The Wire · Showcase
NFTABLES BECOMES KUBE-PROXY DEFAULT, POD RESIZE GETS SCHEDULER PREEMPTION
By RepoJournal · Filed · About Kubernetes
Kubernetes is locking in its networking overhaul while shipping major scheduler and API surface enhancements for 1.37, signaling a fundamental shift in how the cluster routes traffic and scales workloads.
The enhancement desk is moving fast. KEP-5343 commits nftables as the default kube-proxy backend [1], a follow-up to the earlier iptables-to-nftables transition that operators have been testing for releases. This is the moment the old path becomes obsolete. Meanwhile, scheduler preemption for in-place pod resize is landing as alpha in 1.37 [2], which means workloads can now trigger pod evictions when they need to grow without losing data or state. The API surface is expanding too: h2c (HTTP/2 cleartext) container probes are now a KEP [3], and the big one, KEP-6012, proposes a CompositePodGroup API for hierarchical scheduling [4], rolling out in phases to avoid review bottlenecks. Smaller but solid: atomic write volumes get user field support [5]. On the website desk, API reference pages now auto-link back to concept docs that mention them [6], making discovery way cleaner for operators hunting documentation. The community desk logged eligible voters for the 2026 steering election [7] and wrapped the 2025 annual report [8]. Test infra trimmed CI overhead by dropping kueue website verification jobs [9] and removed RangeStream feature gate enablement from the 5k-node performance test [10], keeping the test matrix focused. The apiserver desk landed watch cache refactors that improve encapsulation and fixed benchmark data races [11], technical debt work that keeps observability clean.
Action items
- → Plan kube-proxy migration to nftables as default in 1.37+ kubernetes/enhancements [plan]
- → Review in-place pod resize scheduler preemption design for 1.37 alpha kubernetes/enhancements [monitor]
- → Track CompositePodGroup API RFC through phased implementation kubernetes/enhancements [monitor]
References
- [1] KEP-5343: Make nftables the default kube-proxy backend ↗ kubernetes/enhancements
- [2] KEP 5836: Add KEP for Scheduler Preemption for In-Place Pod Resize (alpha) ↗ kubernetes/enhancements
- [3] Add KEP-5999 for HTTP/2 cleartext (h2c) container probes ↗ kubernetes/enhancements
- [4] KEP-6012: Add initial KEP docs for CompositePodGroup API ↗ kubernetes/enhancements
- [5] KEP-5936: Add user fields to atomic write volumes ↗ kubernetes/enhancements
- [6] Add (automatic) hyperlinks from API reference pages to concept docs that reference them ↗ kubernetes/website
- [7] Initial generation of eligible voters for 2026 steering election ↗ kubernetes/community
- [8] Update 2025 Annual Report with initiatives and updates ↗ kubernetes/community
- [9] Drop CI jobs for periodic-kueue-verify-website-links for releases kubernetes/test-infra
- [10] Remove RangeStream enablement on pull 5k test kubernetes/test-infra
- [11] apiserver/testing: fix data races and ensure uniform node selection in store benchmarks kubernetes/apiserver
FAQ
- What changed in Kubernetes on June 16, 2026?
- Kubernetes is locking in its networking overhaul while shipping major scheduler and API surface enhancements for 1.37, signaling a fundamental shift in how the cluster routes traffic and scales workloads.
- What should Kubernetes teams do about it?
- Plan kube-proxy migration to nftables as default in 1.37+ • Review in-place pod resize scheduler preemption design for 1.37 alpha • Track CompositePodGroup API RFC through phased implementation
- Which Kubernetes repositories shipped on June 16, 2026?
- kubernetes/enhancements, kubernetes/website, kubernetes/community, kubernetes/test-infra, kubernetes/apiserver