Refactor kubernetes metricsets to use single calls per endpoint to improve performance. For instance, state_* metricsets fetch the same amount of data to only store some of them according the target k8s object. This can be improved if metricsets can share the input. Some goes for pod, node, volume, container and system metricsets that get the same input from Kubelet's api.

We can find useful the implementation of cloudfoundry integration which behaves like this.