RabbitMQ Core Knowledge

Full Reference: See advanced.md for producer patterns (Node.js, Java, Python, Go), consumer patterns, security configuration, and high availability setup.

Deep Knowledge: Use mcp__documentation__fetch_docs with technology: rabbitmq for comprehensive documentation.

Quick Start (Docker)

services:
  rabbitmq:
    image: rabbitmq:3-management
    ports:
      - "5672:5672"   # AMQP
      - "15672:15672" # Management UI
    environment:
      - RABBITMQ_DEFAULT_USER=admin
      - RABBITMQ_DEFAULT_PASS=admin

Core Concepts

Concept	Description
Exchange	Routes messages to queues
Queue	Buffer that stores messages
Binding	Rule linking exchange to queue
Routing Key	Message attribute for routing

Architecture

Producer ──▶ Exchange ──binding──▶ Queue ──▶ Consumer
                │
                ├── direct   (exact routing key match)
                ├── topic    (pattern matching)
                ├── fanout   (broadcast to all)
                └── headers  (header-based routing)

Exchange Types

Type	Routing	Use Case
direct	Exact key match	Task queues, RPC
topic	Pattern (*.#)	Event routing
fanout	Broadcast all	Notifications
headers	Header match	Complex routing

# Topic exchange patterns
order.*        → matches order.created, order.updated
order.#        → matches order.created, order.item.added
*.created      → matches order.created, user.created

Queue Configuration

Argument	Description
`x-message-ttl`	Message expiration (ms)
`x-max-length`	Max queue size
`x-dead-letter-exchange`	DLX for rejected messages
`x-dead-letter-routing-key`	DLQ routing key
`x-queue-type`	classic, quorum, stream

When NOT to Use This Skill

Event streaming with replay - Kafka provides persistent log
Ultra-high throughput (>100k msg/s) - Kafka or Pulsar scale better
AWS-native architecture - SQS integrates better
Lightweight messaging - NATS has simpler operations
JMS compliance required - ActiveMQ provides JMS API

Anti-Patterns

Anti-Pattern	Why It's Bad	Solution
No prefetch limit	Consumer overwhelmed	Set `prefetch` to 10-50
Classic queues in HA	Data loss on node failure	Use quorum queues
No DLX configured	Poison messages loop forever	Configure dead letter exchange
Automatic acks	Message loss on crash	Use manual acks
No publisher confirms	Silent message loss	Enable confirms for critical messages

Quick Troubleshooting

Issue	Likely Cause	Fix
Messages piling up	Slow/dead consumer	Check consumer count and processing
High memory usage	Too many unacked messages	Reduce prefetch, add consumers
Messages not routed	Wrong routing key	Check exchange-queue bindings
Duplicate messages	Consumer crashed before ack	Implement idempotent processing
Publisher confirms timeout	Broker overloaded	Reduce publish rate

Production Checklist

Reference Documentation

Available topics: basics, exchanges, queues, consumers, clustering, production