This is a clone of issue OCPBUGS-4998. The following is the description of the original issue:
—
If the cluster enters the installing-pending-user-action state in assisted-service, it will not recover absent user action.
One way to reproduce this is to have the wrong boot order set in the host, so that it reboots into the agent ISO again instead of the installed CoreOS on disk. (I managed this in dev-scripts by setting a root device hint that pointed to a secondary disk, and only creating that disk once the VM was up. This does not add the new disk to the boot order list, and even if you set it manually it does not take effect until after a full shutdown of the VM - the soft reboot doesn't count.)

Currently we report:

cluster has stopped installing... working to recover installation

in a loop. This is not accurate (unlike in e.g. the install-failed state) - it cannot be recovered automatically.

Also we should only report this, or any other, status once when the status changes, and not continuously in a loop.