Fork from apollo-datasource-mongodb

Apollo data source for MongoDB

npm i apollo-datasources-mongodb

This package uses DataLoader for batching and per-request memoization caching. It also optionally (if you provide a ttl) does shared application-level caching (using either the default Apollo InMemoryLRUCache or the cache you provide to ApolloServer()). It does this for the following methods:

• findOneById(id, options)
• findManyByIds(ids, options)
• findByFields(fields, options)

Usage

Basic

The basic setup is subclassing MongoDataSource, passing your collection or Mongoose model to the constructor, and using the API methods:

data-sources/Users.js

import { MongoDataSource } from 'apollo-datasource-mongodb'

export default class Users extends MongoDataSource {
  getUser(userId) {
    return this.findOneById(userId)
  }
}

and:

import { MongoClient } from 'mongodb'

import Users from './data-sources/Users.js'

const client = new MongoClient('mongodb://localhost:27017/test')
client.connect()

const server = new ApolloServer({
  typeDefs,
  resolvers,
  dataSources: () => ({
    users: new Users(client.db().collection('users'))
    // OR
    // users: new Users(UserModel)
  })
})

Inside the data source, the collection is available at this.collection (e.g. this.collection.update({_id: 'foo, { $set: { name: 'me' }}})). The model (if you're using Mongoose) is available at this.model (new this.model({ name: 'Alice' })). The request's context is available at this.context. For example, if you put the logged-in user's ID on context as context.currentUserId:

class Users extends MongoDataSource {
  ...

  async getPrivateUserData(userId) {
    const isAuthorized = this.context.currentUserId === userId
    if (isAuthorized) {
      const user = await this.findOneById(userId)
      return user && user.privateData
    }
  }
}

If you want to implement an initialize method, it must call the parent method:

class Users extends MongoDataSource {
  initialize(config) {
    super.initialize(config)
    ...
  }
}

If you're passing a Mongoose model rather than a collection, Mongoose will be used for data fetching. All transformations defined on that model (virtuals, plugins, etc.) will be applied to your data before caching, just like you would expect it. If you're using reference fields, you might be interested in checking out mongoose-autopopulate.

Batching

This is the main feature, and is always enabled. Here's a full example:

class Users extends MongoDataSource {
  getUser(userId) {
    return this.findOneById(userId)
  }
}

class Posts extends MongoDataSource {
  getPosts(postIds) {
    return this.findManyByIds(postIds)
  }
}

const resolvers = {
  Post: {
    author: (post, _, { dataSources: { users } }) => users.getUser(post.authorId)
  },
  User: {
    posts: (user, _, { dataSources: { posts } }) => posts.getPosts(user.postIds)
  }
}

const server = new ApolloServer({
  typeDefs,
  resolvers,
  dataSources: () => ({
    users: new Users(db.collection('users')),
    posts: new Posts(db.collection('posts'))
  })
})

Caching

To enable shared application-level caching, you do everything from the above section, and you add the ttl (in seconds) option to findOneById():

const MINUTE = 60

class Users extends MongoDataSource {
  getUser(userId) {
    return this.findOneById(userId, { ttl: MINUTE })
  }

  updateUserName(userId, newName) {
    this.deleteFromCacheById(userId)
    return this.collection.updateOne({
      _id: userId
    }, {
      $set: { name: newName }
    })
  }
}

const resolvers = {
  Post: {
    author: (post, _, { users }) => users.getUser(post.authorId)
  },
  Mutation: {
    changeName: (_, { userId, newName }, { users, currentUserId }) =>
      currentUserId === userId && users.updateUserName(userId, newName)
  }
}

Here we also call deleteFromCacheById() to remove the user from the cache when the user's data changes. If we're okay with people receiving out-of-date data for the duration of our ttl—in this case, for as long as a minute—then we don't need to bother adding calls to deleteFromCacheById().

TypeScript

Since we are using a typed language, we want the provided methods to be correctly typed as well. This requires us to make the MongoDataSource class polymorphic. It requires 1-2 template arguments. The first argument is the type of the document in our collection. The second argument is the type of context in our GraphQL server, which defaults to any. For example:

data-sources/Users.ts

import { MongoDataSource } from 'apollo-datasource-mongodb'
import { ObjectId } from 'mongodb'

interface UserDocument {
  _id: ObjectId
  username: string
  password: string
  email: string
  interests: [string]
}

// This is optional
interface Context {
  loggedInUser: UserDocument
}

export default class Users extends MongoDataSource<UserDocument, Context> {
  getUser(userId) {
    // this.context has type `Context` as defined above
    // this.findOneById has type `(id: ObjectId) => Promise<UserDocument | null | undefined>`
    return this.findOneById(userId)
  }
}

and:

import { MongoClient } from 'mongodb'

import Users from './data-sources/Users.ts'

const client = new MongoClient('mongodb://localhost:27017/test')
client.connect()

const server = new ApolloServer({
  typeDefs,
  resolvers,
  dataSources: () => ({
    users: new Users(client.db().collection('users'))
    // OR
    // users: new Users(UserModel)
  })
})

API

The type of the id argument must match the type used in the database. We currently support ObjectId and string types.

findOneById

this.findOneById(id, { ttl })

Resolves to the found document. Uses DataLoader to load id. DataLoader uses collection.find({ _id: { $in: ids } }). Optionally caches the document if ttl is set (in whole positive seconds).

findManyByIds

this.findManyByIds(ids, { ttl })

Calls findOneById() for each id. Resolves to an array of documents.

findByFields

this.findByFields(fields, { ttl })

Resolves to an array of documents matching the passed fields. If an empty object is passed as the fields parameter, resolves to an array containing all documents in the given collection.

fields has this type:

interface Fields {
  [fieldName: string]:
    | string
    | number
    | boolean
    | ObjectId
    | (string | number | boolean | ObjectId)[]
}

Examples

// get user by username
// `collection.find({ username: $in: ['testUser'] })`
this.findByFields({
  username: 'testUser'
})

// get all users with either the "gaming" OR "games" interest
// `collection.find({ interests: $in: ['gaming', 'games'] })`
this.findByFields({
  interests: ['gaming', 'games']
})

// get user by username AND with either the "gaming" OR "games" interest
// `collection.find({ username: $in: ['testUser'], interests: $in: ['gaming', 'games'] })`
this.findByFields({
  username: 'testUser',
  interests: ['gaming', 'games']
})

deleteFromCacheById

this.deleteFromCacheById(id)

Deletes a document from the cache that was fetched with findOneById or findManyByIds.

deleteFromCacheByFields

this.deleteFromCacheByFields(fields)

Deletes a document from the cache that was fetched with findByFields. Fields should be passed in exactly the same way they were used to find with.