Pipelines: [Feature Request] Manual Approval of workflow

[goswamig]

So far KFP has nice UI for DAG. However in production setup we want the final deployment to be approved by human.

This opens up another channel to add notification setup if possible. I am assuming Argo provides those primitives but we might have to expose them via KFP compiler API ?

[amitripshtos]

I raised this kind of request here: https://github.com/kubeflow/pipelines/issues/3101

I also delivered a fully working solution that includes the ability to suspend and resume the pipeline.

Can you please add your opinions to the issue?

Thanks.

[Bobgy]

I agree this is a common need for CI/CD pipelines.

Is suspend and resume the argo way to do it?
I remember seeing in Jenkins, we may make a component that waits for user signal.

Not sure which option will be a better fit. Welcome discussion on this.

[rmgogogo]

I'm more treating this FR as a WORKFLOW feature.
It will have more extensions like

• which role can approve it
• who approved it
• customized workflow logic definition, e.x. 1st people approve it, 2nd audit it or something like it.

So it's more of a workflow feature to me and it can be handled via following options.

1. On top of KFP (another workflow system which has role/user/rule to call KFP API to trigger runs). (preferred)
2. Inside KFP user workload. The pod can sync-wait-for signal (pull other system or msg-queue to wait for msg until continue).
3. code review approval + CI/CD system.

Are you using KFP for ML Pipeline or other case? May I know the detailed use-case?

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[Bobgy]

/lifecycle frozen
this is still a valid feature request