[receiver/k8s_cluster] Missing node metrics for `k8s.node.condition_out_of_disk`, `k8s.node.allocatable_pods` and `k8s.node.allocatable_hugepages-*`

[atoulme]

Component(s)

No response

What happened?

Description

When performing the transition to pdata, I unwittingly didn't account for some newer allocatable resource metrics associated with nodes. The receiver accepts resource names as strings and doesn't perform a validation on them, and therefore they are not caught by the receiver at collection time.

Steps to Reproduce

Try to run the receiver with the following configuration:

clusterReceiver:
  enabled: true
  config:
    receivers:
      k8s_cluster:
        node_conditions_to_report:
          - Ready
          - DiskPressure
          - MemoryPressure
          - PIDPressure
          - NetworkUnavailable
          - OutOfDisk
        allocatable_types_to_report:
          - cpu
          - memory
          - ephemeral-storage
          - storage
          - pods
          - hugepages-1Gi
          - hugepages-2Mi

Expected Result

The collector should emit the following metrics:

• k8s.node.condition_out_of_disk
• k8s.node.allocatable_pods
• k8s.node.allocatable_hugepages_1_gi
• k8s.node.allocatable_hugepages_2_mi

Actual Result

The collector doesn't emit those metrics.

Collector version

0.80.0

Environment information

Environment

OS: (e.g., "Ubuntu 20.04")
Compiler(if manually compiled): (e.g., "go 14.2")

OpenTelemetry Collector configuration

No response

Log output

No response

Additional context

No response

[github-actions]

Pinging code owners for receiver/k8scluster: @dmitryax. See Adding Labels via Comments if you do not have permissions to add labels yourself.

[dmitryax]

@atoulme can this be assigned to you or do you want someone else's help?

[atoulme]

I'm taking a first pass at pods and a few other things but we have a problem with how hugepages is just a wildcard prefix.

[dmitryax]

we have a problem with how hugepages is just a wildcard prefix

Those can be added outside of mdatagen. We need a comment about those metrics in metadata.yaml.

[dmitryax]

Later, I'd say we should move the fluid part of the metric names to an attribute, but that'll be a breaking change. For now we need a bugfix to restore the previous behavior

[atoulme]

Understood. I will break the fix in multiple PRs, starting with pods. I also see we don't catch the node condition OutOfDisk which is apparently a custom cloud provider provided metric.

[dmitryax]

It'd be great to apply an extended configuration to the e2e config so it produces as much metrics as possible