Kubernetes v1.37 Promotes Prometheus Native Histograms to Beta by Default

In Kubernetes v1.37, native histogram support for core metrics has officially graduated to Beta and is now enabled by default. Originating as an Alpha capability under KEP-5808 in v1.36, this update transitions the internal Kubernetes metrics subsystem (k8s.io/component-base/metrics) to leverage Prometheus Native Histograms alongside legacy formats. Instead of generating multi-series static buckets across arbitrary boundaries, the engine now encapsulates exponential, dynamic buckets into structured single time series. Latency and execution duration have long been the backbone of Kubernetes observability, tracking critical paths like API server request latencies and scheduler execution times. Historically, classic Prometheus histograms forced operators and engineers to specify static bucket boundaries upfront. This traditional approach created two pervasive pain points: inaccurate quantile estimations when observed latencies fell outside preconfigured ranges, and severe metric cardinality explosions. Native histograms eliminate the need for guessing bucket distributions, dramatically reducing scrape payload sizes and time-series database storage costs by up to 90% while maintaining roughly 5% relative error bounds across all observations. This graduation reflects the wider cloud-native trend toward high-resolution, low-overhead observability at cloud scale. As Kubernetes clusters support larger fleets and heavier, latency-sensitive workloads—such as real-time microservices and distributed AI training pipelines—the control plane telemetry volume has expanded exponentially. The collaboration between the Kubernetes SIG Instrumentation team and the Prometheus ecosystem highlights an architectural push toward standardized sparse and exponential bucket structures, unifying telemetry models across modern container orchestration frameworks. For platform engineers and SREs, utilizing native histograms is immediate yet safely decoupled. Because Kubernetes v1.37 dual-exposes metrics, existing scraping pipelines configured for classic formats continue functioning without disruption. To unlock native histogram ingestion, teams running modern Prometheus scrapers simply enable the scrape_native_histograms configuration in their scraping jobs. If downstream systems experience parsing issues, administrators can revert individual scrape jobs immediately without restarting cluster components, or disable the NativeHistograms feature gate across the control plane if required. Operators should plan to audit their Prometheus storage configurations to take advantage of reduced TSDB churn.