If a CRD is deleted, stop it's watcher

[bryanl]

When a CRD is deleted, the following is printed to the stderr

E0807 07:41:24.186729   16120 reflector.go:125] k8s.io/client-go/dynamic/dynamicinformer/informer.go:90: Failed to list *unstructured.Unstructured: the server could not find the requested resource

[wwitzel3]

Right now in dynamic_cache.go, currentInformer we create all of the informers using the dc.stopCh. To enable this behavior octant will need to create and store a stop channel for each informer created for a given GVR.

Octant can then lookup the stop channel of an informer for a given GVR and then send the singnal to the stop channel.

[bryanl]

This is more complex than I originally thought: kubernetes/kubernetes#79610

The solution is going to have to work around this inability to remove event handlers from informers. Here is what I'm thinking now:

When a CRD is removed, we'll need to replace the informer. This is not currently possible because DynamicSharedInformerFactory doesn't have a method for this. The factory code isn't complex and could be replicated to include only the features we need (e.g. add and remove informers, start them at create time instead of using start). I haven't looked at the source closely enough to determine what will leak if informers are replaced, but hopefully it should not be a problem if their stop channel is invoked first.

[tedyu]

type SharedInformer interface {

It seems we can add a Stop func to the above interface where the stop channel is closed.