mirror of
https://github.com/Unleash/unleash.git
synced 2024-11-01 19:07:38 +01:00
81 lines
3.6 KiB
Markdown
81 lines
3.6 KiB
Markdown
---
|
|
title: "ADR: Write Model vs Read Models"
|
|
---
|
|
|
|
## Background
|
|
|
|
We keep solving 3 problems in each of the business modules:
|
|
* changing the state of the system through actions
|
|
* displaying UI
|
|
* asking other modules for information
|
|
|
|
In the past we had one pattern of `service + store` for handling all 3 of those. Store was responsible for changing state and returning UI data. Cross module collaboration
|
|
happened by calling services in other modules. While it was convenient and reduced the number of building blocks to learn, it also led to coupling between
|
|
services and overloaded stores - responsible for writing data, reading complex UI data and answering questions from other modules.
|
|
|
|
## Decision
|
|
|
|
To address these challenges categorize your problem as one of 3 models:
|
|
|
|
[3 types of models](/img/write-model-vs-read-models.png)
|
|
|
|
We have 3 types of models serving 3 different purposes:
|
|
* I need to take action that modifies a state of the system (Write Model)
|
|
|
|
Solution: Go through the Command stack aka the *Write Model*. *Application logic* with the use case/workflow/steps resides in the *Application Service*. Application service can delegate
|
|
business logic handling to the *Domain Model* if needed (only for complex subdomains like Change Requests). Simple domain logic (very few business if statements)
|
|
can be handled together with the application logic. Application service calls the store that is responsible for the *generic CRUD operations* serving the write model.
|
|
|
|
* I need to read data for UI display (External Read Model)
|
|
|
|
Solution: Expose *External Read Model* aka *View Model* that returns all the data in one query so that frontend doesn't have to ask multiple write models for data.
|
|
View model will typically join data across a few DB tables since most UI screens required more than one table of data. If we keep the *complex queries* for UI
|
|
in the external read model our stores can be generic and focused on the main table operations.
|
|
|
|
* I need information from another module to proceed with my module application logic (Internal Read Model)
|
|
|
|
Solution: Cross module queries can be handled with the *Internal Read Model*. Internal means it's used internally between our modules and not exposed to the UI.
|
|
Internal read models are typically narrowly focused on answering one question and usually require *simple queries* compared to external read models. By introducing internal
|
|
read model other business modules are not coupled to our module's service/store/write model and can't call write model methods from another module by accident.
|
|
|
|
## Example
|
|
|
|
Before (one multi-purpose class)
|
|
```typescript
|
|
class SegmentStore {
|
|
// used to perform actions on segment
|
|
create(segment: Segment): Promise<Segment> {}
|
|
get(id: number): Promise<Segment> {}
|
|
update(id: number, segment: Segment): Promise<Segment> {}
|
|
delete(id: number): Promise<void> {}
|
|
|
|
// used by UI
|
|
getAll(): Promise<SegmentWithUsageInfo[]> {}
|
|
|
|
// used by another module checking existing names
|
|
getSegmentNames(): Promise<string[]> {}
|
|
}
|
|
```
|
|
|
|
After (3 role-based classes so clients depends only on method they use)
|
|
```typescript
|
|
// used to perform actions on segment
|
|
class SegmentStore {
|
|
create(segment: Segment): Promise<Segment> {}
|
|
get(id: number): Promise<Segment> {}
|
|
update(id: number, segment: Segment): Promise<Segment> {}
|
|
delete(id: number): Promise<void> {}
|
|
}
|
|
|
|
// used by UI
|
|
class SegmentViewModel {
|
|
getAll(): Promise<SegmentWithUsageInfo[]> {}
|
|
}
|
|
|
|
// used by another module checking existing names
|
|
class SegmentReadModel {
|
|
getSegmentNames(): Promise<string[]> {}
|
|
}
|
|
```
|
|
|