[k8s v1.30] kube-state-metrics does not collect Sidecar container resource request or limit metrics

[ibayue]

Description

When using kube-state-metrics v2.16.0 (and later) on Kubernetes 1.30, resource request metrics for Sidecar containers (e.g., kube_pod_container_resource_requests_cpu_cores) are missing.
However, on Kubernetes 1.28 with the same kube-state-metrics version, Sidecar container metrics are collected correctly.

Steps to Reproduce

1. Deploy a Kubernetes 1.30 cluster
2. Deploy kube-state-metrics v2.16.0
3. Deploy Pods containing Sidecar containers with explicitly declared resource requests
4. Query Prometheus metrics and notice the missing Sidecar container resource request metrics

Root Cause Analysis

• Kubernetes 1.30 introduces GA for Sidecar lifecycle, separating Sidecar containers from Pod.Spec.Containers, changing the API structure.
• kube-state-metrics still only iterates over Pod.Spec.Containers when collecting metrics, not considering the new Sidecar container fields or Pod.Status.ContainerStatuses.
• This results in missing resource request metrics for Sidecar containers.

Expected Behavior

kube-state-metrics should support Kubernetes 1.30 and later versions by correctly collecting resource requests and limits for all containers, including Sidecar containers.

Suggested Solution

• Update the collection logic to iterate over all running containers (e.g., Pod.Status.ContainerStatuses) instead of just Pod.Spec.Containers
• Or add support for the new Sidecar container fields introduced in Kubernetes 1.30

Environment

• kube-state-metrics version: v2.16.0
• Kubernetes version: v1.30
• Deployment method: Helm Chart / other

---

[nmn3m]

@ibayue Thanks for raising this! Just to clarify:

• Sidecar lifecycle did not GA in v1.30. Native sidecar containers were introduced as an alpha feature in v1.28 (Kubernetes 1.28 blog) and only reached stable in v1.33 (Kubernetes 1.33 release notes).
• They are implemented as init containers with restartPolicy: Always, not as a new field separate from Pod.spec.containers.
• You’re right about kube-state-metrics — it currently iterates over Pod.spec.containers when exporting metrics (kube-state-metrics repo), so native sidecars defined in spec.initContainers aren’t included yet, which can result in missing resource request metrics.

[rexagod]

/triage accepted
/assign @nmn3m

[rexagod]

We've seen implicit behavioural changes for metrics over time in this repository, which, due to their involved nature, may or may not be seen as a regression.

That being said, ideally we start of with the assumption that it isn't, unless a majorly breaking behaviour is exhibited, as the other approach warrants for a patch release, which is reasonable if a regression can be established over the general consensus of the community.

[rexagod]

👋🏼 Revisiting because I saw this added by you in the agenda. Do you need any inputs here?