Completeness of semantic convention for k8s resources

[Sprakhar97]

What are you trying to achieve?
Currently only few k8s object have a opentelemetry specification but we want a specification that covers all k8s objects that can be monitored.
What did you expect to see?
All k8s objects are modelled as opentelemetry resource with semantic convention for attributes defining these k8s resources.

Should this be an OTEP?

[tigrannajaryan]

Historically we added semantic conventions on demand. I am not sure if it possible to add everything Kubernetes related in one go due to the volume of new attributes needed. It may trigger lots of discussion and take a long to come to an agreement. So perhaps a series of PRs that add missing Kubernetes related conventions would be easier to get merged.

OTEP is likely unnecessary if it is just a set of new attributes added to semantic conventions. OTEP may be necessary if it wants to suggest significant changes to existing attributes or brings some new ideas about how to structure the attributes.

I am going to unassign this from myself and mark as "help wanted" in case anyone wants to tackle this.

[dmitryax]

@Sprakhar97 this PR propose an alternative set of attributes that can represent any k8s object, even a part of an object like container. PTAL in provide you input

[Sprakhar97]

@dmitryax @tigrannajaryan The above PR talks about a possibility of generalised semantics for crds but I'm talking about other well known k8s object that are missing from k8s specifications like pvc, resourcequota etc., as we know opentelemetry is all about having a common standard shared by various vendors and to be vendor agnostic it should be as precise as possible so that there is very little room for making assumptions thus the k8s otel specification should cover exhaustive list of k8s objects and their attributes and metrics and IMOH k8s being a widely used orchestrator, we should atleast try to cover all the essential entities of k8s.

[dmitryax]

I'm not saying that we shouldn't define other objects like pvc, resourcequota etc. But I don't think we should define all possible objects just to complete the list, it's going to be huge. Probably we can add them as long as they are actually being used by open telemetry components. Do you have any specific component in mind that needs other k8s objects definitions?

And I believe we also need a generalized approach to reproduce any object (even a part of an object) for telemetry sources where object can be anything including CRDs, that's why we have the pull request to add an alternative way to represent any k8s object.