Kubernetes v1.37 'Garhwal' Unveils Pod Checkpointing, Cgroup v2 Transition, and Enhanced Autoscaling

August 26, 2026
Kubernetes v1.37 'Garhwal' Unveils Pod Checkpointing, Cgroup v2 Transition, and Enhanced Autoscaling
  • The upcoming Kubernetes release introduces pod-level checkpoint/restore in Alpha (KEP 5823), enabling capture of full pod state, along with KYAML standardization for manifest formatting and improved watch cache initialization to prevent API server startup bottlenecks.

  • Legacy Linux cgroup v1 is being retired in favor of cgroup v2, with v1 discouraged since version 1.35 and only limited, temporary workarounds available.

  • Metrics API moves from beta to general availability after a long beta period, signaling production readiness and a shift in policy history for beta APIs.

  • The release centers on production readiness and deprecation of older components, aligning Kubernetes with current workloads and AI/ML deployment trends observed by the CNCF.

  • API server startup now enforces rate limits, returning HTTP 429 for excess requests; operators should implement Retry-After and exponential backoff per KEP 4568.

  • The development cycle spanned about 15 weeks with contributions from over 1,700 individuals across 212 companies to deliver new features and improvements.

  • A new status condition ScaledToZero indicates workloads scaled to zero, flipping to NotScaledToZero with reason NotScaledToZero when scaling up occurs.

  • The release emphasizes stronger control plane stability, broader autoscaling (including zero-downscale behavior), and configurable admission policies with offline/disk-based options, while continuing to rely on metrics-based autoscaling and API server resilience.

  • Kube-proxy will discontinue Linux IPVS support starting Kubernetes 1.43, with nftables becoming the recommended forward path.

  • Kubernetes v1.37, Garhwal, introduces 67 enhancements across stable, beta, and alpha stages, including notable autoscaling and API server resilience improvements.

  • KYAML enforces a strict YAML subset to improve manifest parsing and consistency without requiring existing manifests to change.

  • Admission control can now load from disk via a new Beta feature staticManifestsDir, enabling immediate startup policy effects even if etcd becomes unavailable.

Summary based on 2 sources


Get a daily email with more Tech stories

More Stories