Bootstrapping one k8s cloud multiple times causes problems during destroy-controller

[Harry Pidcock]

If you perform the following:

juju bootstrap microk8s a
juju bootstrap microk8s b
juju destroy-controller a

You will be left with a broken controller b because several cluster wide resources such as the ClusterRole, ClusterRoleBinding, MutatingWebhookConfiguration, ValidatingWebhookConfiguration and StorageClasses will be deleted using DeleteCollection(model.juju.is/name=controller)

The problem here is the model.juju.is/name label is not unique, unlike with hosted models (unique by namespace name == model name).

We need to ensure that all global resources outside of an owned namespace are labelled by either model uuid (because their life is tied to a model) or labelled by controller uuid (because their life is tied to a controller).

[Harry Pidcock (hpidcock)]

We need to add a model.juju.is/model-uuid label with the model's uuid to all global resources so they are only deleted when that model goes away.