$ the-wire · showcase
kops fixes kubeconfig reconcile, scheduler stops leaking nominated nodes
By RepoJournal · Filed · About Kubernetes · Composed from the cited sources · methodology
![kubernetes-prow[bot]](https://avatars.githubusercontent.com/in/992787?v=4)
kops landed a four-PR kubeconfig effort that changes how --use-kubeconfig behaves during update and reconcile, while kubernetes/kubernetes merged a pod group preemption fix for the case where clearing the nominated node name silently did nothing.
kubeconfig: honor update and reconcile connection settings kubernetes/kops
Control-plane version checks now use the shared connection options and those options propagate through reconcile, and kops preserves the local kubeconfig by default when --use-kubeconfig is selected while retaining explicit exports. Generated connection caches are invalidated after direct infrastructure updates, so later reconcile stages rediscover the API endpoint instead of acting on a stale ...
Fix clearing NNN for low prio pods in podgrouppreemption kubernetes/kubernetes
In pod group preemption, getLowerPriorityNominatedPods always returned 0 pods because the candidate name for cluster-wide preemption is "cluster", so the nominated node name of lower priority pods was never cleared. Merged and backported into 1.38; the fix makes the NNN get cleared for those pods as intended.
kubeconfig: preserve explicit API server URLs with NLBs kubernetes/kops
The second half of the kubeconfig work: explicit API server URLs are now preserved when an NLB sits in front of the cluster, rather than replaced by the derived endpoint. kops also clarified endpoint override and TLS identity precedence in the docs, with 30 handshake regression cases added and runtime TLS behavior stated as unchanged.
Graduate the informer latency test to the release-blocking and presubmit 5k jobs kubernetes/test-infra
The informer latency test graduates into the release-blocking and presubmit 5k scalability jobs, with the three experimental runs from the apiserver restart experiment used as the baseline. Anyone touching informers or the apiserver restart path should expect latency regressions to gate PRs rather than surface in periodic results later.
add kube 1-37 sig-node release branch testing kubernetes/test-infra
Prow now runs sig-node release branch testing against the 1.37 branch, added just ahead of the 1.38 code freeze; the long tail also includes a resource-manager presubmit move to us-east1-b after the periodics held there, and KEP-5328 marked implemented in enhancements.
