Prior to Rancher v2.5.6, the agent did not have native Windows manifests on downstream clusters with Windows nodes. This would result in a failing agent pod for the cluster.

If you are upgrading from an older version of Rancher to v2.5.6+, you can deploy a working agent with the following workflow in the downstream cluster:

  1. Cordon all Windows nodes.
  2. Apply the below toleration to the agent workload.
  3. Uncordon all Windows nodes.
  4. Delete all agent pods. New pods should be created with the new toleration.
  5. Once the agent pods are running, and auto-update is enabled for Fleet, they should be updated to a Windows-compatible agent version.
tolerations:
- effect: NoSchedule
  key: cattle.io/os
  operator: Equal
  value: linux