Coordinate and/or create Spark operator example use case

[chuckbelisle]

Let's have at least one use case for the new Spark feature or even have an active project try it out in one of our demo/sandbox namespaces.

Use Strategy

The following would be the suggested workflow:

• Users will develop their spark applications, likely using PySpark on jupyter notebooks.
• These applications will be containerized and pushed to either AAW-contrib-containers or Arti -> note that we do not have docker available on notebooks, I'm not entirely sure what this would look like
• The users will also create a SparkApplication manifest and submit this to the cluster, including their image in the manifest

To enable this workflow:

• the kubeflow-edit role will need to have rules added for spark-application resources, or something along those lines.
• support for pyspark should be added to kubeflow notebooks

Side Note

Confining Pods to the Appropriate Classification

Currently our implementation is very primitive, so there is no support for pro-b or GPU nodes. To-do would be to enable mutating webhooks on the spark operator and allow tolerations to be injected onto the spark driver pod when it's spawned by the operator. This would allow the spark apps to run on pro-b or gpu nodes.