Dominance

Example Ontology

// User.js
module.exports = {
  connection: 'ourMySQL',
  attributes: {
    email: 'string',
    wishlist: {
      collection: 'product',
      via: 'wishlistedBy'
    }
  }
};

// Product.js
module.exports = {
  connection: 'ourRedis',
  attributes: {
    name: 'string',
    wishlistedBy: {
      collection: 'user',
      via: 'wishlist'
    }
  }
};

The Problem

It's easy to see what's going on in this cross-adapter relationship. There's a many-to-many ( N->... ) relationship between users and products. In fact, you can imagine a few other relationships (e.g. purchases) which might exist, but since those are probably better-represented using a middleman model, I went for something simple in this example.

Anyways, that's all great... but where does the relationship resource live? "ProductUser", if you'll pardon with the SQL-oriented nomenclature. We know it'll end up on one side or the other, but what if we want to control which database it ends up in?

IMPORTANT NOTE

This is only a problem because both sides of the association have a via modifier specified!! In the absence of via, a collection attribute always behaves as dominant: true. See the FAQ below for more information.

The Solution

Eventually, it may even be possible to specify a 3rd connection/adapter to use for the join table. For now, we'll focus on choosing one side or the other.

We address this through the concept of "dominance." In any cross-adapter model relationship, one side is assumed to be dominant. It may be helpful to think about the analogy of a child with multinational parents who must choose one country or the other for her citizenship

Here's the ontology again, but this time we'll indicate the MySQL database as the "dominant". This means that the "ProductUser" relationship "table" will be stored as a MySQL table.

// User.js
module.exports = {
  connection: 'ourMySQL',
  attributes: {
    email: 'string',
    wishlist: {
      collection: 'product',
      via: 'wishlistedBy',
      dominant: true
    }
  }
};

// Product.js
module.exports = {
  connection: 'ourRedis',
  attributes: {
    name: 'string',
    wishlistedBy: {
      collection: 'user',
      via: 'wishlist'
    }
  }
};

Choosing a "dominant"

Several factors may influence your decision:

  • If one side is a SQL database, placing the relationship table on that side will allow your queries to be more efficient, since the relationship table can be joined before the other side is communicated with. This reduces the number of total queries required from 3 to 2.
  • If one connection is much faster than the other, all other things being equal, it probably makes sense to put the connection on that side.
  • If you know that it is much easier to migrate one of the connections, you may choose to set that side as dominant. Similarly, regulations or compliance issues may affect your decision as well. If the relationship contains sensitive patient information (for instance, a relationship between Patient and Medicine) you want to be sure that all relevant data is saved in one particular database over the other (in this case, Patient is likely to be dominant).
  • Along the same lines, if one of your connections is read-only (perhaps Medicine in the previous example is connected to a read-only vendor database), you won't be able to write to it, so you'll want to make sure your relationship data can be persisted safely on the other side.

FAQ

What if one of the collections doesn't have via?

If a collection association does not have a via property, it is automatically dominant: true.

What if both collections don't have via?

If both collections don't have via, then they are not related. Both are dominant, because they are separate relationship tables!!

What about model associations?

In all other types of associations, the dominant property is prohibited. Setting one side to dominant is only necessary for associations between two models which have an attribute like: { via: '...', collection: '...' } on both sides.

Can a model be dominant for one attribute and not another?

Keep in mind that a model is "dominant" only in the context of a particular relationship. A model may be dominant in one or more relationships (attributes) while simultaneously NOT being dominant in other relationships (attributes). e.g. if a User has a collection of toys called favoriteToys via favoriteToyOf on the Toy model, and favoriteToys on User is dominant: true, Toy can still be dominant in other ways. So Toy might also be associated to User by way of its attribute, designedBy, for which it is dominant: true.

Can both models be dominant?

No. If both models in a cross-adapter/cross-connection, many-to-many association set dominant: true, an error is thrown before lift.

Can neither model be dominant?

Sort of... If neither model in a cross-adapter/cross-connection, many-to-many association sets dominant: true, a warning is displayed before lift, and a guess will be made automatically based on the characteristics of the relationship. For now, that just means an arbitrary decision based on alphabetical order :)

What about non-cross-adapter associations?

The dominant property is silently ignored in non-cross-adapter/cross-connection associations. We're assuming you might be planning on breaking up the schema across multiple connections eventually, and there's no reason to prevent you from being proactive. Plus, this reserves additional future utility for the "dominant" option down the road.