清洁架构中的数据库登录？

Question

I'm pretty new to Clean Architecture and like the principals. If we define data access interfaces and the implementation comes later that's all good.

In my case my application will have data from several sources. Now if I have a database instead of using the file system for the implementation, I will need a database login and therefore prompt a user. How does this fit in with Clean Architecture since the domain and application layers shouldn't know about implementation?Only implementing with a database brings a requirement for a login.

Cheers, Alan

Answer 1

A good question, but sadly there is no simple answer to most good questions.

A naive approach would be to ask the user for the credentials in the UI and then pass them as a user case request object to the use case. The use case can then pass them to the repository.

But like any simple solutions for complex problems it has pitfalls.

The simple solution means that the credentials must be passed to the use case (and forwarded to the repository) on every use case invokation. Since you don't want the user to enter the password on every use case, you must store the password in the application state in a secure manner. This strongly depends on the kind of application you develop. Is it a desktop app, a mobile, a web or even cloud application.

If you have a desktop app you might want to use repository instances that are session scoped. Thus you must introduce some kind of session in your application state. Maybe it is a simple map, maybe more. You can then either create use cases by passing them the repository from the session (if already logged in) or you can use a singleton use case instance that refers to a repository proxy (also a singleton) that delegates to the actual (logged in) repository.

If you have a web application you might have multiple server nodes. Thus there is no single session. Every web request might be processed on a different node. You can use sticky sessions to prevent this, but on the cost of scalability. Otherwise you must find a way to also distribute the credentials in a secure manner. Maybe you can use some kind of token.

Sadly I can't give you a oneline answer since a good answer depends on much more and there dozens of solutions based on other requirements. I hope I could give you a feeling of the complexity and it helps you to narrow your question.

Answer 2

Accepting login data from user does not mean you expose domain to implementation details, it only means you get that input from presentation layer (and of course should save it in a secured way for later use) rather than save it ahead somewhere in your infrastructure layer.

In a way, ui client and infrastructure are similar, they are both considered as implementation details. That is why they are both presented in the same outer circle of the CA diagram.