A pattern where one upstream request triggers many downstream calls, the classic source of cascading load surprises.
Fan-out is when one incoming request causes the service to make multiple downstream calls, e.g. a single user-feed request that fetches 50 user profiles, 100 post documents, and 20 ad slots. Fan-out is fine when the downstream calls are fast and independent, but it amplifies load: 1K rps with a fan-out of 50 puts 50K rps on the downstream. The pattern's twin is fan-in, where many results are aggregated back to one response.
Most 'why is everything slow' incidents trace back to a fan-out that grew without anyone noticing, the personalization service used to call 5 downstreams, now calls 80, and the slowest of the 80 is the new p99. Bounding fan-out (concurrency limits, timeouts per call, partial-result tolerance) is the discipline that keeps fan-out shapes from becoming load multipliers.
See the part of the platform that handles fan-out in production.