The Wire · Showcase
KUBERNETES 1.37 INCOMING, KOPS STRIPS UNNECESSARY CREDENTIALS
By RepoJournal · Filed · About Kubernetes
Kubernetes 1.37 sneak peek is live on the website, while kOps 1.37's removal of gossip DNS and protokube unlocks a major security hardening: stripping API credentials from control-plane nodes that never needed them.
The website shipped a sneak peek blog for Kubernetes 1.37 [1] alongside a Gateway API v1.6 release announcement [2], setting expectations for what's coming in the next major version. Simultaneously, kOps is tightening the screws on cloud permissions [3]. The removal of gossip DNS and protokube in kOps 1.37 means DigitalOcean, Hetzner, and Scaleway credentials no longer get exported to control-plane user data, and Route 53 permissions drop from the AWS control-plane instance role when `useServiceAccountExternalPermissions` is enabled. This follows the principle that dedicate services should have dedicated credentials, not shared node-wide access. Over in the autoscaler, the team is undergoing a mechanical migration to extract core Cluster Autoscaler logic into a dedicated library [4], which includes removing the benchmark CI job and migrating testing and reference providers [7]. The ingress-gce controller picked up a critical fix for NEGBinding ownership [5], preventing multiple CRs from fighting over the same NEG by name, plus an allowMissing implementation for filtered informers to handle supervisor nodes without labels [6]. Finally, the release tooling got a patch bump for sigstore-go [8] that resolves GHSA-wqqc-jjcq-vfxm.
One email a day. Unsubscribe in one click.
Keep up with Kubernetes in about 3 minutes a day: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
One email a day. Unsubscribe in one click. Read a past issue →
Action items
- → Review kOps 1.37 release notes for credential management changes, especially if you run external DNS or Route 53 kubernetes/kops [plan]
- → Update any custom NEGBinding deployments to account for single-syncer-per-NEG enforcement kubernetes/ingress-gce [plan]
- → Monitor cluster-autoscaler library migration if you maintain custom cloud providers kubernetes/autoscaler [monitor]
References
- [1] Add Kubernetes 1.37 sneak peek blog ↗ kubernetes/website
- [2] Create a blog post for gateway api v1.6 release. ↗ kubernetes/website
- [3] Narrow instance cloud permissions to actual consumers ↗ kubernetes/kops
- [4] Merge pull request #10002 from Choraden/remove-cluster-autoscaler-core kubernetes/autoscaler
- [5] Merge pull request #3177 from anyush/negbinding-conflicting-crs kubernetes/ingress-gce
- [6] Merge pull request #3188 from priyapande/pr1-allow-missing kubernetes/ingress-gce
- [7] Remove testing and reference providers migrated to sigs.k8s.io kubernetes/autoscaler
- [8] Bump github.com/sigstore/sigstore-go from 1.2.0 to 1.2.1 ↗ kubernetes/release