How do you manage request-level timeouts in a synchronous, batch-based inference architecture?

Interview question asked to Data Scientists interviewing at Audible, Benchling, Ally and other companies. Original question asked: How do you manage request-level timeouts in a synchronous, batch-based inference architecture?.