A use case would be you are developing your models locally e.g. in an IDE but deploying on your cluster and you want to send your prediction requests.
Kubeflow should be exposed through a reachable ingress that is secured (e.g. with IAP or Basic Auth). So the client library should obtain a credential (e.g. a JWT) and attach it to the request.
jtfogarty
commented
4 years ago
How about attaching authorization credentials to prediction requests? e.g here: https://github.com/kubeflow/fairing/blob/642530567d0255f88b0f83b0a40b3325c303bfa2/fairing/ml_tasks/tasks.py#L69
A use case would be you are developing your models locally e.g. in an IDE but deploying on your cluster and you want to send your prediction requests.
Kubeflow should be exposed through a reachable ingress that is secured (e.g. with IAP or Basic Auth). So the client library should obtain a credential (e.g. a JWT) and attach it to the request.
The pipelines SDK has an auth module https://github.com/kubeflow/pipelines/blob/master/sdk/python/kfp/_auth.py
And there is an open issue to support IAP kubeflow/pipelines#1104