$ the-wire · showcase
Kueue presubmits stop firing 51 jobs at a time
By RepoJournal · Filed · About Kubernetes · Composed from the cited sources · methodology
The day's load-bearing change is in CI capacity: kueue's presubmits now skip paths they never tested, easing pod_unscheduled_timeout on the shared EKS build cluster, while kube-scheduler picks up a cel-go bump that pins its cost model.
kueue: skip more paths, raise unscheduled timeout kubernetes/test-infra
A default Kueue PR starts 51 presubmits, most of them heavy, even for a kueueviz dependency bump that most of them don't test. The job now skips cmd/kueueviz/, test/e2e/kueueviz/ and hack/ paths and raises the unscheduled timeout, targeting the pod_unscheduled_timeout failures that hit during dependabot runs, cherry-pick waves and /ok-to-test sweeps.
Merge pull request #143439 from cici37/celbump33 kubernetes/kube-scheduler
cel-go moves to v0.33.0 with environment and cost model versions pinned, so admission and validation expressions evaluate against a fixed cost budget rather than the library's default. Anyone vendoring cel-go alongside Kubernetes should match that pin.
Add an on-demand eksctl e2e presubmit for aws-fsx-csi-driver kubernetes/test-infra
Adds pull-aws-fsx-csi-driver-e2e-eksctl, which runs make test-e2e-eksctl on an eksctl cluster with EKS Pod Identity and walks the new cluster tag isolation rollout (disable, backfill, enforce) with negative tests in the controller and IAM. The existing kops job, pull-aws-fsx-csi-driver-e2e, stays in place.
Add an on-demand eksctl e2e presubmit for aws-fsx-csi-driver kubernetes/test-infra
The job is opt-in (/test pull-aws-fsx-csi-driver-e2e-eksctl) because the make target only exists once kubernetes-sigs/aws-fsx-csi-driver#517 merges, and a run takes about 1h40m, so it carries a 3h timeout.
Remove saad and add gnufied kubernetes/org
Ownership churn in kubernetes/org: gnufied replaces saad, with lsviben and onkar717 added to kubernetes-sigs, DavidXU12345 added to the kubernetes org, and another membership batch landing for 2026-10-29. And ci: bump cluster-capacity for k8s 1.37 moves that job to the newer release.