Quick and dirty Redis cluster taking advantage of Redis Sentinel for automatic failover. Persistence is turned off by default.

0. Setup docker swarm
1. Create a overlay network:

docker network create --attachable --driver overlay redis

2. Modify scripts/docker-compose.yml to how you want to deploy the stack.
3. Run scripts/bootstrap.sh.

bash scripts/bootstrap.sh latest

4. Connect to with redis-cli

docker run --rm --network redis -ti redis:4.0.11-alpine redis-cli -h redis

To access the redis cluster outside of docker, port 6379 needs to be expose. This can be done by adding ports to the docker-compose file:

...
  redis:
    image: thomasjpfan/redis-look
    ports:
      - "6379:6379"
...

A docker service called redis-zero is created to serve as the initial master for the redis sentinels to setup. The redis-look instances watches the redis sentinels for a master, and connects to redis-zero once a master has been decided. Once the dust has settled, remove the redis-zero instance and wait for failover to take over so a new redis-master will take over. Use redis-utils to reset sentinels so that its metadata is accurate with the correct state.

The use of redis-zero as a bootstrapping step allows for the docker-compose.yml to provide only the long running services:

version: '3.1'

services:

  redis-sentinel:
    image: thomasjpfan/redis-sentinel
    environment:
      - REDIS_IP=redis-zero
      - REDIS_MASTER_NAME=redismaster
    deploy:
      replicas: 3
    networks:
      - redis

  redis:
    image: thomasjpfan/redis-look
    environment:
      - REDIS_SENTINEL_IP=redis-sentinel
      - REDIS_MASTER_NAME=redismaster
      - REDIS_SENTINEL_PORT=26379
    deploy:
      replicas: 3
    networks:
      - redis

networks:
  redis:
    external: true

From now on just scale redis to expand the number of slaves or scale redis-sentinel to increase the number of sentinels.