A healthy application is divided into macro and sub components that represent the contexts at play, whether they are related to business domain or infrastructure code.
As GitLab code has so many features and components it’s hard to see what contexts are involved. We should expect any class to be defined inside a module/namespace that represents the contexts where it operates.
When we namespace classes inside their domain:
- Similar terminology becomes unambiguous as the domain clarifies the meaning:
- Top-level namespaces could be associated to one or more groups identified as domain experts.
- We can better identify the interactions and coupling between components.
For example, several classes inside
MergeRequests::domain interact more with
Ci::domain and less with
# bad class MyClass end # good module MyDomain class MyClass end end
A good guideline for naming a top-level namespace (bounded context) is to use the related
feature category. For example,
Continuous Integration feature category maps to
Alternatively a new class could be added to
Groups:: if it’s either:
- Strictly related to one of these domains. For example
- A new component that does not have yet a more specific domain. In this case, when a more explicit domain does emerge we would need to move the class to a more specific namespace.
Do not use the stage or group name since a feature category could be reassigned to a different group in the future.
# bad module Create class Commit end end # good module Repositories class Commit end end
On the other hand, a feature category may sometimes be too granular. Features tend to be treated differently according to Product and Marketing, while they may share a lot of domain models and behavior under the hood. In this case, having too many bounded contexts could make them shallow and more coupled with other contexts.
Bounded contexts (or top-level namespaces) can be seen as macro-components in the overall app.
Good bounded contexts should be deep
so consider having nested namespaces to further break down complex parts of the domain.
For example, instead of having separate and granular bounded contexts like:
ContainerNetworkSecurity::, we could have:
module ContainerSecurity module HostSecurity end module NetworkSecurity end module Scanning end end
If classes that are defined into a namespace have a lot in common with classes in other namespaces, chances are that these two namespaces are part of the same bounded context.