The nodes directive is the only required section in the cluster.yml file. It’s used by RKE to specify cluster node(s), ssh credentials used to access the node(s) and which roles these nodes will be in the Kubernetes cluster.

    - address:
      user: ubuntu
      - controlplane
      - etcd
      ssh_key_path: /home/user/.ssh/id_rsa
      port: 2222
    - address:
      user: ubuntu
      - worker
      ssh_key: |-
        -----BEGIN RSA PRIVATE KEY-----

        -----END RSA PRIVATE KEY-----
    - address:
      user: ubuntu
      - worker
      hostname_override: node3
        app: ingress

Node Options

Within each node, there are multiple directives that can be used.


The address directive will be used to set the hostname or IP address of the node. RKE must be able to connect to this address.

Internal Address

The internal_address provides the ability to have nodes with multiple addresses set a specific address to use for inter-host communication on a private network. If the internal_address is not set, the address is used for inter-host communication.

Overriding the Hostname

The hostname_override is used to be able to provide a friendly name for RKE to use when registering the node in Kubernetes. This hostname doesn’t need to be a routable address, but it must be a valid Kubernetes resource name. If the hostname_override isn’t set, then the address directive is used when registering the node in Kubernetes.

Note: When cloud providers are configured, you may need to override the hostname in order to use the cloud provider correctly. There is an exception for the AWS cloud provider, where the hostname_override field will be explicitly ignored.

SSH Port

In each node, you specify which port to be used when connecting to this node. The default port is 22.

SSH Users

For each node, you specify the user to be used when connecting to this node. This user must be a member of the Docker group or allowed to write to the node’s Docker socket.

SSH Key Path

For each node, you specify the path, i.e. ssh_key_path, for the SSH private key to be used when connecting to this node. The default key path for each node is ~/.ssh/id_rsa.

Note: If you have a private key that can be used across all nodes, you can set the SSH key path at the cluster level. The SSH key path set in each node will always take precedence.


Instead of setting the path to the SSH key, you can alternatively specify the actual key, i.e. ssh_key, to be used to connect to the node.

Kubernetes Roles

You can specify the list of roles that you want the node to be as part of the Kubernetes cluster. Three roles are supported: controlplane, etcd and worker. Node roles are not mutually exclusive. It’s possible to assign any combination of roles to any node. It’s also possible to change a node’s role using the upgrade process.

Note: Prior to v0.1.8, workloads/pods might have run on any nodes with worker or controlplane roles, but as of v0.1.8, they will only be deployed to any worker nodes.

  • etcd

With this role, the etcd container will be run on these nodes. Etcd keeps the state of your cluster and is the most important component in your cluster, single source of truth of your cluster. Although you can run etcd on just one node, it typically takes 3, 5 or more nodes to create an HA configuration. Etcd is a distributed reliable key-value store which stores all Kubernetes state. Taint set on nodes with the etcd role is shown below:

Taint Key Taint Value Taint Effect true NoExecute
  • controlplane

With this role, the stateless components that are used to deploy Kubernetes will run on these nodes. These components are used to run the API server, scheduler, and controllers. Taint set on nodes with the controlplane role is shown below:

Taint Key Taint Value Taint Effect true NoSchedule
  • worker

With this role, any workloads or pods that are deployed will land on these nodes.

Docker Socket

If the Docker socket is different than the default, you can set the docker_socket. The default is /var/run/docker.sock


You have the ability to add an arbitrary map of labels for each node. It can be used when using the ingress controller’s node_selector option.