GraphQL: компромиссы

GraphQL wins при diverse clients и graph-shaped data; loses на simple CRUD, file transfer и HTTP caching ubiquity.

Разбор

• GraphQL wins когда clients diverse, data graph-shaped — kills over-fetching и waterfalls, typed schema + introspection для tooling/codegen.
• When NOT: simple CRUD, file up/download, public API где HTTP caching matters — REST проще и cache-friendly. GraphQL для two-endpoint app — over-engineering.
• You inherit a server. Schema design, resolver performance (DataLoader), query-cost limits, client caching — ongoing work. Complexity moves client → system.
• Caching downgrade vs REST HTTP layer — rebuild client normalized cache, optionally edge persisted queries.
• Monitoring/rate-limiting harder — every call same POST; rate-limit by query complexity instead of per-endpoint.

Итог

GraphQL — когда many clients нужны different field subsets одного graph; REST default для CRUD и cacheable public APIs.