Area(s)

area:k8s

Propose new conventions

Description

I would like to propose a new set of semantic conventions to capture node-level system daemon metrics (e.g., kubelet, runtime, pods) exposed by the Kubelet Summary API via Summary.Node.SystemContainers.

Currently, these metrics represent cgroup-level resource trackers on the node but are different from standard application containers because they are not associated with any k8s Pod.

Following the discussion in opentelemetry-collector-contrib#46703, the consensus is to introduce a dedicated namespace (k8s.node.system_container.*) and a self-describing attribute (k8s.node.system_container.name) to cleanly separate these contexts.

Why current k8s.container.* namespace is not suitable

Using the existing k8s.container.* namespace for system containers is problematic because it:

1. Violates Pod Assumptions: It mixes node-level daemon metrics with application container metrics under the same metric name.
2. Breaks Downstream Queries: Users assume that k8s.container.* metrics always belong to a pod and will have standard pod-level fields (e.g., k8s.pod.name, k8s.namespace.name), while system containers do not have these fields.

Therefore, creating a new k8s.node.system_container.* namespace with a specific k8s.node.system_container.name attribute ensures we cleanly separate application container metrics from node-level daemon metrics.

Proposed Attribute

Proposed Metrics

Description: System container level metrics captured under the namespace k8s.node.system_container.

Metric: k8s.node.system_container.cpu.usage

Attributes:

• k8s.node.system_container.name (Required)

Metric: k8s.node.system_container.cpu.time

Attributes:

• k8s.node.system_container.name (Required)

Metric: k8s.node.system_container.memory.working_set

Attributes:

• k8s.node.system_container.name (Required)

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}