Bounded Contexts Working Group

Attributes

Summary

Today code is namespaced using Ruby namespaces but we don’t have explicit rules or guidelines on how to organize code. New namespaces are constantly being created and often there is no need to since new concepts could be nested inside existing namespaces to better represent bounded contexts. Having consistently namespaced code is the pre-requisite for a modularized codebase.

With this working group we want to:

• Identify the bounded contexts that compose the GitLab Rails codebase
• Stop the proliferation of new top-level namespaces without explicit reason.
• Have modules with rich domain logic that are aligned to the engineering organization.

Context

The proposal to split GitLab Rails monolith into a modular monolith describes in details why modularization is important for the current stage of the GitLab codebase. In order to make progress towards a modularized monolith we first need to understand what are the modules and bounded-context that are present today in GitLab Rails monolith.

Activity

Members of the working group will:

• Classify the files related to their bounded context.
• Participate to sync meetings, bring questions and communicate progress.
• Provide feedback on the process and the guidelines to identify bounded contexts.
• Act as modularization evangelist for their respective stage/domain.
• Help review the final list of bounded contexts to ensure it’s comprehensive.

Communication

• The working group would meet weekly initially alternating between a APAC and AMER friendly timezone. Then we can make it bi-weekly.
• A #wg_bounded_contexts slack channel would be created for anyone to ask questions and communicate async.
• After the initial meeting we expect the work to be carried out asynchronously. Each team member can work off their own list of files to categorize.
• Working on a spreadsheet (or alternatives) will make it easy to track progress on categorized files and remaining work to do.
• DRI to facilitate the meetings and communicate weekly status updates will be @fabiopitino and anyone else who wants to contribute.
• Eventual issues and follow-up ideas would be created using ~modular_monolith label.

Exit Criteria

The working group can be disbanded when:

Details

The working group would map the Ruby codebase, in particular the domain code into app/ and lib/ folder and come up with a list of bounded contexts. This may exclude app/controllers and app/views initially since we want to focus primarily on the domain layer (the core) of the Hexagonal Architecture.

During the process of mapping the codebase we would need to do:

• List all the Ruby files in a spreadsheet and categorize them into components following the defined guidelines. Files under the same directory can generally be categorized in bulk, so this should not take a lot of time.
• Inside the lib directory, distinguish between generic code that should be extracted as gem and domain code that should be namespaced the same as code in app. All code related to the same domain should have the exact same namespace. For example, we should not have Ci:: (from app) and Gitlab::Ci:: (from lib).
  • We will have a special platform category for code that should be extracted as gems and cross-cutting concerns in app and lib (generic base classes, utility, loggers, framework code, etc.).
• It’s important to focus only on the top-level bounded contexts without getting carried away with renaming nested classes. This can be a refactoring activity at the discretion of the team owning this feature, but not in scope of the working group.
• Identify similar namespaces and the merging strategy. It could be any of the below:
  • Create a new namespace that represents a larger domain and move the classes in there.
  • Identify a representative namespace for the bounded context and move all other classes in there.
• Distinguish ambiguous namespaces by putting them into separate bounded context. When 2 namespaces seem similar but they operate in completely different contexts, move them under the respective bounded context.
• Identify namespaces that have multiple responsibilities and should be split. In this case classes could go into other namespaces that better represent their domain.
• Try to reuse existing top-level namespaces whenever possible.
• Ignore everything else (non Ruby code, config, specs, controllers, views, rake tasks, Grape API, GraphQL).

The list of bounded-context doesn’t have to be perfect. It should be a good enough representation of the domains at play in the GitLab Rails codebase where each class has only 1 namespaces where it can exist and there is no ambiguity. We will iterate on this list and will need to have a process in place for creating, deleting, renaming bounded contexts.

Roles and Responsibilities

The Working Group should be composed of representative of all the various GitLab stages. Each representative would be in a functional lead role representing their domain as a backend expert for their given GitLab DevOps stage.

Participants would ideally be interested in driving results related to modularization and Domain-Driven Design. We will communicate the creation of the working group in #development, #backend and #engineering-fyi slack channels. For stages still missing participants we will seek support from Engineering leaders who work in the given DevOps stage.

View page source - Edit this page - please contribute.