A checkout API calls inventory, payment, and shipping synchronously. One slow downstream service exhausts its request workers. Which design best limits the cascading failure?

1. Retry every failed call indefinitely in the same request worker.
2. Set bounded timeouts, isolate concurrency, and apply a circuit breaker where appropriate.
3. Increase log retention as the only architectural change.
4. Remove all downstream timeouts to avoid premature failures.