Description of problem

Spin off of OCPBUGS-30192

The daemon process can exit due to health check failures in 4.16+, after we added apiserver server CA rotation handling. The came with the side effect that if the MCD happens to exit in the middle of the update (e.g. image pull portion), the files/units would have been updated but the OS upgrade would not, blocking the upgrade indefinitely when the new container comes up.

Version-Release number of selected component

4.16

How reproducible

Only in BM CI so far, unsure if other issues contribute to this.

Steps to Reproduce

Get lucky and have api-int DNS break while the machine-config daemon is deploying updated files to disk. Unclear how to reliably trigger this, or distinguish from OCPBUGS-30192 and other failure modes.

Actual results

Expected results

Additional info