Consensus

Consensus is the distributed-systems problem of getting a group of independent processes to agree on a single value or ordering of events, even when some of them crash, restart, or lose messages. It is one of the foundational problems in computer science, and the algorithms that solve it — Paxos, Raft, and the Byzantine-fault-tolerant protocols behind blockchains — are what let a cluster behave like one reliable machine. Without consensus, replicated systems cannot agree on what actually happened.

• Agreement - Every non-faulty node decides on the same value, so replicas never diverge.
• Fault tolerance - The system keeps making progress as long as a required quorum of nodes stays healthy.
• Ordering - A consistent sequence of operations is established across nodes, which is the basis of replicated logs and state machines.
• Safety and liveness trade-offs - Protocols balance never producing a wrong answer against always eventually producing one, under real network conditions.

Consensus sits underneath the databases, message queues, and coordination services that back most APIs — when a provider promises strong consistency or high availability, a consensus protocol is usually why it can. It surfaces directly in APIs for distributed data stores and blockchain networks, where clients must reason about finality and quorum. For anyone assessing an API’s reliability, understanding the consensus model beneath it explains a lot about its real consistency and failure behavior.