What is a Quorum

A quorum is the minimum number of nodes in a distributed system that must participate in an operation for it to be considered successful. For a cluster of N nodes, the standard majority quorum is floor(N/2) + 1. A 3-node cluster requires 2 nodes to agree. A 5-node cluster requires 3. This majority rule ensures that any two quorums overlap by at least one node, which guarantees that the latest written value is always visible to a subsequent read.

How it works

Quorums are used in two main contexts: consensus algorithms and replicated storage systems.

In consensus algorithms like Raft, a leader must receive acknowledgment from a quorum of followers before a log entry is considered committed. If the leader crashes, a new leader election also requires a quorum. Because any two quorums share at least one member, the new leader is guaranteed to have seen all committed entries.

In replicated databases like Cassandra and DynamoDB, quorums are configured separately for reads and writes using the parameters W (write quorum) and R (read quorum). The constraint W + R > N guarantees that a read quorum and write quorum always overlap, so reads see the most recent write. Common configurations include W=2, R=2, N=3 (balanced) or W=3, R=1, N=3 (fast reads, slower writes).

Quorum systems tolerate up to N - quorum node failures. A 3-node cluster tolerates 1 failure. A 5-node cluster tolerates 2. Increasing cluster size improves fault tolerance but adds latency because more nodes must respond. This is why most systems use 3 or 5 nodes -- odd numbers avoid ties during leader election, and the fault tolerance gain from 5 to 7 nodes rarely justifies the cost.

Why it matters

Quorums are the mechanism that makes distributed systems both fault-tolerant and consistent. Every time a distributed database acknowledges a write or a consensus algorithm commits a decision, a quorum is what guarantees the result survives node failures.

See How Consensus Works and How Replication Works for quorum mechanics in practice.