How to Build a High‑Availability RabbitMQ Cluster on CentOS with Docker

Demand Analysis

During RabbitMQ deployment we analyzed existing connections (631) and queues (80,418). Cloud service could not meet the required queue count and cost was high, so a self‑built three‑node cluster was chosen, each node using 8 CPU, 16 GB RAM, 100 GB disk, 5 Mbps network.

Server Specification Comparison

Selected self‑built nodes (8 core/16 GB) cost ¥2,485 per month, compared with Tencent Cloud options where higher specs still fell short of queue requirements.

Cluster Construction Requirements

Future expansion must be supported; node specifications may be adjusted to 4 core/8 GB with 150 GB storage for new workloads. Memory usage threshold should be set to 70 %.

Environment Setup

All nodes run CentOS 7.9. Hostnames are set, and the init.sh script installs basic tools, disables firewalld and SELinux, increases file descriptor limits, applies kernel tuning, reduces swap usage, and installs performance tools.

Data Disk Formatting

mkfs.ext4 /dev/vdb
mount /dev/vdb /data
echo "/dev/vdb /data ext4 defaults 0 0" >> /etc/fstab
mount -a
mkdir -p /data/{apd,logs,prog,setup,backup,www}

Docker and Docker‑Compose Installation

yum install -y yum-utils device-mapper-persistent-data lvm2 git htop
sudo yum-config-manager --add-repo https://mirrors.aliyun.com/docker-ce/linux/centos/docker-ce.repo
sudo sed -i 's+download.docker.com+mirrors.aliyun.com/docker-ce+' /etc/yum.repos.d/docker-ce.repo
yum makecache fast
yum -y install docker-ce-18.03.1.ce
systemctl enable --now docker
mkdir -p /etc/docker
cat > /etc/docker/daemon.json <<'EOF'
{
  "registry-mirrors": ["https://rbmo5xql.mirror.aliyuncs.com"],
  "log-driver":"json-file",
  "bip": "192.168.1.5/24",
  "log-opts": { "max-size": "50m", "max-file": "1" }
}
EOF
systemctl daemon-reload && systemctl restart docker
cd /data/setup
wget -O docker-compose https://github.com/docker/compose/releases/download/1.29.2/docker-compose-Linux-x86_64
chmod +x docker-compose
cp docker-compose /usr/local/bin/
ln -sf /usr/local/bin/docker-compose /usr/bin/docker-compose
docker-compose -v

RabbitMQ Cluster Deployment

Use the rabbitmq:3.8-management image (avoid the Alpine variant due to known vulnerabilities). Pull the image on all three nodes.

# docker pull rabbitmq:3.8-management

Cookie Generation

cat > rabbitmq-cookie.sh <<'eof'
docker run -d --name mq rabbitmq:3.8-management
sleep 10
docker exec -it mq cat /var/lib/rabbitmq/.erlang.cookie > .erlang.cookie
chmod 600 .erlang.cookie
docker rm -f mq
docker volume prune
eof
sh rabbitmq-cookie.sh

RabbitMQ Configuration

cat > rabbitmq.conf <<'eof'
loopback_users.guest = false
listeners.tcp.default = 5672
cluster_formation.peer_discovery_backend = rabbit_peer_discovery_classic_config
cluster_formation.classic_config.nodes.1 = rabbit@pos_rabbitmq_1
cluster_formation.classic_config.nodes.2 = rabbit@pos_rabbitmq_2
cluster_formation.classic_config.nodes.3 = rabbit@pos_rabbitmq_3
eof

Docker‑Compose Files

Each node uses a similar docker-compose.yaml with host networking, extra hosts mapping, volume mounts, and environment variables. Memory high‑watermark is set to 0.7 (70 %).

version: "3.6"
services:
  pos_rabbitmq_1:
    image: rabbitmq:3.8-management
    restart: always
    container_name: pos_rabbitmq_1
    network_mode: host
    extra_hosts:
      - "pos_rabbitmq_1:172.17.80.27"
      - "pos_rabbitmq_2:172.17.80.32"
      - "pos_rabbitmq_3:172.17.80.6"
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /data/apd/rabbitmq:/var/lib/rabbitmq
      - ./rabbitmq.conf:/etc/rabbitmq/rabbitmq.conf
      - .erlang.cookie:/var/lib/rabbitmq/.erlang.cookie
      - ../enabled_plugins:/etc/rabbitmq/enabled_plugins
      - /data/logs/rabbitmq:/var/log/rabbitmq
    environment:
      - LANG=C.UTF-8
      - RABBITMQ_DEFAULT_USER=root
      - RABBITMQ_DEFAULT_PASS=xxxxxx
      - RABBITMQ_VM_MEMORY_HIGH_WATERMARK=0.7

Cluster Startup and HA Policy

# docker-compose up -d
# docker logs pos_rabbitmq_1 -f
# docker exec -it pos_rabbitmq_1 rabbitmqctl set_policy ha-all "^" '{"ha-mode":"all"}'

Testing

The three nodes form a cluster, HA mirroring is enabled, and memory usage is limited to 70 % of the 8 GB RAM, leaving about 5.2 GB for applications.