คิว Quorum สำหรับ HA สมัยใหม่

Classic mirrored queues replicate data but suffer from split-brain risks and unpredictable failover behavior under network partitions.

RabbitMQ now recommends quorum queues instead.

A quorum queue is a replicated queue built on the Raft consensus algorithm. A majority of replicas must agree before a message is considered committed.

With Raft, writes are durable once a quorum (majority) of nodes acknowledges them. With 3 replicas, 2 must agree, so the cluster tolerates losing 1 node.

Set the queue type argument to quorum.

Each quorum queue has one leader replica and several followers. The leader handles client traffic; if it fails, followers elect a new leader automatically.

Use an odd number of replicas (3 or 5) so a clear majority always exists. Even counts waste a node without improving fault tolerance.

Quorum queues are always durable and store messages on disk across replicas. There is no transient quorum queue.

Quorum queues track a per-message delivery count, enabling automatic dead-lettering of repeatedly failing poison messages, something mirrored queues lacked.

• Mirrored: deprecated, split-brain prone
• Quorum: Raft-based, predictable failover, poison handling

Migrate mirrored queues to quorum for new deployments.

Quorum queues use more memory and disk for replication and have slightly higher latency per write, the cost of strong consistency. They are best for important, durable workloads.

The management UI shows each quorum queue's leader, online replicas, and replica lag, helping you confirm the cluster is healthy before maintenance.

Quorum queues use the Raft consensus algorithm to provide predictable, durable high availability, replacing the deprecated mirrored queues.

Declare them with x-queue-type: quorum, use an odd replica count, and gain built-in poison message handling for resilient clusters.