0

I have installed Kubeflow on a private GKE Cluster from the GCP Console using below service AI Platform --> Pipelines The kubeflow pipeline UI can be accessed fine.

But my confusion how can I access the UI from internet as there are no ingress/LBs created for this workload as only services of type Cluster IP are created as a part of kubeflow installation

enter image description here

The Kubeflow Pipeline URL is as follows which resolves to a Public IP Address.

 https://xxxxxxxxxxx-dot-us-xxxxxx.pipelines.googleusercontent.com

I don't see any entry in CloudDNS but seems like the URL is exposed similar to how CloudRun/App Engine URLs are exposed.

Under Application TAB from GKE Console , following can be seen

enter image description here

Please ignore the error with respect to "Status" from above image as I changed the number of Nodes to 0.

Zama Ques
  • 1,497
  • 5
  • 26
  • 45
  • in short, you want to understand how is possible that your can access to `https://xxxxxxxxxxx-dot-us-xxxxxx.pipelines.googleusercontent.com` when you don't have any load balancer nor ingress on your GKE cluster ? – Frank Dec 03 '20 at 17:48
  • Yes..Correct..Thanks – Zama Ques Dec 04 '20 at 04:29
  • 1
    According to the [docs](https://cloud.google.com/solutions/machine-learning/setting-up-an-mlops-environment#using_a_standalone_deployment_of_kubeflow_pipelines), the web-based UI and REST APIs for Kubeflow Pipelines are exposed through a public URL using an [inverting proxy and agent](https://github.com/google/inverting-proxy). You should be able to see a proxy agent in your cluster workloads. – Fcojavmelo Dec 04 '20 at 22:38

0 Answers0