cascading failures on upstream services. then you get 20 different services failing instead of the single one.