How do you prevent duplicate generation requests at the API boundary?

How do you prevent duplicate generation requests at the API boundary?

Duplicate requests can come from double clicks, retries, or a lost response. Disabling a button improves the interface but does not solve duplication at the server boundary.

Define request identity

An idempotency key can represent one intended operation within a user or project scope. Associate it with the request’s meaningful input and saved response or job identifier. If the key is reused with different input, reject the conflict instead of silently returning an unrelated job.

Handle concurrent arrivals

Use a database-enforced uniqueness rule or equivalent atomic operation so two simultaneous requests cannot both create the canonical job. Define retention and retry behavior for the key. Test two parallel submissions and a response lost after job creation. The second attempt should resolve to the existing intended operation rather than creating another billable pipeline.