Why API gateway?

One client, one checkout call, three backend services. Watch what changes when a gateway sits in front of them.

Watch it explained

~15s animated walkthrough: duplicated auth/rate-limit/logging logic across services vs. one gateway.

Loading player…

Direct fan-out

Each checkout call fans out to 3 services, and Orders/Inventory each re-check the token themselves.

sent

success

—

overloaded drops

gateway 429s

—

Authhealthy

queue 0.0 / drains 8/s

Ordershealthy

queue 0.0 / drains 8/s

Inventoryhealthy

queue 0.0 / drains 8/s

Auth checks performed

≈ — checks per request (client + 2 services re-check)

Last decisions

Run traffic to see decisions.

Why this matters

Cross-cutting concerns move to one place. Direct mode re-validates the token 3 times per request across 3 services; the gateway validates it once and forwards trust downstream.
Backpressure protects the fleet. Trigger a burst in Direct mode and watch services turn amber/red with no protection. Switch to Gateway and the same burst gets absorbed as fast, cheap 429s before it ever reaches a service.
The client gets simpler, not the system. The gateway still has to know how to route, retry, and aggregate — that complexity doesn't vanish, it centralizes.

Traffic console

Run steady traffic or fire a spike and compare modes.

baseline req / secondSteady client traffic in requests per simulated second, before any burst multiplier.service capacity (req/s)How many requests per second each backend service can actually drain. Past 2× this, the service is treated as overloaded and starts dropping calls.gateway capacity (req/s)Only used in Gateway mode. The gateway's own token-bucket rate limit — requests beyond this get a fast 429 instead of reaching services.simulation speedOnly changes animation/simulation playback speed.

Try this

Hit “Send spike” and watch Auth turn red first — it's taking 3 hits per request while Orders and Inventory only take 1. Then switch to Gateway and repeat.