What common mistake do developers make with Retries, why does the mistake cause problems, and what is the safer approach?

What common mistake do developers make with Retries, why does the mistake cause problems, and what is the safer approach? One recurring problem with retries is applying it without understanding the boundary it protects. Assuming a queued job runs exactly once; workers can retry jobs, so external side effects should be made idempotent where possible. In addition, Transient failures can be retried according to job retry configuration and worker settings. Safer engineering approach: keep the responsibility in the correct layer, use the framework's supported API, validate and authorize inputs, avoid hidden side effects, and verify the behavior with tests.