Reconciliation does not work for EtcdCluster resources that failed to be created.

[kvaps]

Hey team. Just discovered a bug:

1. Installed an old version of the operator v0.0.1
2. Applied an invalid EtcdCluster for it (with podSpec), it started crashing with CrashLoopBackOff - which is okay.
3. Installed a newer version of the operator v0.1.0, it started up and did not reconcile the previous resource - which is not okay.
4. Deleted the EtcdCluster and recreated it, now it's working.

Any ideas what the issue might be?

[aobort]

Attempt 1 to reproduce issue:

1. Checked out to v0.0.1
2. Deployed with v0.0.1 image using make install deploy (CRD match v0.0.1)
3. Applied manifest with PodSpec field

Observation: failed to apply (as expected)

Error from server (BadRequest): error when creating "cluster.yaml": EtcdCluster in version "v1alpha1" cannot be handled as a EtcdCluster: strict decoding error: unknown field "spec.podSpec", unknown field "spec.storage.emptyDir"

Attempt 2 to reproduce issue:

1. Checked out to tag v0.1.0
2. Deployed with v0.0.1 image using make install deploy (CRD match v0.1.0)
3. Applied sample manifest with PodTemplate field

Observation:

• statefulset for etcd cluster is created, running normally
• warnings in controller logs:

  manager 2024-05-03T08:10:49Z    INFO    KubeAPIWarningLogger    unknown field "spec.storage.size"                                                                           
  manager 2024-05-03T08:10:49Z    INFO    KubeAPIWarningLogger    unknown field "spec.storage.storageClass"