Description of problem:

In 4.15 and previous versions of Openshift, the MCO deploys MachineConfigNode CRDs in both default and techpreview clusters. However, since MCN is techpreview feature gated in 4.15, the CRD should not be deployed in default clusters. This functionality leads to upgrade failures, as described in OCPBUGS-59567, due to conflicts arising from having both the v1alpha1 CRD pre-existing and trying to point to the new v1 CRD.

Version-Release number of selected component (if applicable):

4.15 and prior

How reproducible:

Always when updating from <=4.15 to 4.20

Steps to Reproduce:

1. Start with a <=4.15 cluster.
2. Upgrade through each version up to 4.20.

Actual results:

The upgrade between 4.19 & 4.20 will fail with the following message being emitted by the CVO:

"Unhandled Error" err="error running apply for customresourcedefinition \"machineconfignodes.machineconfiguration.openshift.io\" (789 of 936): CustomResourceDefinition.apiextensions.k8s.io \"machineconfignodes.machineconfiguration.openshift.io\" is invalid: status.storedVersions[0]: Invalid value: \"v1alpha1\": must appear in spec.versions" logger="UnhandledError"

Expected results:

Upgrade should succeed.

Additional info:

See debugging Slack thread here for additional insight from djoshy & rh-ee-pabrodri