The Whole Chapter, in One Component
There is no gateway-specific theory to learn. An API gateway is where every mechanism in this chapter ends up living once there is more than one service.
What is new is not any single feature. It is that all of them are in one place, owned by one team, on one deployment schedule, on the critical path of everything.
The stages are ordinary and individually cheap. What makes them consequential is that there is no way to opt one service out.
That simplification is the product. It is also the dependency that makes the service unable to run without the gateway.
Notice what has happened to the service's security model. It now trusts a header it did not verify, because the gateway verified it, and that is only true while the gateway is the only way in.
What the Concentration Costs, Measured
Five real features, including verifying an HMAC and validating a body against a schema, added nineteen microseconds in total.
Going through the gateway at all costs 0.160 milliseconds and does nothing. Adding five real features on top of that costs 0.019.
Which reframes the usual argument. People debate whether to enable a gateway feature because of its cost, and the features are nearly free. The decision that costs something is whether there is a gateway in the path at all, and that one is usually made once and never revisited.
The larger cost is not .
