3.4 API Gateways
One entry point for many services, and what it means for the client.
Modern applications rarely talk to a single backend. An API gateway gives the frontend one address and one set of rules, hiding however many services sit behind it.
Diagram source
flowchart LR
A[Frontend]
B[API Gateway]
C[IAM Service]
D[Payment Service]
E[Merchant Service]
F[Notification Service]
A --> B
B --> C
B --> D
B --> E
B --> FWithout one, the frontend needs to know every service's address, implement auth against each, and handle CORS for all of them. Services become directly coupled to clients and cannot be split or moved without breaking them.
What belongs in the gateway
Cross-cutting concerns that would otherwise be reimplemented in every service:
Routing — path or host decides the destination. /api/payments/* goes to
the payment service.
Authentication — validate the token or session once, at the edge. Services receive an already-authenticated request with the user identity attached, and never see raw credentials.
Rate limiting — per user, per IP, per endpoint, applied before the request costs anything downstream.
Request IDs — attach a correlation ID that follows the request through every service. This is what makes distributed tracing possible.
Logging and metrics — one consistent place where every request is recorded.
Versioning — route /v1 and /v2 to different implementations during a
migration.
Authorization is the interesting case: the gateway can check that you are authenticated, but whether you may modify this particular invoice is a business rule that belongs in the service that owns invoices.
Gateway vs load balancer vs reverse proxy
These overlap enough to be genuinely confusing, and often the same software does all three.
Reverse proxy forwards requests to backends; the base mechanism
Load balancer distributes across identical instances; about capacity
API gateway routes to different services and applies API policyThe distinction is intent, not technology. Nginx can do all three. An API gateway is the layer that knows about APIs — tokens, rate limits, versions — rather than just about hosts and ports.
What this changes for the frontend
Three practical consequences.
One origin. If the gateway sits on the same domain as your app, CORS stops being your problem entirely.
Consistent errors. The gateway can normalize failures so the client is not
writing per-service error handling. It also means a 401 might come from the
gateway rather than the service — worth knowing when debugging.
New failure modes. A rate limit returns 429 with a Retry-After header
that your client should respect rather than immediately retrying. And the
gateway itself is a single point of failure, which is why it is usually run
behind a load balancer with several
instances.
The BFF pattern
Backend for Frontend: a gateway-like service owned by the frontend team, shaped to what the UI actually needs.
A screen that requires data from four services means four round trips from the browser, each paying full network latency. A BFF makes those calls server-side on a fast internal network and returns one response shaped for that screen.
Browser → BFF → [ four services, in parallel, internally ]Useful when mobile and web need different shapes of the same data, or when the number of round trips is hurting perceived performance. It is also where GraphQL usually sits.
The cost is another service to own, deploy and monitor — and a tendency for business logic to migrate into it, which is the thing to watch for.
If you are on Next.js or a similar framework, route handlers and server components already give you a BFF. Server-side data fetching that aggregates several APIs and returns exactly what the page needs is this pattern, without a separate deployment.