feat: Rewrite the Tensorboard Controller Charm following the sidecar pattern

[phoevos]

Rewrite the Tensorboard Controller Charm following the sidecar pattern. Summary of changes:

• Add upstream (CRD and Auth) manifests as Jinja templates
• Add helpers for managing K8s resources, retrieving the service environment, and creating Pebble layers following the example of the Jupyter Controller Charm rewrite
• Stop managing the workload using the Pod spec
• Apply K8s CRDs and other manifests directly
• Use Pebble to manage the workload service and environment
• Add remove event handler
• Stop exposing the Tensorboard Controller containerPort as configurable
• Rewrite unit tests
• Add unit tests to test the Pebble layer, K8s resource handler, and update status execution
• Add integration tests to test that the deployment is successful and the controller is working as expected

Upgrade Instructions

No special steps required, apart from trusting the tensorboard-controller application and providing the image on refresh since we changed the oci-image resource to tensorboard-controller-image:

1. Trust the application

   juju trust tensorboard-controller

2. Refresh the charm providing either the path to the local charm (e.g. --path ./tensorboard-controller_ubuntu-20.04-amd64.charm) or the channel if the desired release is published on Charmhub.

   juju refresh tensorboard-controller \
   --resource tensorboard-controller-image=kubeflownotebookswg/tensorboard-controller:v1.7.0

[ca-scribner]

@phoevos can you open (and link in description) a PR on kubeflow-roles-operator that removes the Roles that you migrated over here to the charm? Without that, we'll have some resource conflict issues on final bundle testing

[ca-scribner]

Have you tested upgrading from the previous version to this one? Please include some upgrade instructions here (could be as simple as juju refresh or could have additional steps. Could even be "you can't" :D )

[ca-scribner]

Oh, and we should document how port config was removed. That might actually mess with the upgrade procedures (not sure what a charm does if you upgrade and a config was removed). Good to test and document here

[i-chvets]

I would even go furthern and add upgrade test.

[ca-scribner]

You've added charm-level integration tests - we need to add those to the CI as well

[phoevos]

@phoevos can you open (and link in description) a PR on kubeflow-roles-operator that removes the Roles that you migrated over here to the charm? Without that, we'll have some resource conflict issues on final bundle testing

I don't see anything associated with the TB Controller there (while I do see the TWA-specific roles), am I missing something?

[DnPlas]

@phoevos can you open (and link in description) a PR on kubeflow-roles-operator that removes the Roles that you migrated over here to the charm? Without that, we'll have some resource conflict issues on final bundle testing

I don't see anything associated with the TB Controller there (while I do see the TWA-specific roles), am I missing something?

You are right @phoevos, the kubeflor-roles-operator does not handle any tensorboard-controller aggregation clusterRole. No further action is needed.