Description of problem:
After specifying a bad secret (one that doesn't contain tls.crt/tls.key) as a serving cert for the api server, modifying the secret or modifying the apiserver/cluster resource to point to a good secret does not fix the issue.

Version-Release number of selected component (if applicable):
4.1.0-rc.4

How reproducible:
Always

Steps to Reproduce:
1. Create a serving cert secret with wrong keys (say crt,key instead of tls.crt,tls.key) in the openshift-config namespace.

2. Modify the apiserver/cluster resource and specify the new secret as the default serving cert:

apiVersion: config.openshift.io/v1
kind: APIServer
metadata:
  name: cluster
spec:
  servingCerts:
    defaultServingCertificate:
      name: bad-creds

3. Wait for kube-apiserver operator to apply the configuration and observe that one of the kube-apiserver pods starts crashlooping (as expected).
4. Modify the original secret to contain proper keys (tls.crt/tls.key). Wait for a change in the kube-apiserver pods.
5. Modify apiserver/cluster to point to a newly created secret that contains the right keys:

apiVersion: config.openshift.io/v1
kind: APIServer
metadata:
  name: cluster
spec:
  servingCerts:
    defaultServingCertificate:
      name: good-creds

Wait for a change in the kube-apiserver pods.

Actual results:

After step 4 and/or 5, nothing changes. The one apiserver pod keeps crashlooping saying it can't find tls.crt.


Expected results:

After modifying the secret or the configuration, the secret should be updated on the master and the new serving cert should take effect.

Additional info:

Removing the serving certificate configuration completely:
apiVersion: config.openshift.io/v1
kind: APIServer
metadata:
  name: cluster
spec: {}

and waiting for the kube-apiserver to become stable again does get around this issue.