Quorum
Also known as:
Majority Quorum
Voting Quorum
Definition
A quorum is the smallest group of nodes that has to agree before a distributed operation counts as done. The most common rule is the majority quorum, N/2 + 1. It works because any two majorities share at least one node, and that shared node is what keeps the system consistent when others fail.
Key Takeaways
- Majority quorum guarantees overlap. In a five node cluster, any group of three shares at least one node with any other group of three.
- A quorum based system can lose up to N/2 minus 1 nodes at the same time and still stay safe.
- Read and write quorums can be different sizes. As long as R plus W is greater than N, every read sees the latest committed write.
- Quorum is the budget Paxos, Raft, Dynamo style replication, and ZooKeeper all spend on fault tolerance.
How It Works
- An operation is sent to all N replicas in a group.
- Each replica handles the operation and sends an acknowledgement back.
- Once the coordinator collects acknowledgements from a quorum, the operation is committed.
- Slow or failed replicas catch up later through repair, log replay, or hinted handoff.
Where It Is Used
- Cassandra and DynamoDB let you pick the read and write quorum per query, like ONE, QUORUM, or ALL.
- Raft and Paxos commit log entries only after a majority quorum has stored them.
- Kafka’s in-sync replica set is a quorum that decides which producer writes are safe to commit.