Skip to content

Understanding Clusters

RancherOS bootstraps a node with Kubernetes (k3s/rke2) and Rancher such that all future management of Kubernetes and Rancher can be done from Kubernetes. This is done by running Rancherd once per node on boot. Once the system has been fully bootstrapped it will not run again. Rancherd is ran from cloud-init and it's configuration is embedded in the cloud-config file.

Cluster Initialization

Creating a cluster always starts with one node initializing the cluster, and all other nodes joining the cluster by pointing to a server node. The node that will initialize a new cluster is the one with role: server and server: "" (empty). The new cluster will have a token generated or you can manually assign a unique string. The token for an existing cluster can be determined by running rancherd get-token on a server node.

Joining Nodes

Nodes can be joined to the cluster as the role server to add more control plane nodes or as the role agent to add more worker nodes. To join a node you must have the Rancher server URL (which is by default running on port 8443) and the token. The server and token are assigned to the server and token fields respectively.

Node Roles

Rancherd will bootstrap a node with one of the following roles

  1. server: Joins the cluster as a new control-plane,etcd,worker node
  2. agent: Joins the cluster as a worker only node.

Server discovery

It can be quite cumbersome to automate bringing up a clustered system that requires one bootstrap node. Also there are more considerations around load balancing and replacing nodes in a proper production setup. Rancherd support server discovery based on go-discover.

To use server discovery you must set the role, discovery and token fields. The discovery configuration will be used to dynamically determine what is the server URL and if the current node should act as the node to initialize the cluster.

Example

role: server
discovery:
  params:
    # Corresponds to go-discover provider name
    provider: "mdns"
    # All other key/values are parameters corresponding to what 
    # the go-discover provider is expecting
    service: "rancher-server"
  # If this is a new cluster it will wait until 3 server are 
  # available and they all agree on the same cluster-init node
  expectedServers: 3
  # How long servers are remembered for. It is useful for providers
  # that are not consistent in their responses, like mdns.
  serverCacheDuration: 1m