Description of problem:

All jobs perma-fail on 4.20 after the 4.19 branching. The symptom is various cluster operators failing because of missing workers. Worker Machines are in Provisioned state, no Nodes are created, no CSR's are recorded. From the serial console logs, it looks like CoreOS started booting but never finished or crashed. SSH access does not work.

The problem can be reproduced locally using dev-scripts on 4.20 with a CI payload.