Strong relations with referential integrity #2331

Essentially, we see two levels of relations:

• Weak relations, where the developer primarily wants to navigate related models (think of GraphQL) and referential integrity is either not important or already enforced by the (SQL) database.
• Strong relations, where the referential integrity is guaranteed by the framework together with the database.

IMO, referential integrity must be guaranteed at repository level (or lower). For example, when a test helper invokes repository APIs to seed the database with sample data, then we want the referential integrity to be preserved. If we implemented referential integrity at controller level, then consumers of Repository API would be bypassing these checks and would be able to created corrupted data.

Implementation wise, it's important to implement referential integrity in way that does not create any race conditions. For example, a naive implementation calling findById on the source model before creating a new instance of the target model creates a race condition - if another request deletes the source model instance after findById received database response but before create command is set, then we end up with corrupted data.

With SQL databases, referential integrity is typically implemented by defining FOREIGN KEY constraints. In the context of LB4, this means that lb4 migrate must create such constraints based on the relations configured by the app.

With NoSQL databases, users should use different relation types than hasMany, for example referencesMany or embedsMany.