Kubernetes 1.37 Adds Autoscaling and Memory Cuts

Kubernetes 1.37 makes Dynamic Resource Allocation’s extended resource support stable, letting DRA drivers satisfy traditional extended resource requests without a separate device plugin. It also moves ResourceClaim status updates for device details, including network interface data, and device taints and tolerations to stable, with further alpha, beta and stable improvements in the release.

Kubernetes 1.37 adds beta support for scaling workloads down to zero replicas with HorizontalPodAutoscaler when using suitable object or external metrics. This helps queue-based and batch workloads save resources, but request-driven services still need buffering because Services do not queue traffic when no Pods are running.

The video explains how to connect the Sidero Labs documentation MCP server to Claude, either through Claude.ai or Claude Code in the terminal. It says the setup helps ground answers in official documentation and reduce inaccurate responses, and points viewers to the server at docs.siderolabs.com/mcp.

etcd RangeStream graduates to beta in Kubernetes 1.37 and, with etcd v3.7, reduces memory use and makes large list reads more predictable. The streaming RPC breaks large responses into chunks, which lowers peak memory pressure on both etcd and the API server.