Tutorial: Using Flocker with Docker, Swarm, Compose¶

What You’ll Need¶

• A Flocker cluster with Swarm installed.

  • Use one of our Flocker with Swarm installation options to get one.

• A Client machine with Docker Compose and access to the Swarm master.

  • If you used the CloudFormation installer:

    • The Client EC2 instance is preconfigured with Docker Compose. Use the following command to ssh in:

      ssh -i <KeyPath> ubuntu@<ClientNodeIP>
      

    • <KeyPath> is the path on your machine to the .pem file you downloaded from AWS, for example: ~/Downloads/flocker-test.pem.

    • You will need <ClientNodeIP>, <ControlNodeIP>, <AgentNode1IP> and <AgentNode2IP> from the CloudFormation Outputs tab.

    • The rest of this tutorial will assume you are logged into the Client instance.

  • If you did not use the CloudFormation installer:

    • Install Docker Compose on any machine which has network access to the Swarm master that you created when you installed Swarm manually.
    • The rest of this tutorial will assume you are logged into the machine you installed Compose on.

Step 1: Set DOCKER_HOST¶

Compose uses the environment variable DOCKER_HOST to know how to talk to the Swarm master. If you used the CloudFormation installer, it is listed in the Outputs tab of your CloudFormation stack.

Use the following commands to set the DOCKER_TLS_VERIFY and DOCKER_HOST environment variables:

export DOCKER_TLS_VERIFY=1
export DOCKER_HOST=tcp://<ControlNodeIP>:2376

<ControlNodeIP> is the IP address of your Swarm master.

Step 2: Deploy the app on the first node¶

The two Docker Compose files below need to be saved on your Client machine, in a directory named swarm-compose-tutorial.

tutorial-downloads/flocker-swarm-tutorial-node1.yml

tutorial-downloads/flocker-swarm-tutorial-node2.yml

You can either click the cloud icons to save the files locally, and then move them onto your Client machine using using a transfer medium such as scp, or right click each file, and copy the link address and run the following commands with the tutorial URLs:

mkdir swarm-compose-tutorial
cd swarm-compose-tutorial
wget <Tutorial1Url>
wget <Tutorial2Url>

The Docker Compose files both have the same layout, as illustrated below, except the node2 file has constraint:flocker-node==2 instead of constraint:flocker-node==1.

version: "2"

volumes:
   postgres:
      driver: "flocker"
      driver_opts:
        size: "10GiB"
        profile: "bronze"

services:
   web:
     image: binocarlos/moby-counter
     ports:
        -  "80:80"
     environment:
        - "constraint:flocker-node==1"
        - "USE_POSTGRES_HOST=postgres"
        - "POSTGRES_USER=flocker"
        - "POSTGRES_PASSWORD=flockerdemo"
     network_mode: "bridge"
     links:
        - "postgres:postgres"
   postgres:
     image: postgres:9.5
     ports:
        -  "5432:5432"
     environment:
        - "constraint:flocker-node==1"
        - "POSTGRES_USER=flocker"
        - "POSTGRES_PASSWORD=flockerdemo"
        - "POSTGRES_DB=postgres"
        - "PGDATA=/var/lib/postgresql/data"
     network_mode: "bridge"
     volumes:
        - 'postgres:/var/lib/postgresql/data'

• The moby-counter app container is exposed on port 80.
• The app is configured with the same database credentials as the database.
• The postgres container uses a volume_driver of flocker and uses a named Flocker volume called postgres.
  • Flocker will automatically provision this volume on-demand if it doesn’t already exist.

Now deploy the app by running:

docker-compose -f flocker-swarm-tutorial-node1.yml up -d

Open <AgentNode1IP> in a browser, and click around to add some Docker logos on the screen. For example:

The locations of the logos get stored (persisted) in the PostgreSQL database, and saved to the Flocker volume.

Step 3: Move the app¶

Now we will demonstrate stopping the app on one machine and starting it on the other.

docker-compose -f flocker-swarm-tutorial-node1.yml down
docker-compose -f flocker-swarm-tutorial-node2.yml up -d

Note that we are destroying the first set of containers and then starting the second compose file which has the constraint to force Swarm to schedule the containers onto the second node.

Flocker will detach and attach the storage so that the container starts up with the expected data.

Open <AgentNode2IP> in a browser, and you’ll be able to see that your data has persisted! For example:

Cleaning up¶

To clean up the containers and Docker’s references to the volumes, run:

docker-compose -f flocker-swarm-tutorial-node2.yml down
docker volume rm swarmcomposetutorial_postgres

To actually delete the volumes, we need to use flockerctl. For more information, see About the Docker Integration and The flockerctl Command.

unset DOCKER_HOST
unset DOCKER_TLS_VERIFY
export FLOCKER_CERTS_PATH=/etc/flocker
export FLOCKER_USER=user1
export FLOCKER_CONTROL_SERVICE=<ControlNodeIP>
flockerctl ls
flockerctl destroy -d <DatasetID> # use <DatasetID> from output above

Note that this will destroy the postgres volume and all the data in it.

Next steps¶

Now try deploying your own Docker Compose app!

• Set volume_driver: flocker for any stateful containers you have.
• Specify the Flocker volumes using "flocker_volume_name:/path_inside_container" syntax for one or more of the list elements under the volumes key.