grafana/k8s-monitoring-helm k8s-monitoring-4.1.5

on GitHub

What's changed

• Fix OpenCost failing to deploy with duplicate entries for key [name="CONFIG_PATH"] when customPricing is enabled. The GKE GCP-provider workaround no longer sets CONFIG_PATH via extraEnv (which collided with the CONFIG_PATH the OpenCost chart sets for custom pricing); instead it mounts a writable emptyDir at OpenCost's default config path (/var/configs). (#2692) (@petewall)
• Require labelMatchers (rather than accepting a namespace alone) when connecting Cluster Metrics (kube-state-metrics), Host Metrics (Node Exporter, Windows Exporter, Kepler), or Cost Metrics (OpenCost) to an existing service that is not deployed via telemetryServices. Previously a namespace-only configuration rendered an empty label = "" selector that matched every pod in the namespace. (@petewall)
• Add a best-effort validator that uses Helm lookup to confirm the configured labelMatchers actually select running pods in the specified namespace for an existing kube-state-metrics, Node Exporter, Windows Exporter, Kepler, or OpenCost. The check is skipped during helm template/--dry-run (no cluster connection) and only runs when a namespace is set. (@petewall)
• Update Alloy Operator to 0.5.9, Beyla to 1.16.8, and OpenCost to 2.5.23 (@petewall)* Add a best-effort validator that checks (via Helm lookup) for an existing Node Exporter using the same port when deploying the bundled telemetryServices.node-exporter. If a port conflict is detected, the install fails with guidance to either skip the deployment and point Host Metrics at the existing Node Exporter, or deploy on a unique port. (@petewall)

Full changelog: CHANGELOG.md