Rationale

http4k was created as the distillation of 15 years worth of experience of using various server-side libraries and we've stolen good ideas from everywhere we can. For instance - the routing module is inspired by UtterlyIdle, the "Server as a function" and filter model is stolen from Finagle, and the contract module OpenApi generator is ported from Fintrospect. With the growing adoption of Kotlin, we wanted something that would fully leverage the features of the language and it felt like a good time to start something from scratch.

For our purposes, we wanted something that:

  1. Was based on simple functional concepts and embraced immutability.
  2. Embraced the "Server as a Function" model a uniform server/client API.
  3. Absolutely no magic involved: No reflection. No annotations.
  4. Lightweight with minimal dependencies (apart from the Kotlin StdLib, http4k-core has zero).
  5. Embraced Test-Driven approaches, was testable outside of an HTTP container, and testing should require no custom infrastructure.
  6. Starts/stops ultra quickly.
  7. Provides typesafe HTTP message deconstruction/construction.
  8. Automatically dealt with contract breaches to remove boilerplate.
  9. Automatic generation of OpenApi documentation (including JSON Schema models).

http4k ticks all of these boxes.

It allow us to construct entire suites of services which can be tested either wired together without HTTP, or spun up in containers using a single line of code. The symmetric HTTP API also allows Filter chains (often called "Middleware" or "Interceptors" in other frameworks) to be constructed into reusable units/stacks for both server and client sides (eg. logging/metrics/caching...) since they can be composed together for later use.

As a bonus, we can also easily create simple Fake servers for any HTTP contract, which means (in combination with CDC suites) you can end-to-end test micro-services in an outside-in way (using GOOS-style acceptance tests).

Scenarios such as "what happens if this HTTP dependency continually takes > 5 seconds to respond?" are easily modelled - answers you can't easily get if you're faking out your dependencies inside the HTTP boundary.