Support for the "kubectl top" command across multiple clusters

[XiShanYongYe-Chang]

We have added component fleet-apiserver, which provides users with APIs compatible with the Kubernetes single cluster experience. Related proposal PR #4317

For most kubectl commands, we can obtain the same experience as that of a single Kubernetes cluster.

Currently, for the kubectl top command, there is a phenomenon:

As we can see, the number of resources obtained has doubled. We want to obtain the format of the lower part, but there is no related resource(cpu/memory) use data.

This is because for PodMetrics and NodeMetirics resources, requests are forwarded to the karmada-metrics-adapter component for processing. The difference is that the karmada-metrics-adapter component is not processed in the same way as fleet-apiserver (adding the suffix .clusterspacce.{cluster-name} to resources).

[XiShanYongYe-Chang]

Solution 1: Modify the resource request response of the karmada-metrics-adapter component and add the .clusterspace.{cluster-name} suffix to the pod and node names. like this:

Modification:

1. Change the API return value of the metrics.k8s.io group in the karmada-metrics-adapter component.
2. Adapt to the logic of the federatedhpa-controller to process the suffix of the pod resource.

Impact on Users:
The returned information of resources in the metrics.k8s.io group of the karmada-metrics-adapter component changes, and users are aware of the behavior change.

Note: The return values of custom metrics in the karmada-metrics-adapter component do not change.

[XiShanYongYe-Chang]

Solution 2: Replace metrics.k8s.io with metrics.karmada.io in the karmada-metrics-adapter component. The return value remains unchanged. Add a fleet-metrics-adapter component to work with fleet-apiserver to process requests from the metrics.k8s.io group. The .clusterspace.{cluster-name} suffix is added to the resource name in the returned value.

Modification:

1. Change the API return value of the metrics.k8s.io group in the karmada-metrics-adapter component.
2. Adapt to the logic of the federatedhpa-controller to process the suffix of the pod resource.
3. Add a new component fleet-metrics-adapter.

Impact on Users:
To be compatible with the previous bussiness, users need to use the new metrics.karmada.io group. (Only the group name is changed, and the return value remains unchanged.)

[XiShanYongYe-Chang]

Solution 3: Use the top command provided by karmadactl to allow users to query node and pod information in the member cluster.

Note: Currently, the karmadactl top command can be used to query pod information but cannot be used to query node information.

[XiShanYongYe-Chang]

I shared this topic at today's community meeting.

https://docs.google.com/document/d/1y6YLVC-v7cmVAdbjedoyR5WL0-q45DBRXTvz5_I7bkA/edit#heading=h.h535bf1fgq9k

The comments were that solutions 1 and 2 have a direct impact on user behavior and are not recommended. Solution 3 is recommended.