What this tests
monitoringsemantics
Point your client at this specimen and find out — before prod does — whether it surfaces a real signal, or reports green while users suffer.
Anatomy
For four days in November 2023 Cloudflare's edge kept happily serving traffic while its control plane — the dashboard, the APIs, the logging and analytics — faceplanted, because the Flexential data centres hosting it lost power and the "independent" control plane turned out to have opinions about which buildings it lived in. A gorgeous failure shape: the product works, the management of the product does not. Customers could serve traffic but couldn't configure, observe, or bill it. If your monitoring treats "the API returns 200" and "the API is usable" as the same check, this is the exhibit that ends that complacency.
Source: Cloudflare blog · 2023-11-02
Replay it
curl -i https://faultydata.com/r/FD-2026-091
// response appears hereSignature
{Status:500 Headers:map[Content-Type:application/json] Body:{"success":false,"errors":[{"code":7003,"message":"control plane unavailable"}],"result":null} Corrupt: Delay:<nil> Flaky:<nil> Sequence:<nil> Drip:<nil> StallAfter:0 Hang:false Reset:false Raw: Phases:[]}
Keep digging
FD-2026-023 — Global 502s from a bad WAF rule (Cloudflare, Jul 2019) FD-2026-048 — Origin unreachable (Cloudflare 523) FD-2026-096 — The endpoint that never came back (Code Spaces, Jun 2014) FD-2026-094 — Stale crew-scheduling data melts operations (Southwest, Dec 2022) FD-2026-103 — Zombie health check (200 while dead inside) FD-2026-105 — HTTP 520 — the Cloudflare unknown-error junk drawer