Kubernetes CPU metrics should be counting # of cores used

Right now we are summing the CPU utilization across all nodes. Prior to that we were averaging each data point across all CPU's.

Both are, in fact, wrong I believe. Part of the value (and feature) of k8s is that it will spread your pods around to nodes that have available resources.

You shouldn't have to know, or care, what node they are running and how much heavily utilized (or not) each node is. In fact, it's a great feature of an orchestrator to bin pack each node as effectively as possible for maximum efficiency.

With this in mind, CPU % utilization doesn't really make sense at all. What we should show instead is how many cores your app is consuming. This can be compared to how many it is reserving, etc.