@open-telemetry/specs-semconv-approvers should the .count suffix be applied here based on the guideline from

We don't use a pluralized metric name here but I want to double check if this rule should be taken into account.

It would be preferable if we define this metrics as is to minimize breaking changes of the Collector.

I added the replication_controller metrics in this PR since the names are already the same and group well here.

In any case, we need to have this aligned with #1637 and #1644.

Taking back my approval until discussion in #1637 (comment) is resolved

I have updated the PR to include the statefulset and hpa metrics since those are related to the consistency discussion (#1637 (comment)).

If want to achieve consistency while also being semantically correct we either need to use the _pods or the _replicas suffix. I chose to use the _pods suffix as more "user friendly".

Providing bellow the current state (copied from the PR's description):

• k8s.deployment.desired_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.deployment.available_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.replicaset.desired_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.replicaset.available_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.replication_controller.desired_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.replication_controller.available_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.statefulset.desired_pods (emitted by collector) [ ✅ not a breaking change for the Collector]
• k8s.statefulset.ready_pods (emitted by collector) [ ✅ not a breaking change for the Collector]
• k8s.statefulset.current_pods (emitted by collector) [ ✅ not a breaking change for the Collector]
• k8s.statefulset.updated_pods (emitted by collector) [ ✅ not a breaking change for the Collector]
• k8s.hpa.desired_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.hpa.current_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.hpa.max_pods (emitted by collector) [⚠️ BREAKING change for the Collector]
• k8s.hpa.min_pods (emitted by collector) [⚠️ BREAKING change for the Collector]

Breaking changes between the Collector and the Semconv will be listed in the k8s migration guide once #1636 is merged.

We have a few minor discrepancies in resource/metric names. While I don't think we need to change any existing names, I would like to note them down for future reference.

1. Short/Full names for k8s type - HPA seems to be the only resource type that uses the short name. The full object kind can be a mouthful, but best to stick with the full names as long as they are a reasonable length.
2. k8s object kinds use PascalCase. Some resources mirror this with underscores in the attr/metric names (e.g. k8s.replication_controller for k8s obj ReplicationController), while others don't (e.g. k8s.statefulset for k8s obj StatefulSet). I think we should be consistent with the underscore convention.

I think this observation from @dashpole might be relevant to some of the metrics here. Most of these timeseries seem meaningfully additive

I think this observation from @dashpole might be relevant to some of the metrics here. Most of these timeseries seem meaningfully additive

As discussed in the K8s SemConv WG meeting today, we agree on making the change to updowncounter for these metrics (I think this will also apply to #1649). I will update the PR(s) to reflect this.

@open-telemetry/specs-semconv-maintainers could you also take a look please?

note: I'm suggesting we introduce the metrics as-is and follow-up for enhancing their descriptions more to cover their specific relations with resources.

I added resource correlation note per metric with 1f91666 according to the discussion at #1649.