Setup a Docker Swarm cluster on RHEL/CentOS and Debian/Ubuntu servers using the new Docker Engine's "Swarm Mode" (https://docs.docker.com/engine/swarm/).

None.

Available variables are listed below, along with default values (see defaults/main.yml):

docker_repo: main
# docker_repo: testing
# docker_repo: experimental

The repo from which install Docker. Override the default to install testing or experimental docker builds.

docker_dependencies: "{{ default_docker_dependencies }}"

Extra packages that have to installed together with Docker. The value of default_docker_dependencies depends on the target OS family.

NB: If you are installing Docker on a Raspberry running Raspbian or any other Debian-like OS make sure to set docker_dependencies: [ ] otherwise Ansible will fail because the linux-image-extra-virtual package is not available for the arm architecture (see issue #4).

docker_swarm_interface: "{{ ansible_default_ipv4['alias'] }}"

Setting docker_swarm_interface allows you to define which network interface will be used for cluster inter-communication.

docker_swarm_addr: "{{ hostvars[inventory_hostname]['ansible_' + docker_swarm_interface]['ipv4']['address'] }}"

By default, the ip address for raft API will be taken from desired interface. You can setup listening address where the raft APIs will be exposed, overriding the docker_swarm_addr variable value in your playbook.

docker_swarm_port: 2377

Listening port where the raft APIs will be exposed.

docker_admin_users:
  - "{{ ansible_user }}"

The list of users that has to be added to the docker_group to interact with the Docker daemon. NB: The users must already exist in the system.

skip_engine: False
skip_group: False
skip_swarm: False
skip_docker_py: False

Setting skip_engine: True will make the role skip the installation of docker-engine. If you want to use this role to just install docker-engine without enabling swarm-mode set skip_swarm: True. To skip the tasks adding the docker_admin_users to the docker_group set skip_group: True. Finally, the docker-py installation task can be skipped setting skip_docker_py to True.

Node labels provide a flexible method of node organization. You can also use node labels in service constraints. Apply constraints when you create a service to limit the nodes where the scheduler assigns tasks for the service. You can define labels by swarm_labels variable, e.g:

$ cat inventory
...
[docker_swarm_manager]
swarm-01 swarm_labels=deploy

[docker_swarm_worker]
swarm-02 swarm_labels='["libvirt", "docker", "foo", "bar"]'
swarm-03
...

In this case:

$ docker inspect --format '{{json .Spec.Labels}}'  swarm-02 | jq
{
   "bar": "true",
   "docker": "true",
   "for": "true",
   "libvirt": "true",
}

You can assign labels to cluster running playbook with --tags=swarm_labels

NB: Please note, all labels that are not defined in inventory will be removed

None.

$ cat inventory
swarm-01 ansible_ssh_host=172.10.10.1
swarm-02 ansible_ssh_host=172.10.10.2
swarm-03 ansible_ssh_host=172.10.10.3

[docker_engine]
swarm-01
swarm-02
swarm-03

[docker_swarm_manager]
swarm-01 swarm_labels=deploy

[docker_swarm_worker]
swarm-02 swarm_labels='["libvirt", "docker", "foo", "bar"]'
swarm-03

$ cat playbook.yml
- name: "Provision Docker Swarm Cluster"
  hosts: all
  roles:
    - { role: atosatto.docker-swarm }

MIT

Andrea Tosatto (@_hilbert_)