- Shipped
- November 16, 2026 at 2:29 AM UTC
- Author
- Kamo
- Commit
- f985cff
On 2026-11-16, 48 auto-cert Certificates had been failing for up to 34 days: 41 stuck "Renewing" for domains that no longer exist (kamo.club, theshortterm.rent, joshuatreerentalmanagers.com, domaindaddygo.com, ebonix.club) and 7 that never issued because their DNS points at another host (b11capital.com, breakbeat.store, tech-life.com, makeitpopsigns.com). They held 66 HTTP-01 challenges and their solver pods/services, retrying against the shared Let's Encrypt account. cleanup_orphan_certificates never caught them: it only removes hosts the org registry denies. cleanup_dead_certificates, at startup and every ~30 min: - a Certificate stuck for 7 days (an issuance still running, or not Ready) whose host no longer resolves to any of our public IPs is deleted with its tls-* secret (cascading to requests, orders, challenges and solvers) - static and always-allow hosts, other people's Certificates, and anything that resolves here are never touched - removed hosts go in ConfigMap auto-cert-dead-hosts; ensure_host_certificate skips them until their DNS points here again, then forgets them Also revisionHistoryLimit 1 on every auto-cert Certificate: cert-manager had kept all 740 CertificateRequests and Orders. A dry run against the live cluster would remove 43 now; 3 more are under the 7-day mark. Tests run the real functions against a fake kubectl and DNS.
