This repository contains an example implementation of the clean architecture in pure Java. You can either use this example as a starting point for discussions about the clean architecture and its various implementation details, or you can use it to learn more about the clean architecture.
The Clean Architecture, Robert C. Martin, 13 August 2012
- Screaming Architecture
- Package By Component
- Modules And Components
- Domain Events And The Event Bus
- As Pure As Possible
- Implementation Notes
- More Sources
- The Example Domain
- Any Questions?
Uncle Bob told us that good architecture is focused on the use cases and not on details like frameworks, tools and so on. Following the "screaming architecture" rule I tried to focus on domain aspects. Therefore, you will find domain-related terms at the top level of this project.
All modules in this project are structured using the package by component strategy as described by Simon Brown in the Clean Architecture book - chapter 34, the missing chapter.
The package by component strategy makes use of Java's default access modifier that only allows access from classes within the same package. It would also be possible to use the Java module system or OSGI, but I wanted to use the simplest solution that works.
If you would like to learn more about the other packaging strategies take a look at chapter 34, the missing chapter, in the Clean Architecture book.
Event though the terms module and component are widely used, there is often a lot of confusion about the meaning of them. If you ask 5 developers: "What are the differences?", you might get 5 different answers, so here is mine:
A component is a group of elements. A component can contain other components.
Component is derived from the latin word componere, which means "to put together".
A module is a replaceable component.
The term module is derived from the latin word modulus, which means "measure". Today the meaning in general is "interchangeable part".
The example application is organized in 2 domain modules that you find at the top level of this project, like a screaming architecture should be organized.
A component in the clean architecture spans all layers, so it's hard to explain what a component is if you only use a 2D diagram. So I tried to visualize my thoughts in 3D.
Components in the clean architecture should be completely separated, but sometimes they share common entities. The communication between components can be done using domain events. A domain event is usually the outcome of the execution of a use case and is published somehow.
I would call the publishing mechanism a domain-event bus. The domain-event bus is a component that is responsible for the delivery of domain events. Domain events are published by use cases and can are consumed by controllers or other components within the interface adapters layer, e.g. a gateway that sends the event to an external system, like a message queue.
This example does not use a lot of frameworks and libraries. Especially not for the application bootstrap, the use cases or entities and the object-relational mapping. Thus, you will not find any Spring or Hibernate and even no apache commons library.
Although I like frameworks and libraries as much as you probably do, I want to show how a complete application works that is implemented in pure Java. You will probably be surprised that many frameworks and libraries can be omitted without significantly increasing the implementation effort. We should consider this before using a framework or library, especially if we made the experience that framework updates sometimes make a lot of effort. Also license issues do not arise if we do not use them and a fix is easy to make. I don't want to say that frameworks and libraries are bad, but they come at a cost that we often forget. So if we use them we should know why we use them, and we should separate them from the core of our application to protect it. Most libraries I use in this example application are written by myself.
When you browse the source code you should keep in mind that you see one solution for every particular technical problem that arises when implementing the clean architecture. For a lot of situations there are more than one way to solve them. Anyway, all solutions must honor the dependency rules of the architecture. Feel free to play with the architecture and refactor any aspect you would like to change to see what is easy to change and what is not.
In the context of the clean architecture you will often see a diagram that uses the terms:
- use case interactor
- input and output boundary
- request/response models.
These terms are a mix between the boundary-controller-entity (BCE) by Ivar Jacobson, the hexagonal architecture by Alistair Cockburn and the clean architecture by Robert C. Martin.
In the BCE architecture you have a boundary, a controller and an entity, as the name implies. In the hexagonal architecture you talk about input and output ports and the clean architecture uses the terms interactors, request and response models, and entities.
In this example I use the term use case for the API instead of input boundary and interactor for the use case implementation. Request and Response models are the input and output models of the use case.
In the UI presenters are used by a controller to present response models to the user and to transform user input to request models. The controller is invoked by the UI as a result of an event. It then uses the presenter to create a request model and invokes the use case.
In this point my architecture is different to the diagram I showed above, because it does not decouple a presenter from a controller, and thus it doesn't need to pass an output boundary to a use case interactor. Therefore, my architecture has a lot less complexity in this area, but at the cost that controllers are not separated from presenters. It's true that I don't have the benefits of decoupling them, but I also don't have a lot of implementation issues that arise from the decoupling. It's a tradeoff that I made in the UI/use case integration to reduce the complexity in that area. I made that decision, because the focus of the example application is the use case part, how these classes are structured and how different modules integrate.
Since this example should be used as a starting point for discussions, feel free to refactor the UI/use case integration part and start a discussion.
If you would like to learn more about clean code and the clean architecture, you should take a look at my youtube channel - clean-code.guru (german).
I also would like to recommend the cleancoders.com website, where you can find a lot of videos about clean code. Most of them provided by Robert "Uncle Bob" Martin and Micah Martin.
For this clean architecture example I chose a car rental application which is complex enough to address common implementation issues, but simple enough to stay focused on the architecture and not the domain.
The application is separated in 2 domain module in order to show how communication via domain events can be done. Each module implements several use cases that are the modules components.
-
car-booking
The car-booking module contains the use cases a customer executes, like find car offers and book a car.
-
car-management
The car-management module contains the use cases a salesperson executes, like handover a car to a customer and return one.
You might want to explore the clean architecture further by implementing more use cases.
Here is a list of user story proposals:
- As a sales manager I would like to define multiple rental rates with a different, but never overlapping, validity period, so that rates can vary over time.
- As a sales manager I would like to specify different stations where the customer can pick up cars.
- As a customer I would like to get offers for stations in the city where I live (my address).
The domain modules are completely separated, which means that each one uses its own dedicated data storage. At the moment they are separated by two different database schemas, but they could also use their own database. The management module reacts to domain events that are fired by the car-booking module in order to update its database. I chose that way to demonstrate how domain events can be implemented.
In the current implementation the domain events are synchronously published on the same thread. Since I use local transactions that are bound to the thread, the event subscribers logic is executed in the same transaction as the publisher.
Feel free to refactor the domain event bus to use asynchronous event publishing or even an external event queue.
The example application comes with a very basic swing ui that you can use to try out the application. Since it is a basic ui it neither has any convenience features, like validation, nor it is very nice. The purpose of the user interface is to provide a simple GUI to access the use cases. It also shows the basics of how use cases are connected to controllers and how controllers use presenters in order to update models. Feel free to implement a web server and provide rest controllers in order to support a web ui.
The application uses a H2 file database that is automatically created in the working directory when you start the application. This will contain some test data. If you need to reset the database, just delete the carrental.mv.db in the working directory.
The example data is the same as the data that the unit tests use. So you can also browse it here:
If you have any questions about
- this example application you can start a discussion.
- the clean architecture and clean code in general you can browse my answers on stackoverflow or post a new answer. I usually watch new answers to the clean-architecture and hexagonal-architecture tag.
- clean code in general or software architecture, you can either take a look at my youtube channel - clean-code.guru (german) or
contact me for trainings and coaching or book a session on
.