fix: Introduce k8s-service-info between katib-controller and katib-db-manager

[orfeas-k]

This PR introduces a k8s-service-info between katib-controller and katib-db-manager. This addresses the issue below. Read the summary comment for a quick overview of the issue.

Note that this means that katib-controller will require a relation to katib-db-manager in order to start its pebble service, which we will need to introduce in the latest/edge bundle.

Testing

1. Deploy AKS cluster following the guide. You can always check the deploy-to-aks workflow for the commands the CI follows. Note that Juju 3.5 should be used but right now is broken thus 3.4 is probably a better option (didn't try this though myself). Another option could be using channel 3/stable which points to 3.5.0 which is not broken (apart from the issue with kubeflow-dashboard which you can workaround by refreshing to upstream image).
2. Update the latest/edge bundle.yaml as follows

@@ -64,11 +64,12 @@ applications:
     _github_repo_branch: main
   katib-controller:
     charm: katib-controller
-    channel: latest/edge
+    channel: latest/edge/pr-185
     scale: 1
     trust: true
     _github_repo_name: katib-operators
     _github_repo_branch: main
+    base: [email protected]
   katib-db:
     charm: mysql-k8s
     channel: 8.0/edge
@@ -79,11 +80,12 @@ applications:
     _github_dependency_repo_branch: main
   katib-db-manager:
     charm: katib-db-manager
-    channel: latest/edge
+    channel: latest/edge/pr-185
     scale: 1
     trust: true
     _github_repo_name: katib-operators
     _github_repo_branch: main
+    base: [email protected]
   katib-ui:
     charm: katib-ui
     channel: latest/edge
@@ -297,6 +299,7 @@ relations:
   - [istio-pilot:istio-pilot, istio-ingressgateway:istio-pilot]
   - [istio-pilot:ingress, tensorboards-web-app:ingress]
   - [istio-pilot:gateway-info, tensorboard-controller:gateway-info]
+  - [katib-controller, katib-db-manager]
   - [katib-db-manager:relational-db, katib-db:database]
   - [kfp-api:relational-db, kfp-db:database]
   - [kfp-api:kfp-api, kfp-persistence:kfp-api]

3. Deploy kubeflow using the tox environment tests are using as well

tox -e test_bundle_deployment-latest -- --model kubeflow --keep-models -vv -s

This uses the edited bundle.yaml

5. Create a notebook and run katib UAT

Closes canonical/bundle-kubeflow#893

[NohaIhab]

thanks @orfeas-k, left some comments

this k8s_service_info_requirer_component module looks reusable, I think it can live in chisme, WDYT?

[orfeas-k]

Replied to all comments. Regarding k8s_service_info_requirer_component, it's definitely a resuable component since I copied that from mlmd/envoy but we have the following limitation canonical/charmed-kubeflow-chisme#94 which we would have to solve first and then see if it makes sense to add this to chisme.

[NohaIhab]

thx @orfeas-k !

The base branch was changed.

[NohaIhab]

On second thought discussing with @orfeas-k , we need to merge this PR to main before sending a follow-up PR to modify the tests to not use the local charm. This is because we need the new version of katib-db-manager to be released to latest/edge before we can do the change in the tests.