grafana/k8s-monitoring-helm k8s-monitoring-4.2.1

on GitHub

What's changed

• Add the root, host-network, host-storage, and host-cgroup collector presets and deprecate the privileged preset. root sets the privileged root securityContext, host-network sets hostPID/hostNetwork/dnsPolicy, and host-storage/host-cgroup mount host paths. privileged is unchanged and still works, but cannot be combined with these new presets. (#2796) (@TylerHelmuth)
• Collector presets now append their alloy.mounts.extra and controller.volumes.extra entries instead of overwriting, and the OpenShift SecurityContextConstraints allows host networking and ports when a collector sets hostNetwork. (#2796) (@TylerHelmuth)
• Add per-instance extraDiscoveryRules to the Service Integrations feature, allowing pre-scrape discovery.relabel rules — e.g. a labelmap from pod annotations — to enrich an integration's scraped metrics with custom labels while the __meta_* discovery labels are still present. Set on each integration instance (the targets it discovers are unique); available on all nine Kubernetes service-discovery integrations, with Istio set per sidecar and istiod. (@bradleypettit)
• Fix kubeletResource.extraDiscoveryRules and kubeletProbes.extraDiscoveryRules being silently dropped. (#2743) (@bradleypettit)
• Surface deploy: false defaults for telemetryServices.beyla and telemetryServices.sdkInjector in the top-level values, matching the other telemetry services. This prevents them from deploying unintentionally on Helm versions affected by a subchart value coalescing regression (helm/helm#32132). (#2781, #2782) (@petewall)
• Change the default remote configuration pollFrequency from 5m to 30s. (@TylerHelmuth)
• Add CLUSTER_NAME environment variable to remote-config-enabled collectors alongside NAMESPACE and POD_NAME. (#2775) (@petewall)
• Fix OTLP destinations producing ;-joined duplicate label values on target_info (e.g. k8s_cluster_name="c;c") when a scrape target exposes its own target_info metric. Flat resource attributes from the conversion (service_namespace, deployment_environment, deployment_environment_name, k8s_cluster_name) are reconciled into their dotted equivalents and dropped when the values match. (#2786) (@mbaykara)* Fix OTLP destinations deriving the wrong service identity for metrics converted from a Prometheus scrape. The conversion sets service.name to the full job label (e.g. namespace/workload from Grafana Beyla), breaking service-to-logs correlation in Application Observability. A new normalization step after otelcol.receiver.prometheus prefers the explicit service_name label; OTLP-native telemetry is unaffected. (#2783) (@mbaykara)

Chart released at https://github.com/grafana/helm-charts/releases/tag/k8s-monitoring-4.2.1

Full changelog: CHANGELOG.md