Controlplane/Etcd nodes are stuck at provisioning for a new cluster
Article Number: 000020006
Environment
- Rancher version >= v2.5.6
- An RKE, RKE2 or k3s downstream cluster created in Rancher using either existing nodes, or provisioned using an infrastructure provider
- No nodes have been added with the worker role
Situation
When creating a new cluster:
- RKE: It will get stuck in "Provisioning" status, while the control plane and etcd nodes will show a "Registering" status.
-
RKE2/k3s: It will get stuck in "Updating" status, while the control plane and etcd nodes will show a "Waiting for Node Ref" status.
-
Provisioning logs will show:
[INFO ] waiting for at least one control plane, etcd, and worker node to be registered
Resolution
As of Rancher v2.5.6, new RKE, RKE2 or k3s clusters require at least one node with the worker
or all
roles to begin provisioning. The reason for this is that the downstream cluster requires a worker node to host the Rancher cluster-agent, CoreDNS, metrics-server, and ingress-nginx. Without these workloads, a cluster will not provision correctly.