In a relational database, foreign keys are normally used to associate records stored in different tables, but wouldn’t it be nice to define relationships dynamically without having to add extra columns or tables? And while we’re at it, how about having sparse relationships by associating a record directly with any other record like “post X was last edited by user #123” or “post X was flagged for review by user #456” (who happens to be a moderator)?
So… a polymorphic many-to-many join table?
Yes, that’s correct. Here’s how an entry in the join table looks like:
{ "id": 6, "sourceComp": "user", "sourceId": 2, "targetComp": "post", "targetId": 3, "type": "author", "createdAt": "2024-03-28T13:28:59.175Z", "updatedAt": "2024-03-28T13:28:59.175Z" }Fine for prototyping, but adds a scaling tech debt “time bomb” for a live system. Those associations had better be really sparse.
There’s certainly the danger of creating too many ad-hoc or sparse relationships, which can cause issues. That said, when used for supplementing foreign keys, Tie-in can be a useful tool in a production system as well.
Don’t you want a graph database at this point?
That idea crossed my mind too, but you can’t really use the full capabilities of SQL in graph databases, and that’s a deal breaker for me.
i was thinking the same thing
NoSQL has been a thing since before there was SQL.
AFAIK, no NoSQL database fully supports SQL, and only some offer support for transactions and joins. The idea here is to augment a relational database by adding capabilities for dynamic relationships.