Category: architecture

Running a Apache Cassandra Cluster with Docker-Swarm is quite easy using the official Docker Image. Docker-Swarm allows you to setup several docker worker nodes running on different hardware or virtual servers. Take a look at my example docker-compose.yml file:

version: "3.2"

networks:
  cluster_net:
    external:
      name: cassandra-net  
  
services:  

  ################################################################
  # The Casandra cluster 
  #   - cassandra-node1
  ################################################################        
  cassandra-001:
    image: cassandra:3.11
    environment:
      CASSANDRA_BROADCAST_ADDRESS: "cassandra-001"
    deploy:
      restart_policy:
        condition: on-failure
        max_attempts: 3
        window: 120s
      placement:
        constraints:
          - node.hostname == node-001
    volumes:
        - /mnt/cassandra:/var/lib/cassandra 
    networks:
      - cluster_net

  ################################################################
  # The Casandra cluster 
  #   - cassandra-node2
  ################################################################        
  cassandra-002:
    image: cassandra:3.11
    environment:
      CASSANDRA_BROADCAST_ADDRESS: "cassandra-002"
      CASSANDRA_SEEDS: "cassandra-001"
    deploy:
      restart_policy:
        condition: on-failure
        max_attempts: 3
        window: 120s
      placement:
        constraints:
          - node.hostname == node-002
    volumes:
        - /mnt/cassandra:/var/lib/cassandra 
    networks:
      - cluster_net

I am running each cassandra service on a specific host within my docker-swarm. We can not use the build-in scaling feature of docker-swarm because we need to define a separate data volume for each service. See the section ‘volumes’.

The other important part are the two environment variables ‘CASSANDRA_BROADCAST_ADDRESS’ and ‘CASSANDRA_SEEDS’.

‘CASSANDRA_BROADCAST_ADDRESS’ defines a container name for each cassandra node within the cassandra cluster. This name matches the service name. As both services run in the same network ‘cluster_net’ the both cassandara nodes find each user via the service name.

The second environment ‘CASSANDRA_SEEDS’ defines the seed node which need to be defined for the second service only. This is necessary even if a cassandra cluster is ‘master-less’.

That’s is!