Prometheus query used for pod metrics is:

container_cpu_load_average_10s{pod="

", namespace="{#NAMESPACE}"}

but this retrieve with the value set to "sum" a value that is double because it read also the parent cgroup value

should be better to use