The name should just be container.cpu.usage

The metric. classifier is for our YAML database, not the metric name. (Sorry for confusion)

Same comment for all other metrics.

is this needed separate from system.cpu.*?

(there was some related discussion in open-telemetry/opentelemetry-specification#2388)

Thank's @trask! That's interesting one and it's quite similar to open-telemetry/opentelemetry-specification#2388 (comment).

I will try to share my view on this :).
So, from infrastructure's observability point of view, container metrics would be collected from outside the containers themselves (as a best practice).
For example by using the Docker runtime's APIs or Kubelet's API's.

So the point here is that it does not make a lot of sense to treat a container as a host and collect its metrics by running a collector inside the container.

So for metrics collected through the runtime/orchastrator APIs I think we need to be specific and have a resource specific namespace like container.cpu, pod.cpu etc.

In many cases the CPU/Memory resources are limited as well:

• https://docs.docker.com/config/containers/resource_constraints/#cpu
• https://kubernetes.io/docs/tasks/configure-pod-container/assign-cpu-resource/#specify-a-cpu-request-and-a-cpu-limit

In very specific usecases, if someone wants to treat a container as a host (kind uses containers' as hosts) then the collector should be running on the container/host directly and report metrics under the system.* namespace. In that case the resource is a "host" not a container from the observation/collection point of view.

Also related to #226 (comment).

Should be a counter instead (since values are additive)?

https://github.com/open-telemetry/opentelemetry-specification/blob/065b25024549120800da7cda6ccd9717658ff0df/specification/metrics/supplementary-guidelines.md#instrument-selection

should it be a counter?

should it be a counter? (https://github.com/open-telemetry/opentelemetry-specification/blob/065b25024549120800da7cda6ccd9717658ff0df/specification/metrics/supplementary-guidelines.md#instrument-selection)

and here: it's a counter