Improve GitLab Kubernetes CPU metrics
We should improve two aspects of GitLab's Kubernetes CPU metrics:
- The data can be polluted by a deployment of the same name, in another namespace. We should filter on the namespace of the project.
- We should switch to using
# of Cores
, instead of% utilized
. This better aligns with how we report cluster metrics, and in general CPU metrics in k8s more broadly. - There is an issue in how we were isolating metrics, based on job type to de-dupe pre-1.7 k8s instances.