Kubernetes v1.37 Promotes etcd RangeStream to Beta, Slashing Control Plane Memory Spikes

In Kubernetes v1.37, the upstream project promoted the etcd RangeStream feature (KEP-5966) to Beta, enabling it by default when paired with etcd v3.7 or newer. Developed to overhaul core data-plane retrieval, RangeStream replaces the legacy paginated Range RPC mechanism used by kube-apiserver when fetching large collections of objects from etcd. Instead of assembling full result sets or relying on key-count-based pagination, RangeStream dynamically splits responses into byte-bounded chunks and streams them incrementally over gRPC. This optimization matters immensely for platform teams managing high-density or large-scale clusters containing tens of thousands of Pods, custom resources, or high-churn workloads. Historically, when an API server starts up or needs to rebuild its in-memory watch cache, it must perform a full collection read from etcd. Under the traditional Range implementation, fixed-count key pagination was blind to object payload sizes. A page composed of massive specs or densely annotated objects could consume unpredictable amounts of memory on both etcd and kube-apiserver simultaneously, frequently triggering Out-Of-Memory (OOM) kills during critical failover windows. RangeStream ensures that memory consumption is bounded and deterministic by releasing chunk buffers as soon as they are decoded. This release builds directly upon the broader multi-year effort to modernize Kubernetes control plane scalability and reliability. Alongside complementary enhancements such as resilient watch cache initialization and concurrent watch object decoding, RangeStream addresses the cold-start synchronization bottlenecks and thundering-herd issues that have historically impacted control planes under recovery. As Kubernetes clusters increasingly host memory-heavy AI/ML pipelines and extensive CRD schemas, the byte volume per resource has grown considerably. Hardening the interaction between the API server and the underlying data store is critical for maintaining reliable operations at enterprise scale. In practice, enabling RangeStream requires minimal intervention on Kubernetes v1.37 provided the underlying datastore is updated to etcd v3.7+. The feature gate (`EtcdRangeStream`) is active by default and automatically falls back to classic paginated Range requests if an older etcd version is detected. Platform operators can monitor the `etcd_request_duration_seconds_count` metric with the label `operation="listStream"` to confirm that streaming reads are actively taking place. If compatibility issues arise with intermediate proxies, administrators can temporarily revert to legacy behavior by configuring `--feature-gates=EtcdRangeStream=false` on the API server.