Deprecate TfPort and set default port for users

[DjangoPeng]

In the current CRD, we have a TfPort field here for TensorFlow replicas. While the pod template has a Port field as well.

According to the discussion of TfPort here, we'd like to deprecate it and set a default port for users.

/cc @jlewi @gaocegege @ScorpioCPH

[gaocegege]

Now our pod template spec is not required because we have TfImage and TfPort, if we deprecate it, we should add validation and defaults.

[jlewi]

+1.

For validation we should look at the validation features now available with CRDs. I think we can provide an OpenAPi spec and it will K8s will automatically validate specs.

[gaocegege]

@jlewi

I like the idea! The feature is beta in v1.9. I think we should update the requirement in this repo and kubeflow if we use it.

[gaocegege]

/assign gaocegege

[jlewi]

Opened #437 to use OpenAPI validation.

[jlewi]

Do you think we should fix this issue or just wait for the V2 APi (kubeflow/community#30)? My vote would be for fixing it as part of the v2 API.

[gaocegege]

I will fix the issue in v1alpha2 😄

[gaocegege]

Closed by #492