Avoiding Fallback in Distributed Systems

- This is a technical article from the AWS Builder's Library (a collection of in-depth engineering articles written by Amazon engineers about how they build and operate large-scale systems). - Distributed systems: computer systems where components run on multiple networked machines, communicating by passing messages. They are harder to design than single-machine systems because partial failures (some components break while others keep working) are always possible. - Fallback: a defensive pattern where, when a primary operation fails, a system substitutes a simpler or cheaper operation to avoid a complete outage. The article warns that fallback can do more harm than good if not designed carefully, because it can mask the symptoms of a problem while making the underlying issue worse (e.g., hiding a capacity shortfall until the system collapses). - The piece is aimed at experienced backend engineers; it assumes familiarity with concepts like load shedding, circuit breakers, retries, and idempotency (designing operations so they can be repeated safely without unintended side effects).