The Wire · Showcase
KOPS PLUGS DIGITALOCEAN SSH KEY LEAK, REGISTRY CUTS 404 NOISE
By RepoJournal · Filed · About Kubernetes
DigitalOcean clusters were leaving orphaned SSH keys behind on teardown. kops shipped the fix [ref:6], test-infra is ready to land it [ref:4], and registry.k8s.io rolled out early 404 rejection to slash CloudFront noise [ref:12].
The kops fix [1] adds SSH key cleanup to the DigitalOcean resource deleter, matching what AWS and OpenStack already do. kops names each key after the cluster and fingerprint, uploads it via `dotasks/sshkey.go createKeypair`, but never tore it down. Test-infra's pending change [2] drops the preset-do-ssh label from 8 DigitalOcean e2e jobs and switches to throwaway keypairs per cluster, but it's blocked until kops lands the deletion logic. Over in registry.k8s.io, v0.7.0 shipped [4] with a dedicated Cloud Build service account and AWS CDN support, followed by an aggressive 404 rejection layer [3] that returns failures early instead of letting CloudFront existence checks drain budget. Kube-state-metrics fixed a race condition in config reload serialization [5] where three separate viper watchers were unsynchronized on the same context, and kops patched Akamai instance tag filters to use `+contains` predicates instead of rejected equality checks [6]. The k8s.io infrastructure bumped archeio to 0.7.0 [7] with BQ analytics datasets and ASN capture from GCP [8].
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
- → Merge kops DigitalOcean SSH key deletion [ref:6] before test-infra lands the job changes [ref:4] kubernetes/kops [immediate]
- → Monitor registry.k8s.io 404 rates post-deployment to validate CloudFront savings [ref:12] kubernetes/registry.k8s.io [monitor]
- → Verify kube-state-metrics config reload fix [ref:1] in canary before prod rollout kubernetes/kube-state-metrics [plan]
References
- [1] digitalocean: delete SSH keys when deleting a cluster ↗ kubernetes/kops
- [2] kops: drop preset-do-ssh from DigitalOcean jobs kubernetes/test-infra
- [3] return 404s very early ↗ kubernetes/registry.k8s.io
- [4] v0.7.0 ↗ kubernetes/registry.k8s.io
- [5] fix: serialize config reloads so one does not strand the other ↗ kubernetes/kube-state-metrics
- [6] fix(akamai): Fix Akamai (Linode) Instance tag filter function ↗ kubernetes/kops
- [7] bump archeio to v0.7.0 ↗ kubernetes/k8s.io
- [8] create bq dataset and pass asn values to archeio ↗ kubernetes/k8s.io