High Availability in RabbitMQ: Clustering and Mirrored Queues Explained

Introduction

High Availability (HA) means your system can keep working even if some parts fail. In RabbitMQ, HA is very important because if the broker goes down, messages may be lost, or services may stop.

RabbitMQ offers two main features for HA:

1. Clustering → combine many RabbitMQ nodes into one logical broker.
2. Mirrored Queues (Classic Queues HA) or Quorum Queues → copy queue data across nodes.

This article explains these features in simple terms.

What is Clustering?

A RabbitMQ cluster is a group of RabbitMQ nodes working together.

• All nodes share the same users, vhosts, exchanges, bindings, and queues metadata.
• A producer or consumer can connect to any node.
• Messages are stored on the node that hosts the queue (unless mirrored).

Benefits of Clustering

• Load balancing: clients can connect to different nodes.
• Failover: if one node dies, others are still alive.
• Scalability: add more nodes to handle more connections.

Limitation

• By default, queues live only on one node. If that node fails, the queue and its messages are gone.
• To fix this, use mirrored queues or quorum queues.

Mirrored Queues (Classic Queue HA)

A mirrored queue keeps copies of its data on multiple nodes.

• One node is the master (leader).
• Other nodes are mirrors (followers).
• When the master node goes down, one of the mirrors is promoted to master.
• Clients can continue without losing messages.

Example Flow

flowchart LR
  P[Producer] --> N1((Node 1: Master Queue))
  N1 --> N2((Node 2: Mirror Queue))
  N1 --> N3((Node 3: Mirror Queue))
  N1 --> C[Consumer]

• Producer sends message → goes to Master (Node 1).
• Master replicates the message to Node 2 and Node 3.
• Consumer reads from the master.
• If Node 1 dies → Node 2 becomes the new master.

Quorum Queues

RabbitMQ also supports Quorum Queues, a newer, more reliable option than mirrored queues.

• Based on the Raft consensus algorithm.
• Messages are replicated across nodes.
• Provides strong consistency (only committed messages survive).
• Recommended for new systems instead of mirrored queues.

Example Flow

flowchart LR
  P[Producer] --> QQ1((Quorum Queue Leader))
  QQ1 --> QQ2((Follower))
  QQ1 --> QQ3((Follower))
  QQ1 --> C[Consumer]