Avoid duplicate task execution in Celery when using SQS as the broker by tuning visibility_timeout against task runtime

Steps

1. Set broker_transport_options = {'visibility_timeout': <seconds>} in the Celery config — this only takes effect when Celery creates the SQS queue itself, not when using a pre-existing queue
2. Know the documented default is 1800 seconds (30 minutes) and the AWS/SQS-imposed ceiling is 43200 seconds (12 hours)
3. Set visibility_timeout to at least the longest expected task runtime, including any ETA/countdown delay, since a task that runs past the visibility timeout becomes eligible for SQS redelivery to another worker
4. Keep task_time_limit (or task_soft_time_limit) comfortably under visibility_timeout so a legitimately slow task doesn't get redelivered and double-executed before it finishes
5. Recognize that Celery redelivers messages on worker shutdown regardless of visibility_timeout — a long timeout only delays redelivery after a crash, it doesn't prevent duplicate execution in that scenario
6. Make task logic idempotent as a backstop, since visibility_timeout tuning reduces but does not eliminate the chance of duplicate delivery