Kubernetes: Storage Cleanup and Scheduler Fixes
The Kubernetes project merged 15 pull requests on May 15-16, 2026, focusing on API server storage improvements, scheduler preemption fixes, and kubelet reliability enhancements. Notable changes include storage metrics refactoring, CPU manager checkpoint reverts, and Windows test fixes.
Duration: PT2M17S
Transcript
Good morning, this is your Kubernetes development briefing for May 16th, 2026.
The Kubernetes team merged 15 pull requests yesterday with significant focus on storage layer improvements and scheduler reliability.
Ben Luddy merged KEP-3926 implementing corruption detection for unsafe delete operations in the API server storage layer, adding 429 lines across 12 files to ensure data integrity during object deletion.
The storage system saw major refactoring as Yedou37 moved storage list metrics to a shared package, consolidating metric definitions from etcd3 and cacher components. Shadowofs contributed storage key function cleanup, extracting default KeyRootFunc setup into a helper for clearer code organization.
Scheduler improvements came from mm4tt who fixed preemptor eligibility behavior in pod group preemption, ensuring consistent preemption logic across scheduling scenarios. Argh4k added selective CPU profiling capabilities to scheduler benchmarks, enabling targeted performance analysis.
Lukasz Wojciechowski reverted the generalized CPU manager checkpoint changes due to failing tests, removing 614 lines to restore stability. Marosset fixed flaky Windows kubelet unit tests by improving log directory cleanup handling.
Additional reliability improvements included Hoteye's kubelet image manager logging cleanup, Saschagrunert's direct I/O fix for ContainerMetrics tests on overlayfs, and Sohankunkerkar's test helper standardization.
Kubeadm received a new warning when users configure kube-proxy with IPVS mode, and kubectl proxy now displays appropriate warnings. A documentation fix corrected an autoscale example inconsistency.
Carlory improved goroutine management in the resource pool status request controller, and several test fixes addressed flaky behavior in scheduler and kubelet components.
What's next: Watch for continued storage layer unification work and scheduler preemption improvements in upcoming releases.
That's your Kubernetes development update. I'm your host, reporting from the community.