description | last_modified |
---|---|
Some ways to avoid service classes getting bigger and bigger |
2020-06-13 13:29:49 +0200 |
- In a lot of applications, you have service classes that act as a facade for lower-level domain logic and also contain coordination and control logic
- These service classes tend to grow over time and can become so "fat" that they are difficult to maintain
Basic idea:
- Identify different areas of functionality within the fat service class
- Potential way to split: identify sub-concepts
- Potential way to split: retrieval of data vs. changing data
- Create a separate service for each area
- If needed, you can create a helper service for sharing common functionality
- Either the old fat service class acts as a facade to these more specific service classes, or clients call the sub-services directly
Basic idea:
- Identify the actions that the fat service class is performing
- This could be as simple as "1 public method = 1 action"
- Create a separate class to represent each action
- It might make sense to have all of these actions implement a common interface, especially if there are some "cross-cutting concerns" that need to be taken care of regardless of the specific type of action
- The old fat service class creates the actions and then delegates to them
- Typically each call to the fat service class will create one or more instances of the action classes based on the specific input received
- The old fat service class takes care of cross-cutting concerns if needed
This can be seen as a form of the Command pattern
Benefits:
- Class per action means that we get some very focused classes
- Class per action means we can easily compose higher-level actions out of lower-level actions
- This approach makes it relatively easy to provide undo functionality or show a history of actions (if needed)
Example: virtual file system
- Situation:
- Fat service class with functionality for file creation, file update, file deletion, replacing a folder's contents with contents from an archive, ...
- For every call to the service class, all changes must happen within a single DB transaction
- All changes generate events that other components can listen to, plus they generate updates to an in-memory representation of the current state of the file system. However, those events and updates are only valid once the entire transaction is committed.
- Solution:
- Each file action (create, update, delete, ...) is implemented as its own class
- All of these file action classes share the same interface, which specifies that they return events and in-memory cache updates
- Bigger actions (archive import etc.) delegate work to smaller actions that they create
- For every call to the service class, it creates the necessary actions based on the received input and also passes a transactional DB connection on construction
- Service class collects event and cache updates
- When all actions for a service class call have completed, the service class commits the DB transaction, sends events and applies updates to the in-memory file system representation
- In this particular case, file actions depended on several lower-level services. Solution: file action factory service with methods for creating each type of action. Each method takes action input + DB connection and calls the action's constructor with action input, DB connection and instances of lower level services.
export interface FileAction {
public executeAndGetResult(): Promise<FileActionResult>;
}
export interface FileActionResult {
events: FileEvent[];
cacheUpdates: FileCacheUpdate[];
}