Skip to main content

Layer your app, keep Express within its boundaries



Separate component code into 3 layers​

The root of every component should hold 3 folders that represent common concerns and stages of every transaction:

my-system
├─ apps (components)
│  ├─ component-a
   │  ├─ entry-points
   │  │  ├─ api # controller comes here
   │  │  ├─ message-queue # message consumer comes here
   │  ├─ domain # features and flows: DTO, services, logic
   │  ├─ data-access # DB calls w/o ORM

- Entry-points - This is where requests and flows start, whether it's REST API, Graph, message queue, scheduled jobs or any other door to the application. This layer's responsibility is quite minimal - adapt the payload (e.g., JSON) to the app format, including first validation, call the logic/domain layer and return a response. This is typically achieved with a few lines of code. Many use the term "controller" for this type of code also technically, its just an adapter

- Domain - This is where the app flows, logic and data live. This layer accepts protocol-agnostic payload, plain JavaScript object and returns one as well. Technically it contains common code objects like services, dto/entities, and clients that call external services. It also typically calls the data-access layer to retrieve or persist information

- Data-access - This is where the app holds code that interacts with DB. Ideally, it should externalize an interface that returns/gets plain JavaScript object that is DB agnostic (also known as the repository-pattern). This layer involves DB helper utilities like query builders, ORMs, DB drivers and other implementation libraries

What is the merit? - When having flexible infrastructure that allows adding more API calls and DB queries promptly, a developer can code a feature faster by focusing on the domain folder. In other words, less time is spent on technical activities and more on activities with added value. This is a ubiquitous trait that is at the heart of most software architectures like DDD, hexagonal, clean-architecture and others. On top of this, when the domain layer is not aware of any edge protocol, it can serve any client and not only HTTP calls

Why not MVC or clean architecture? - The 3-tier pattern strikes a great balance between achieving the separation goal while still keeping the structure simple. It also lacks abstractions - each tier represents real-world physical tier where every request will visit. On the other hand, MVC is a simplistic pattern where the letters VC represent a few lines of a code only and the letter M means anything else. Clean architecture is architecture with high level of abstractions that can achieve even greater separation but the price tag is unproportionally higher due to the increased complexity