Interpret Retry-After without creating a retry storm

The useful distinctions

• 429 communicates a request-rate limit.
• Retry-After can express a delay in seconds.
• Retry-After can alternatively use an HTTP date.

Separate delay advice from quota policy

The synthetic response asks for a ninety-second delay. That is useful timing information, but it does not identify whether the limit is per user, per address, per endpoint, or shared across a larger group. Do not reverse-engineer a quota from one response. Record the status and field exactly, then describe a next-attempt plan that respects the advice and remains within the service's documented rules.

Make a later attempt a new observation

Waiting the indicated interval does not guarantee success. Conditions can change, and a later response needs its own status and timing record. For a date-form value, distinguish a server-provided wall-clock time from a local relative timer and account for the observation's clock context. An automatic client should also have bounded attempts and a way to stop when the operation is no longer needed; otherwise a delayed retry can become a permanent loop. The example does not implement a client or measure a service. It gives a reviewable timing contract without asserting that any request was sent or that a particular service enforces a ninety-second policy. A retry worksheet can record observation time plus ninety seconds as an earliest planned attempt, while retaining the original delay value. It should not infer a reset schedule or quota from that single number.

Worked example · synthetic

The numeric value is ninety seconds, not ninety requests. It provides a delay instruction for the synthetic response; it does not describe a quota size or a guaranteed recovery time.

Synthetic text example; no network operation has been performed.

HTTP/MIME line endings are CRLF. Request URL context, where labelled, is commentary rather than a wire field. No message was transmitted.

HTTP/1.1 429 Too Many Requests
Date: Sun, 08 Nov 2026 08:00:00 GMT
Retry-After: 90
Content-Length: 0

Check the interpretation

• Parse numeric seconds separately from HTTP dates.
• Do not derive quota scope from one response.
• Bound retries and inspect each later result.