Currently, it is not possible for cluster admins to revert from a pool that is opted into on-cluster builds and layered MachineConfig updates. See https://issues.redhat.com/browse/OCPBUGS-16201 for details around what happens.

It is worth mentioning that this is mostly an issue for UPI (user provided infrastructure) / bare metal users of OpenShift. For IPI cases in AWS / GCP / Azure / et. al., one can simply delete the node and the machine, which will cause the Machine API to provision a fresh node to replace it, e.g.:

#!/bin/bash

node_name="$1"
node_name="${node_name/node\//}"
machine_id="$(oc get "node/$node_name" -o jsonpath='{.metadata.annotations.machine\.openshift\.io/machine}')"
machine_id="${machine_id/openshift-machine-api\//}"
oc delete --wait=false "machine/$machine_id" -n openshift-machine-api
oc delete --wait=false "node/$node_name"

Done When

• The MCD can revert from a node from on-cluster builds / layered MachineConfigs into the legacy behavior.
• Or we've determined that the above is either infeasible or undeisrable.