a quest for the right level of coupling
Distributed systems (e.g., microservices, web apps, or web services) promise more flexibility to developers because each node is loosely coupled with each other. Indeed, each node evolves on its own and is free to use its own technology stack (e.g., JavaScript, Python, JVM, etc.). However, I argue that this loose coupling also introduces more maintenance burden, which negatively impacts the developers productivity. Typically, when a server is modified to require a new request parameter, the clients and the documentation may not automatically catch up on that change. This results in additional work on the client and the documentation to carry that change forward, or, even worse, in a failure at run-time if developers didn’t notice that they introduced an incompatible change.
To address this problem, I’ve created the endpoints4s library. It introduces just enough coupling between servers, clients and documentation such that they are automatically consistent regarding the communication protocol, while keeping the flexibility of using different technology stacks on servers and clients.
In this talk, I will show a demo of endpoints, explain the design decisions I’ve made, and compare the library with other approaches.
This project contains both the slides and the demo code. To build the demo you only need sbt. To build the slides you need Pandoc, and to show them you need Python 3.
$ sbt
> server/reStart
Use the makeSlides sbt task to build the slides (into the target/slides/ directory). Or the showSlides task
to run an HTTP server showing the slides. Typical developer workflow:
$ sbt
> ~showSlides