News › performance
By Zayden R., September 12, 2026
Kubernetes 1.37 introduces native histogram support and gang scheduling in beta. These updates enhance metrics accuracy and optimize resource allocation, particularly for GPU workloads, making it a crucial update for system administrators.
Kubernetes version 1.37 has been released, bringing with it a significant enhancement in observability and resource scheduling. This latest update marks the graduation of native histogram support to beta status, a development that promises to improve the accuracy of Kubernetes metrics while reducing telemetry storage and scraping overhead. This change, enabled by default, replaces the classic Prometheus histograms with dynamic, exponential buckets, addressing longstanding issues such as the 'bucket guessing game' and interpolation error in quantiles.
The transition to native histograms means Kubernetes components can now expose latency and duration metrics with far greater accuracy. By adopting Prometheus native histograms, the system consolidates multiple bucket time series into structured spans within a single series, potentially reducing the number of time series by up to 90%. This is a solid improvement for system administrators who have been grappling with high cardinality and storage costs.
In addition to histogram enhancements, Kubernetes 1.37 also introduces gang scheduling in beta. This feature is designed to optimize resource allocation by reducing GPU idle costs, which is particularly beneficial for workloads requiring simultaneous scheduling of multiple interdependent tasks. With these updates, Kubernetes continues to refine its ability to efficiently manage complex workloads, an essential capability as container orchestration scales in complexity and scope.
For engineers and system administrators, upgrading to Kubernetes 1.37 will require testing and validation of these new features within their specific environments. The improvements in metric accuracy and resource scheduling efficiency are compelling reasons to consider this upgrade as part of regular maintenance and optimization routines.