How do you make media-generation retries safe?

How do you make media-generation retries safe?

A safe retry preserves the meaning of one intended operation. Repeating a request should not accidentally produce conflicting final results or duplicate publication.

Identify the intended operation

Assign a stable job identity and separate attempt identifiers. Tie each attempt to an immutable input version. Write intermediate output to attempt-specific locations so concurrent attempts cannot overwrite one another. Decide which successful attempt can become the canonical result.

Commit completion carefully

Validate output before atomically or conditionally updating the job’s result reference. Ignore late completion from an obsolete attempt. Keep external side effects separate from pure rendering and reconcile uncertain outcomes before repeating them. Test duplicate delivery by executing the same intended job twice and checking that the project exposes one consistent final result. Also verify that abandoned attempt artifacts remain identifiable for later cleanup.