The Consensus Problem

In distributed systems, Consensus is the fundamental algorithmic problem of getting a set of independent, distributed computing nodes to agree on a single data value, transaction outcome, or sequential state machine mutation, even when some nodes crash or network packets are delayed and dropped.

Consensus is the core engine powering modern distributed infrastructure:

• Leader Election: Reaching agreement on which server is the active Primary (e.g., ZooKeeper, Raft).
• State Machine Replication (SMR): Ensuring every database node executes the exact same sequential log of transactions in the identical order.
• Distributed Commit: Agreeing whether a multi-partition transaction should COMMIT or ABORT (Two-Phase Commit).

The Four Formal Properties of Distributed Consensus:

1. Agreement (Safety): Every non-faulty node must decide on the exact same value.
2. Validity: If the cluster decides on value v, then v must have been proposed by at least one participant.
3. Integrity: No node can decide more than once.
4. Termination (Liveness): Every non-faulty node eventually reaches a decision.