The Wire · Showcase
AZURE KOPS TESTS FINALLY STABLE, STATEFULSET RECREATION LANDS
By RepoJournal · Filed · About Kubernetes
Three weeks of Azure e2e failures are fixed: kindnet pods can now reach the kube-apiserver, and StatefulSets get their long-awaited recreate update strategy.
Kops shipped the fix that broke its entire Azure test suite [1]. The root cause was elegant and brutal: a security rule scoped to NAT gateway IPs blocked pod traffic from kindnet (which preserves source IPs) to the control plane, coredns never came up, and kubetest2.Up failed 100%. The fix adds a dedicated pod CIDR allow rule. A parallel fix disables the uksouth-3 region [2] where Standard_D4ls_v6 instances have chronic allocation failures. Together these unblock Azure e2e. On the core side, StatefulSets now support recreate update strategy [3] [9] [10], matching Deployment and DaemonSet behavior, with feature gate and codegen already in place. The scheduler gains podgroup-level metrics [8], letting you observe entities in queues alongside individual pods. Minikube's helm addon [4] [5] [6] [7] got a complete type-safety pass: HelmVersion now returns parsed semver and errors instead of silent failures, exit codes are consistent across all runners, and test comments document real use cases instead of implementation details. Test infrastructure cleaned up its own backyard: cloud-provider-gcp reboot jobs deleted [11], DRA node jobs bumped to 2Gi memory [12], and perf-tests got a presubmit to catch benchmark changes early [13].
Action items
- → If running kops on Azure: upgrade immediately to get pod-to-API access working kubernetes/kops [immediate]
- → Review minikube helm addon changes if you have custom helm-based addons kubernetes/minikube [plan]
- → Test StatefulSet recreate strategy in dev; it's feature-gated and ready kubernetes/kubernetes [monitor]
References
- [1] azure: allow kindnet pod CIDR to reach the kube-apiserver ↗ kubernetes/kops
- [2] tests/e2e: disable uksouth-3 in Azure e2e zone pool kubernetes/kops
- [3] Add recreate update strategy to statefulsets (#137187) kubernetes/api
- [4] helm: improve error handling, type safety, and test clarity ↗ kubernetes/minikube
- [5] test: rewrite helm test comments to document real use cases kubernetes/minikube
- [6] test: verify HelmVersion returns ErrHelmNotInstalled kubernetes/minikube
- [7] addons: return parsed version and error from HelmVersion kubernetes/minikube
- [8] Add podgroup level metrics ↗ kubernetes/kubernetes
- [9] Add recreate update strategy to statefulsets ↗ kubernetes/kubernetes
- [10] Add recreate update strategy to statefulsets (#137187) kubernetes/kubernetes
- [11] delete cloud-provider-gcp reboot jobs ↗ kubernetes/test-infra
- [12] DRA: increase resource requirements for E2E node jobs ↗ kubernetes/test-infra
- [13] Add pull-perf-tests-benchmark-list-proto presubmit ↗ kubernetes/test-infra