How to Test Retries in a Bulk Video Generation Queue
A batch can fail while submitting a request, polling a task or copying an output. Keep the provider task ID and the durable asset path in your job record so retries can resume the work at the point of failure.
Worked example
Three-row dry-run fixture
row,input_event,expected_next_action ad-001,submit rejected with HTTP 429,delay and retry submission ad-002,submission timed out before response,reconcile before resubmitting ad-003,output download failed after success,retry download for existing task ID
Steps
- Give each input row a stable internal key and store state in a durable queue or database. Claim each row atomically before attempting submission.
- Simulate a documented rate-limit rejection, a timeout before receiving the submission response, and a failed download after success. For an ambiguous submission, reconcile with a provider task or use documented idempotency support before submitting again.
- Apply the provider's backoff guidance to explicit rate-limit rejections. For download failures on known tasks, retry the download while the URL is usable. Keep the provider job ID and mark completed rows with their permanent output locations.
What to check
- Replay an already completed row and verify that it reuses its stored asset.
- Complete the fixture without provider calls. Assert that only the explicit rejection can be submitted again automatically in this example.