Cannot connect to GCP standalone kubeflow pipeline

[superjcd]

I used the GCP standalone kubeflow pipeline as my pipeline runtime with No Authneitcation(only set the Kubeflow Pipelines API Endpoint), but here i got error like:

Can anyone helps me?

[ptitzler]

Hi @superjcd,

what happens when you click on the Kubeflow Pipelines API endpoint link for your runtime configuration? If the Kubeflow Pipelines Central dashboard doesn't open, then the URL you've entered is likely invalid.

[superjcd]

The dashborad opened correctly:

I guess it's something to do with the GCP access token?

[ptitzler]

I don't have access to GCP and therefore can't really troubleshoot this. Based on the information in https://cloud.google.com/ai-platform/pipelines/docs/connecting-with-sdk, this might be a problem with the set up. Can you confirm whether your environment meets the prerequisites listed in the Before you start section. To determine whether this is an Elyra-specific issue, you can write a small Python program that attempts to connect to your KFP endpoint and run it in the same environment where Elyra is deployed:

import kfp
client = kfp.Client(host='https://...pipelines.googleusercontent.com...')
client.list_pipelines()

If that works it could be an issue with our dummy authentication implementation. To confirm if that's the case, you can run this code

import requests
from urllib.parse import urlsplit

url = urlsplit("https://...pipelines.googleusercontent.com...")._replace(path="").geturl()
requests.get(url, allow_redirects=True)

[superjcd]

I think I've walked through the prerequisites before, but the kfp sdk still throw me a warning and i cant't get any of my pipelines:

But interesting thing is when I add a existing_token (which I got via gcloud auth print-acess-token on cloud shell) to the kfp.Client, then it works perfectly(my windows terminal, not cloud shell):

So I do believe it's related to the GCP access token, but I don't know how to set on Elyra

PS

Some related issuses:

• how to connect kfp.Client to standalone KFP v0.2 deployment on GCP? kubeflow/pipelines#2954
• Unable to use kfp.Client() from local / VM kubeflow/pipelines#4143

[ptitzler]

This is very useful information, Thank you!

As a stop gap it appears that we could add a new authentication provider that takes as input an existing token. Such an approach would be vendor agnostic, because Elyra doesn't need to implement the process of obtaining the token via the vendor's proprietary authentication flow. It's not very user friendly though, because the token cannot be automatically obtained or refreshed and it's up to the user to manage it.