A set of Ansible Roles that manage a DC/OS cluster lifecycle on RedHat/CentOS Linux.

To make best use of these roles, your nodes should resemble the Mesosphere recommended way of setting up infrastructure. Depending on your setup, it is expected to deploy to:

• One or more master node ('masters')
• One bootstrap node ('bootstraps')
• Zero or more agent nodes, used for public facing services ('agents_public')
• One or more agent nodes, not used for public facing services ('agents_private')

[bootstraps]
bootstrap1-dcos112s.example.com

[masters]
master1-dcos112s.example.com
master2-dcos112s.example.com
master3-dcos112s.example.com

[agents_private]
agent1-dcos112s.example.com
remoteagent1-dcos112s.example.com

[agents_public]
publicagent1-dcos112s.example.com

[agents:children]
agents_private
agents_public

[common:children]
bootstraps
masters
agents
agents_public

The Mesosphere DC/OS Ansible roles make use of two sets of variables:

1. A set of per-node type group_var's
2. A multi-level dictionary called dcos, that should be available to all nodes

[bootstraps:vars]
node_type=bootstrap

[masters:vars]
node_type=master
dcos_legacy_node_type_name=master

[agents_private:vars]
node_type=agent
dcos_legacy_node_type_name=slave

[agents_public:vars]
node_type=agent_public
dcos_legacy_node_type_name=slave_public

dcos:
  download: "https://downloads.dcos.io/dcos/stable/1.13.4/dcos_generate_config.sh"
  download_checksum: "sha256:a3d295de33ad55b10f5dc66c9594d9175a40f5aaec7734d664493968a9f751fd"
  version: "1.13.4"
  enterprise_dcos: false
  selinux_mode: enforcing

  config:
    cluster_name: "examplecluster"
    security: strict
    bootstrap_url: http://int-bootstrap1-examplecluster.example.com:8080
    exhibitor_storage_backend: static
    master_discovery: static
    master_list:
      - 172.31.42.1

Please see the official Mesosphere DC/OS configuration reference for a full list of possible parameters. There are a few parameters that are used by these roles outside the DC/OS config.yml, specifically:

• bootstrap_url: Should point to http://your bootstrap node:8080. Will be used internally and conviniently overwritten for the installer/upgrader to point to a version specific sub-directory.
• ip_detect_contents: Is used to determine a user-supplied IP detection script. Overwrites the build-in enviroment detection and usage of a generic AWS and/or on premise script. Official Mesosphere DC/OS ip-detect reference
• ip_detect_public_contents: Is used to determine a user-supplied public IP detection script. Overwrites the build-in enviroment detection and usage of a generic AWS and/or on premise script. Official Mesosphere DC/OS ip-detect reference
• fault_domain_detect_contents: Is used to determine a user-supplied fault domain detection script. Overwrites the build-in enviroment detection and usage of a generic AWS and/or on premise script.

Due to the nested structure of the dcos configuration, it might be required to set Ansible to 'merge' instead of 'replace', when combining config from multiple places.

# ansible.cfg
hash_behaviour = merge

When invoking these roles interactively (for example from the operator's machine), the DCOS.bootstrap role will require a manual confirmation of the cluster to run against. This is a safeguarding mechanism to avoid unintentional upgrade or config changes. In non-interactive plays, a variable can be set to skip this step, e.g.:

ansible-playbook -e 'dcos_cluster_name_confirmed=True' dcos.yml

Mesosphere DC/OS is a complex system, spanning multiple nodes to form a full multi-node cluster. There are some constraints in making a playbook use the provided roles:

1. Order of groups to run their respective roles on (e.g. bootstrap node first, then masters, then agents)
2. Concurrency for upgrades (e.g. serial: 1 for master nodes)

The provided dcos.yml playbook can be used as-is for installing and upgrading Mesosphere DC/OS.

• CentOS 7, RHEL 7
• DC/OS 1.12, both open as well as enterprise version

Apache 2.0

This role was created by team SRE @ Mesosphere and others in 2018, based on multiple internal tools and non-public Ansible roles that have been developed internally over the years.