I think the article handles both cases (timeouts and the first error) by cancelling the other running async operations as quickly as it can.