Spreading state without a leader
A gossip protocol lets a cluster share information without any central coordinator. On a fixed interval each node picks a few random peers and exchanges what it knows: who is alive, who is dead, and recent metadata. Over many rounds every fact reaches every node.
The spread is epidemic: each informed node infects a few more each round, so information reaches the whole cluster in a number of rounds that grows only with the logarithm of the cluster size.
Why teams like it
- Scalable because each node talks to only a handful of peers per round.
- Robust because there is no single coordinator to fail.
- Eventually consistent because all nodes converge given enough rounds.
The cost is that updates are not instant and there is a small steady stream of background chatter.
Gossip commonly carries failure detection with heartbeats: if no node has heard from a peer recently, the cluster marks it suspect and then dead.
Key idea
Gossip spreads cluster knowledge through random peer exchanges, scaling without a leader at the cost of delayed convergence.