Limites par utilisateur et par IP, 429 avec Retry-After, limites selon l’offre.
Spec — Rate limiting
Status: draft · Created: 2026-11-10
1. Context and goal
The API protects itself and its customers from bursts and abuse, with limits that clients can see and plan around.
2. Functional requirements (EARS)
- FR-1: THE SYSTEM SHALL count requests per authenticated user and per client IP in a sliding window of 60 seconds, shared by all instances.
- FR-2: WHERE a request is authenticated, THE SYSTEM SHALL apply the limit of the caller's plan (for example 60, 600 and 6,000 requests a minute), and 30 requests a minute per IP to anonymous requests.
- FR-3: WHEN a client exceeds a limit, THE SYSTEM SHALL answer 429 with
Retry-Afterin whole seconds until a request would be accepted, without running the handler. - FR-4: THE SYSTEM SHALL send
RateLimit-Limit,RateLimit-RemainingandRateLimit-Resetheaders on every response. - FR-5: WHEN an operator changes the limit of a plan, THE SYSTEM SHALL apply it within 60 seconds without a deploy.
- FR-6: IF the counter store is unreachable, THEN THE SYSTEM SHALL let requests through, log the failure and raise an alert within 1 minute.
3. Out of scope
- Billing for usage above the limit.
- DDoS protection at the network edge.