knowing the algorithms and their failure modes matters for choosing HA architectures.
The architecture
Bully algorithm: highest-ID node wins. Simple; on leader failure, higher-ID nodes claim leadership. Not fault-tolerant against partitions.
Ring algorithm: nodes arranged in ring; on failure, initiator sends election message around; highest-ID responds. Better failure detection but complex.
How it works end to end
Consensus-based election: use Raft or Paxos to elect leader. Handles partitions correctly (minority side can't elect a leader). Standard in modern systems.
ZooKeeper-based: services register ephemeral znodes; first to create wins. ZooKeeper's ZAB consensus handles the underlying agreement.
Lease-based: leader holds a time-limited lease. On expiration, candidates race to acquire. Simple but requires clock sync.