[feature] support customizing Tensorboard images

Feature Area

/area frontend

What feature would you like to see?

Allow users to customize the tensorboard image they want to use in KFP component visualizations. https://www.kubeflow.org/docs/components/pipelines/sdk/output-viewer/

There are mainly two ways I can think of to support this feature:

Option 1:

1. add an input textbox for image in Tensorboard plugin that users can directly edit
2. (optional) let the textbox remember historic entries

Option 2 - let component output decide

1. when a component emits visualization metadata, add a new field image they can specify tensorboard image
2. KFP UI loads the mlpipeline-ui-metadata and default to show the image option in component metadata

What is the use case or pain point?

Existing issues that can be addressed by this feature: https://github.com/kubeflow/pipelines/issues/5449

It’s currently not possible to choose any tensorboard image version unless it is hard-coded in KFP UI.

Users or pipeline components know the best about which tensorboard version/image type can visualize its output, not the KFP cluster operator.

Is there a workaround currently?

No

---

Love this idea? Give it a 👍. We prioritize fulfilling features with the most 👍.

[Bobgy]

/assign I will work on this shortly, the best way to help is confirm whether above suggested options fit your use-case.

I will implement option 2 directly, because that seems like the fundamental requirement.

[ConverJens]

@Bobgy Being able to pass full image paths is a must for most on-prem installations since they use an internal docker registry. For instance, our k8s klusters have no external access and any tensorflow-based image needs our internal certificates added.