[feature] Configure Kale to use different KFP server #680
Description
Feature Area
/area frontend
/area backend
What feature would you like to see?
- KFP endpoint is configurable: When a user sets a custom KFP API endpoint (host/port), then Kale uses that endpoint for all pipeline submissions and run tracking instead of the default in-cluster service.
- Configuration persists across sessions: When a user saves the KFP endpoint configuration, then it persists across Jupyter kernel restarts and notebook sessions without requiring re-entry.
- Authentication credentials are supported: When the target KFP server requires authentication (e.g., bearer token, cookie, or mTLS), then the user can provide those credentials as part of the configuration, and Kale uses them transparently on each request.
- Connection is validated before use: When a user saves the new KFP endpoint, then Kale performs a connectivity check and surfaces a clear success or error message indicating whether the server is reachable and responding.
- Invalid configuration is rejected gracefully: When the provided endpoint is malformed or unreachable, then Kale displays a descriptive error and does not attempt pipeline compilation or submission.
- Default behavior is preserved: When no custom endpoint is configured, then Kale falls back to the in-cluster KFP service discovery (e.g., ml-pipeline.kubeflow.svc.cluster.local) as it does today.
What is the use case or pain point?
- Let developers use same KFP Server
- Let the users to us different KFP Server flavours/locations on their kubernetes instances.
Is there a workaround currently?
No there is no workaround, if KFP Server is in a different location, or non-standard location it won't work.
Love this idea? Give it a 👍.