Kubernetes v1.37 Graduates Native Histograms to Beta, Enabled by Default

Native Histograms Reach Beta

Native histogram support for Kubernetes metrics has graduated to Beta and is enabled by default in Kubernetes v1.37. This feature, previously introduced as Alpha in Kubernetes v1.36 under KEP-5808, brings high-resolution, low-cardinality observability to Kubernetes metrics.

Addressing Limitations of Classic Histograms

Historically, Kubernetes observability for duration and latency metrics relied on classic Prometheus histograms, which required defining static bucket boundaries. This approach led to challenges such as loss of visibility when workload latency profiles changed, high cardinality and storage costs due to each bucket boundary being a separate time series, and interpolation errors in quantile calculations.

Classic histograms could multiply the number of time series by 10 for a histogram with 10 buckets across multiple labels, increasing memory consumption and storage costs.

Benefits of Prometheus Native Histograms

Prometheus Native Histograms replace static user-defined buckets with dynamic, exponential buckets. Instead of emitting a separate time series for each bucket boundary, a native histogram is stored as a single time series with a rich schema.

This new approach offers high resolution automatically, as exponential buckets dynamically adjust to any value range without pre-configured boundaries. It also reduces the number of time series by up to 90% by consolidating buckets into structured spans within a single time series.