Automatically determine the entity relationships that must be JOINed in a TypeORM query to satisfy nested object fields selected by a client in a GraphQL query.

Can be used as a potentially higher performance alternative to the DataLoader pattern.

Installation

npm i typeorm-relations-graphql

This library is written in TypeScript, so type definitions are included in the box.

Your project must also install the following as peer dependencies (you should have them already):

• typeorm v0.3.x
• graphql v14.x or higher

Note: typeorm v0.3.0 changed the way relations and data sources work. If you are still using typeorm v0.2.x, please install typeorm-graphql-joiner@^1 and read the usage instructions for that version.

Introduction

When your GraphQL server is backed by TypeORM entities, you may have object relationships like the following example:

{
  // Product entity
  "product": {
    "id": "1234",
    "name": "Some product",
    // nested Owner entity
    "owner": {
      "id": "4321",
      "name": "Some owner"
    }
  }
}

Let's say product corresponds to a Product entity in TypeORM, and it has a many-to-one relationship to an Owner entity defined on the product.owner property. In your database, you have a table for each of these entities.

Now you want to expose Product as an object in your GraphQL schema with the same relationship. You could simply resolve product.owner using a database query to fetch the related Owner object, but if you create a GraphQL resolver that returns a list of n products then your server will need to perform n + 1 database queries to fully resolve it. This problem multiplies exponentially as your schema grows more complex and you have levels of nested relationships.

typeorm-relations-graphql can help here by optimizing these relationships into SQL JOINs. Instead of fetching the product and then each owner individually, it enables you to fetch the product with all requested relationships in a single database query by making use of TypeORM's relations option on find methods.

So in this simple example, instead of your resolvers producing this SQL:

SELECT * FROM product;
SELECT * FROM owner WHERE product_id = :x;
SELECT * FROM owner WHERE product_id = :y;
SELECT * FROM owner WHERE product_id = :z;

You can optimize it to:

SELECT * FROM product LEFT JOIN owner ON product.id = owner.product_id;

The value of this optimization increases as you have greater levels of nesting, of course.

You could join these relations manually (or eagerly) with TypeORM, but then you are likely to end up overfetching - retrieving relations that were not requested by the client and producing SQL that is more expensive than necessary. typeorm-relations-graphql only joins relations that were requested in the client's GQL query.

You could also use a DataLoader to batch requests, but this will usually still result in more database queries than are produced by joining relations. Beware however that large joins with many levels of nesting can be bad for performance, too, so you may need to mix approaches.

Usage

First, create a GraphRelationBuilder instance, passing in a TypeORM DataSource object (which provides access to entity metadata):

import { GraphRelationBuilder } from 'typeorm-relations-graphql';
import { dataSource } from './datasource';

const graphRelationBuilder = new GraphRelationBuilder(dataSource);

Inside a GraphQL query resolver (where you have a GraphQLResolveInfo object available) you can use GraphRelationBuilder to determine the relations you need to join to fulfill the query.

The build and buildForQuery methods of GraphRelationBuilder return a RelationMap instance. This is a class provided by the typeorm-relations package and contains methods that you can use to manipulate the relations before passing them to TypeORM. Read the typeorm-relations documentation to learn more about it.

GraphRelationBuilder

buildForQuery(entity: Constructor<Entity>, info: GraphQLResolveInfo): RelationMap<Entity>

Builds a RelationMap for an entity class by mapping from the root of the GraphQL query.

The entity passed in should be an entity class constructor (not an instance of the entity).

For example, if you have a products query in your GQL schema which returns a list of Product entities (where the Product entity and Product GQL object type are equivalent), you can simply map Product relations in this way:

import { GraphQLResolveInfo } from 'graphql';
import { dataSource } from './datasource';

// Example resolver function for a "products" query in your GQL schema
function products(source: any, args: any, context: any, info: GraphQLResolveInfo): Promise<Product[]> {
  const graphRelationBuilder = new GraphRelationBuilder(dataSource);

  const productRelationMap = graphRelationBuilder.buildForQuery(Product, info);

  return dataSource.getRepository(Product).find({
    relations: productRelationMap.toFindOptionsRelations(),
  });
}

In this example if your Product entity has an owner property that relates to another entity, and the owner field is selected by the client's GraphQL query, then calling toFindOptionsRelations() will produce:

{
  owner: true
}

Or, if owner contains an additional relationship to an address entity which is also selected by the client, you can get a nested structure like:

{
  owner: {
    address: true
  }
}

buildForQuery(entity: Constructor<Entity>, info: GraphQLResolveInfo, path: string): RelationMap<Entity>

In some cases you may need to map relations to entity fields where the GQL object type for the entity is not the root node in the query. A common example of this is in a mutation which returns a payload object containing the modified object rather than the object directly. In this case you can pass a path string as the last argument to buildForQuery:

import { GraphQLResolveInfo } from 'graphql';
import { dataSource } from './datasource';

// Example resolver function for a "createProduct" mutation in your GQL schema
async function createProduct(
  source: any,
  args: any,
  context: any,
  info: GraphQLResolveInfo,
): Promise<CreateProductPayload> {
  const graphRelationBuilder = new GraphRelationBuilder(dataSource);

  // Create the new product
  const product: Product = await dataSource.getRepository(Product).save(
    dataSource.getRepository(Product).create({
      name: 'New Product',
    }),
  );

  // Create payload and re-fetch the new product to retrieve all requested relations
  const payload: CreateProductPayload = {
    success: true,
    product: await dataSource.getRepository(Product).findOneOrFail({
      where: { id: product.id },
      relations: graphRelationBuilder.buildForQuery(Product, info, 'product').toFindOptionsRelations(),
    }),
  };

  return payload;
}

A GraphQL query for this mutation might look like:

mutation {
  createProduct {
    success
    product {
      id
      name
      owner {
        id
        name
      }
    }
  }
}

The Product entity here exists below the root level of the object being resolved (createProduct), at a field called product. So the path 'product' must be given to buildForQuery.

Dotted path notation can be used when the entity is at an even lower level in the node tree. For example, the path 'product.owner' could be used to map the Owner entity in this example.

build(entity: Constructor<Entity>, baseNode: SelectionNode, fragments?: Record<string, FragmentDefinitionNode>): GraphRelationBuilder<Entity>

This method works like buildForQuery (and is called by it internally), but it can operate on an arbitrary SelectionNode rather than requiring an entire GraphQLResolveInfo object.

If your GQL for the selection may contain named fragments, the definition of those fragments must be passed through. The required data can be retrieved from the fragments property on the top level GraphQLResolveInfo object.

License

MIT