System Design
System Design/senior/freq 4/5

API Gateway: When and What For

An API gateway centralizes cross-cutting concerns (auth, rate limit, routing, transformation). It's a force multiplier when used narrowly and a god-object when used as an integration platform.

gatewayapiarchitecture

Deep dive

Good fits

  • Token validation + authn at the edge.
  • Per-route rate limiting and quota.
  • TLS termination + WAF integration.
  • BFF (backend-for-frontend) routing.

Bad fits

  • Business logic, data transformation, orchestration → use proper services.
  • "Single source of API truth" → version-skew nightmares.

Build vs buy

  • Cloud-native (APIM, AWS API Gateway): great for managed services, slow to iterate, opaque cost.
  • Self-managed (Kong, Tyk, Envoy + custom plugins): more control, more ops.
  • Service-mesh sidecar (Istio/Envoy): consistent L7 inside the cluster, but not a replacement for an external edge.

Real-world example

From production

Team used the API gateway as the integration layer. Every new service required a gateway PR + config deploy. Lead time per change: 2 weeks. Refactored: gateway only does auth/rate-limit/routing; transformations moved into BFFs. Lead time per service change: <1 day.

Interview questions

1 senior-level
Q1What belongs in an API gateway?

Cross-cutting concerns that benefit from centralization: TLS, authn, rate limiting, basic routing. What doesn't: business logic, data composition, orchestration — those go in services or BFFs.

Common mistakes

  • Putting business rules in gateway scripts.

  • Single shared gateway for many teams with no isolation.

Trade-offs

  • Edge gateway adds a hop and a SPoF — make it HA and observable.

Related