Style Example: An Order and Its Items in REST, GraphQL and gRPC

Putting one screen through REST, GraphQL and gRPC shows how each handles over-fetching, receiving fields the client does not use, and under-fetching, needing extra requests to get everything. The phone screen shows an order number, and for each of its three items the product name and quantity. The counts and sizes are illustrative.

REST under-fetches and over-fetches at once. Seven requests, each paying the network's delay on a phone, and most of the bytes are fields nobody reads. An include parameter, or a dedicated endpoint for the screen, fixes the calls but not the size.

GraphQL fits the screen exactly. One request, and only the fields named. That is what it was designed for: many screens, each wanting a different slice of the same data.

gRPC depends on the procedure. With a procedure designed for this screen, one call returns a compact binary message. With fine-grained procedures, GetOrder then GetItem three times, it under-fetches like REST, though each call is cheaper.

What it does not show. The REST responses can be cached by every HTTP cache on the way; the GraphQL query cannot without extra work. The smallest response is not always the fastest system.