implementing-api-patterns

Choose REST for public third-party APIs, GraphQL when frontends need flexible queries or mobile bandwidth is constrained, gRPC for high-performance service-to-service communication, and tRPC for TypeScript full-stack projects needing end-to-end type safety.

Cursor-based pagination scales to billions of records without skipped or duplicate rows and handles real-time data changes. Offset pagination is simpler and supports direct page access but should only be used for small, static datasets under 10k records.

Axum (Rust) leads at roughly 140k requests per second with sub-millisecond latency, followed by Gin (Go) at 100k+, Hono (TypeScript) at 50k with under 5ms cold starts for edge deployment, and FastAPI (Python) at 40k with the best developer experience.

Yes, FastAPI generates OpenAPI specs automatically from Python type hints with zero configuration. Interactive Swagger UI is available at /docs, ReDoc at /redoc, and the raw spec at /openapi.json.

Use the DataLoader pattern to batch and cache resolver requests into single database queries. DataLoader groups related fetches by key, so querying a list of users with their posts requires only two queries instead of one per user.

Avoid gRPC for browser clients due to limited native support, simple CRUD APIs where REST is simpler, and public APIs where human-readable debugging matters. Use Connect-Go when browser compatibility with gRPC semantics is required.