What this tests

retry-logicheaders

Point your client at this specimen and find out — before prod does — whether it backs off, retries, and knows when to stop.

Anatomy

A 503 with Retry-After: 0 — technically valid, semantically a dare. "Service unavailable, come back in zero seconds" is the server's way of saying "I'm broken but please continue hammering me", which a compliant client will do, at full speed, forever, converting a temporary overload into a self-inflicted denial of service on both ends of the connection. A loop with the server's blessing. Real servers emit this when the header is templated from a config value that defaults to zero, or when a health-check response leaks into the error path — the Retry-After was meant for a different context, like a sign saying "no waiting" applied to a queue. A sensible client treats Retry-After as a minimum, not a target, and adds jitter anyway. A naive one becomes the load it was complaining about — the client equivalent of shouting "why is everyone shouting" at a volume of 10,000 req/s.

Replay it

curl -i https://faultydata.com/r/FD-2026-073

// response appears here

Signature

{Status:503 Headers:map[Retry-After:0] Body:{"error":"service unavailable","retry_after":0} Corrupt: Delay:<nil> Flaky:<nil> Sequence:<nil> Drip:<nil> StallAfter:0 Hang:false Reset:false Raw: Phases:[]}

Keep digging

FD-2026-038 — Lying rate-limit headers FD-2026-041 — Retry-After as HTTP-date FD-2026-102 — Rate-limit headers that lie FD-2026-001 — Stripe-style rate limit FD-2026-019 — JSON body, text/plain label FD-2026-069 — Unsolicited 304 to an unconditional request