Gossip protocol (DB)
A decentralized way for nodes in a cluster to spread information (which nodes exist, their status, ring topology) by periodically "gossiping" with a few random peers. The information spreads epidemically until everyone knows, without a central coordinator.
Mechanic: at regular intervals, each node picks a few others and exchanges its latest view of the cluster state; new information spreads exponentially (like a rumor/an epidemic) and converges quickly across the whole cluster. Win: no central single point of failure, robust to node failures, scales to thousands of nodes. Used for membership discovery, failure detection (which nodes are down), and propagation of metadata in Cassandra, DynamoDB, Consul, Serf. Often builds on a "phi accrual failure detector" to decide whether a silent node is dead. A fundamental mechanism in decentralized, leaderless architectures. Related to consistent hashing and leaderless replication.