Kubernetes: DRA Stability and Scheduler Backports
Most activity was backports to releases 1.34 through 1.37 fixing device allocation status flapping and taint eviction, scheduler namespace selectors, and Job expectations. Open proposals address legacy Service cluster IPs, failed CRD migrations, and Workload-aware Scheduling beta.
Duration: PT2M35S
Episode overview
This episode is a short developer briefing from Kubernetes.
It explains recent repository work in plain language.
- Show: Kubernetes
- Published: 2026-10-17T13:03:17Z
- Audio duration: PT2M35S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning, it's October 17, 2026.
Almost all of yesterday's merges were backports to releases 1.34 through 1.37, centered on device resource allocation and scheduling correctness.
First, device resource allocation. Three cherry-picks of the same fix stop pod resource claim status from flapping between syncs, where each sync dropped entries it didn't include. That's PR 142025 and related backports. Paired with that, PR 142492 includes extended resource claims in pod iteration and taint…
Second, scheduling and batch. The scheduler regression fix in PR 141563 corrects affinity and anti-affinity with negative namespace selectors, so pods in excluded namespaces no longer skew filtering and scoring. It was backported to 1.34 and 1.35. For batch, the Job controller fix in PR 142390 and its siblings…
Finally, storage reliability for old objects. Open PR 143113 fixes legacy Services losing their default cluster IPs on watch and status reads, which could trigger deletion of allocated addresses and duplicate assignment. Open PR 143103 keeps stored versions after a failed custom resource migration so unmigrated…
Looking ahead, the kubelet checkpoint change in PR 141039 to store the full pod…
Nearby episodes from Kubernetes
- Watch Cache Scaling and Quantity Hardening
- Node Env Fixes and Control-Plane Reliability
- Gang Scheduling Speedups and API Server Efficiency
- Scheduler Hardening, Job Counting Fixes, and kubectl WebSockets
- Watch Costs and Pod Group Preemption
- Weekly Recap - Go 1.27 Hardening, Scheduling and Storage Correctness
- Storage Efficiency and Scheduler Resilience
- API Stabilization and Controller Reliability Fixes