When I decided to learn GraphQL I knew that the best way to do it was implementing its concepts, so I figure out that develop a chat application were a way to achieve my goal because would allow me to put into practice all GraphQL features, that’s what this post is about: Learning some GraphQL concepts by building a chat application.

Our application will be split into two parts, back-end, and front-end, as well as these posts, in this first post we will develop the server-side, to do so we’ll use NodeJS, Apollo Server, and of course GraphQL, we also will need a database and a query builder module, I used Knex and MySQL.

Before we continue, all the code is in this [repository](https://github.com/ricardo93borges/graphql-chat-server).

### Initial setup

Ok, first things first, let's start by creating the project and installing its dependencies.

Inside project folder:

```npm init```

And:

```npm i apollo-server bcrypt dotenv graphql jsonwebtoken knex lodash mysql```

``` npm i --save-dev @babel/cli @babel/core @babel/node @babel/plugin-transform-runtime @babel/preset-env babel-jest jest nodemon standard ```

In the scripts section of `package.json` put the following commands:

```
   "start": "nodemon --exec babel-node ./src/index.js",
    "test": "jest",
    "test:watch": "jest --watch",
    "migrate": "knex migrate:latest",
    "unmigrate": "knex migrate:rollback",
    "seed": "knex seed:run",
    "lint": "standard",
    "lint:fix": "standard --fix"
```

In the root folder create a `.babelrc` file:

```
{
  "presets": [
    "@babel/preset-env"
  ],
  "env": {
    "test": {
      "plugins": [
        "@babel/plugin-transform-runtime"
      ]
    }
  }
}
```

Also in the root folder create a `.env` file, this file contains the project’s environment variables: 

```
NODE_ENV=development

DB_HOST=localhost
DB_USER=root
DB_PASSWORD=toor
DB_NAME=chat

SECRET=secret
```

The first variable is the environment, let's leave it as `development` for now, the next four variables is database host, user, password, and name, for these you can set the values accordingly to your database configuration. The last one is the secret value that we’ll use later in the authentication.

Feel free to configure any relational database, I used MySQL, if want to use another, like PostgreSQL, you'll just have to do a different setup in the `knexfile.js`.

### Database and models

In this section we’ll configure our database and implement our models, in the root folder create a `knexfile.js` file, it contains database configuration for development, test, and production environments:

```js
require('dotenv').config()

module.exports = {

  development: {
    client: 'mysql',
    connection: {
      host: process.env.DB_HOST,
      user: process.env.DB_USER,
      password: process.env.DB_PASSWORD,
      database: process.env.DB_NAME
    },
    migrations: {
      directory: './src/data/migrations'
    },
    seeds: { directory: './src/data/seeds' }
  },

  test: {
    client: 'mysql',
    connection: {
      host: process.env.DB_HOST,
      user: process.env.DB_USER,
      password: process.env.DB_PASSWORD,
      database: process.env.DB_NAME
    },
    migrations: {
      directory: './src/data/migrations'
    },
    seeds: { directory: './src/data/seeds' }
  },

  production: {
    client: 'mysql',
    connection: {
      host: process.env.DB_HOST,
      user: process.env.DB_USER,
      password: process.env.DB_PASSWORD,
      database: process.env.DB_NAME
    },
    migrations: {
      directory: './src/data/migrations'
    },
    seeds: { directory: './src/data/seeds' }
  }
}
```

In `src/data/` we can store our database migrations, seeds, and a file that export a database object with the configurations from the `knexfile.js`:

```js

# src/data/db.js

import knex from 'knex'
import knexfile from '../../knexfile'

const env = process.env.NODE_ENV || 'development'
const configs = knexfile[env]
const database = knex(configs)

export default database
```

Now let’s create our migrations, run: 

`knex migrate:make user`
`knex migrate:make message`

The generated files are in the directory configured in `knexfile.js`, they must have the following contents:

```js
// src/data/migrations/20200107121031_user.js

exports.up = (knex) =>
  knex.schema.createTable('user', table => {
    table.bigIncrements('id').unsigned()
    table.string('name').notNullable()
    table.string('email').notNullable()
    table.string('password').notNullable()
  })

exports.down = (knex) => knex.schema.dropSchemaIfExists('user')
```

```js
// src/data/migrations/20200107121034_message.js

exports.up = (knex) =>
  knex.schema.createTable('message', table => {
    table.bigIncrements('id').unsigned()
    table.string('message').notNullable()
    table.bigInteger('senderId').unsigned().references('id').inTable('user')
    table.bigInteger('receiverId').unsigned().references('id').inTable('user')
  })

exports.down = function (knex) {
  knex.schema.dropSchemaIfExists('message')
}
```

can run our migrations, the following commands will create `user` and `message` tables in the database and populate it.

`npm run migrate`

Next, we create our models, let’s start by creating the `Model` class, it contains common methods used by other models that will extend it:

```js
// src/model/Model.js

export default class Model {
  constructor (database, table) {
    this.database = database
    this.table = table
  }

  all () {
    return this.database(this.table).select()
  }

  find (conditions) {
    return this.database(this.table).where(conditions).select()
  }

  findOne (conditions) {
    return this.database(this.table).where(conditions).first()
  }

  findById (id) {
    return this.database(this.table).where({ id }).select().first()
  }

  insert (values) {
    return this.database(this.table).insert(values)
  }
}
```

Then we create `User` and `Message` models, notice that in the `User` model there is a method to generate a token using the environment variable `SECRET` that we defined before, also there are methods to find a user by a token and to retrieve a user's messages.

```js
// src/model/User.js

import Model from './Model'
import bcrypt from 'bcrypt'
import jwt from 'jsonwebtoken'

export class User extends Model {
  constructor (database) {
    super(database, 'user')
  }

  async hash (password) {
    return bcrypt.hash(password, 10)
  }

  async compare (hash, password) {
    return bcrypt.compare(password, hash)
  }

  generateToken (user) {
    /* knex return a RowDataPacket object and jwt.sign function
      expects a plain object, stringify and parse it back does the trick */
    return jwt.sign(
      JSON.parse(JSON.stringify(user)),
      process.env.SECRET,
      {
        expiresIn: 86400
      }
    )
  }

  async getUserByToken (token) {
    try {
      const decoded = jwt.verify(token, process.env.SECRET)
      return decoded
    } catch (error) {
      console.log(error)
      return null
    }
  }

  async getMessages(senderId, lastId) {
    return this.database('message')
      .where('id', '>', lastId)
      .andWhere(q => q.where({ senderId: senderId })
        .orWhere({ receiverId: senderId }))
      .limit(10)
  }
```

```js
// src/model/Message.js

import Model from './Model'

export class Message extends Model {
  constructor (database) {
    super(database, 'message')
  }

  async getConversation (senderId, receiverId, lastId) {
    return this.database('message')
      .where('id', '>', lastId)
      .andWhere({ senderId })
      .andWhere({ receiverId })
      .limit(10)
  }

}
```

Now we have to export all those models, for the sake of organization, I’ve created a `index.js` file in `src/model` that export an object `models` containing all our models.

```js
// src/model/index.js

import database from '../data/db'
import { User } from '../model/User'
import { Message } from '../model/Message'

const user = new User(database)
const message = new Message(database)

const models = {
  user,
  message
}

export default models
```


### Schema

Finally, we’ll deal with GraphQL, let’s start with the schema, but what is the schema? The schema uses GraphQL schema language to define a set of types that our application will provide, a type can be, among others, a query, a mutation, a subscription, an object type, or a scalar type.

The query type defines the possible queries that our application will provide, for example, fetch all messages.

**Mutation type** is like queries but allows to modify data, for example, send a message.

**Subscription** allows the server to send data to a client when an event happens, usually is implemented with WebSockets, for example, in our chat application when a client sends a message, the receiver client must receive that message without request it to the server.

**Object type** defines an object that our application allows to be fetched, like user or message.

And **scalar types**, well, Object type has fields and these fields must have a value of some type, like string or int, these types are scalar types, the possible scalar types are Int, String, Float, Boolean, and ID. In some GraphQL implementations is possible to specify custom scalar types. When we use *!* means that the field is non-nullable and our service promises to return a non-nullable value. If we want to specify that our service will return an array we use *[]*, for example, `[String]!`.

Our GraphQL schema could be defined entirely in a single file, but as our application grows, that file would become a mess, so I decide to separate the schema into files that represent entities, so we’ll have a file to define user schema and another to define message schema, also there will be a file to bring all schema together, let’s start with this file:

```js
// src/schema/index.js

import { merge } from 'lodash'
import { gql, makeExecutableSchema } from 'apollo-server'
import {
  typeDef as User,
  resolvers as userResolvers
} from './user'

import {
  typeDef as Message,
  resolvers as messageResolvers
} from './message'

const Query = gql`
  type Query {
    _empty: String
  }
  type Mutation {
    _empty: String
  }
  type Subscription {
    _empty: String
  }
`
export const schema = makeExecutableSchema({
  typeDefs: [Query, User, Message],
  resolvers: merge(userResolvers, messageResolvers)
})

``` 

Next, we create the user and message schemas, you will notice that in each file there is an object called `resolvers` we will talk about it in a bit. Also notice that when we define the schema in the const `typeDef` we are extending the types Query, Mutation, and Subscription, we have to do this way because a GraphQL schema must have only one of each of these types.

```js
// src/schema/message.js

import { gql } from 'apollo-server'

export const subscriptionEnum = Object.freeze({
  MESSAGE_SENT: 'MESSAGE_SENT'
})

export const typeDef = gql`
  extend type Query {
    messages(cursor: String!): [Message!]!
    conversation(cursor: String!, receiverId: ID!): [Message!]!
  }
  extend type Subscription {
    messageSent: Message
  }
  extend type Mutation {
    sendMessage(sendMessageInput: SendMessageInput!): Message!
  }
  type Message {
    id: ID!
    message: String!
    sender: User!
    receiver: User!
  }
  input SendMessageInput {
    message: String!
    receiverId: ID!
  }
`

export const resolvers = {
  Query: {
    messages: async (parent, args, { models, user }, info) => {
      if (!user) { throw new Error('You must be logged in') }

      const { cursor } = args
      const users = await models.user.all()
      const messages = await models.user.getMessages(user.id, cursor)

      const filteredMessages = messages.map(message => {
        const sender = users.find(user => user.id === message.senderId)
        const receiver = users.find(user => user.id === message.receiverId)
        return { ...message, sender, receiver }
      })

      return filteredMessages
    },

    conversation: async (parent, args, { models, user }, info) => {
      if (!user) { throw new Error('You must be logged in') }

      const { cursor, receiverId } = args
      const users = await models.user.all()
      const messages = await models.message.getConversation(user.id, receiverId, cursor)

      const filteredMessages = messages.map(message => {
        const sender = users.find(user => user.id === message.senderId)
        const receiver = users.find(user => user.id === message.receiverId)
        return { ...message, sender, receiver }
      })

      return filteredMessages
    }
  },

  Subscription: {
    messageSent: {
      subscribe: (parent, args, { pubsub, user }, info) => {
        if (!user) { throw new Error('You must be logged in') }

        return pubsub.asyncIterator([subscriptionEnum.MESSAGE_SENT])
      }
    }
  },

  Mutation: {
    sendMessage: async (parent, args, { models, user, pubsub }, info) => {
      if (!user) { throw new Error('You must be logged in') }

      const { message, receiverId } = args.sendMessageInput

      const receiver = await models.user.findById(receiverId)

      if (!receiver) { throw new Error('receiver not found') }

      const result = await models.message.insert([{
        message,
        senderId: user.id,
        receiverId
      }])

      const newMessage = {
        id: result[0],
        message,
        receiver,
        sender: user
      }

      pubsub.publish(subscriptionEnum.MESSAGE_SENT, { messageSent: newMessage })

      return newMessage
    }
  }
}
```

```js
// src/schema/user.js

import { gql } from 'apollo-server'

export const typeDef = gql`
  extend type Query {
    users: [User!]!
  }
  extend type Mutation {
    createUser(createUserInput: CreateUserInput!): User!
    login(email: String!, password: String!): String!
  }
  type User {
    id: ID!
    name: String!
    email: String!
    password: String!
  }
  input CreateUserInput {
    name: String!
    email: String!
    password: String!
  }
`

export const resolvers = {
  Query: {
    users: async (parent, args, { models, user }, info) => {
      if (!user) { throw new Error('You must be logged in') }

      const users = await models.user.all()
      return users
    }
  },

  Mutation: {
    createUser: async (parent, args, { models }, info) => {
      const { name, email, password } = args.createUserInput
      const user = await models.user.findOne({ email })

      if (user) { throw new Error('Email already taken') }

      const hash = await models.user.hash(password)

      const result = await models.user.insert([{
        name,
        email,
        password: hash
      }])

      return {
        id: result[0],
        password: hash,
        name,
        email
      }
    },

    login: async (parent, args, { models }, info) => {
      const { email, password } = args

      const user = await models.user.findOne({ email })

      if (!user) { throw new Error('Invalid credentials') }

      if (!await models.user.compare(user.password, password)) { throw new Error('Invalid credentials') }

      return models.user.generateToken(user)
    }
  }
}
```

Each file has the schema defined in the const `typeDef` and the resolvers for this schema are in the resolver object.

So what are that resolvers objects? Resolvers contain the logic that will be executed when a query, mutation, or subscription defined in our application schema is called. They are functions that accept the following arguments:

**parent** The object that contains the result returned from the resolver on the parent field

**args** The arguments passed to the query, for example, the login mutation receives `email` and `password` arguments

**context** Is an object shared by all resolvers, in our application it contains the model object that we defined before and the logged in user.

**info** Contains information about the execution state of the query

So if you want to define the resolvers for the type Query put them in the `Query`, if want to define for Mutation type, put inside the `Mutation` object, and so on.

About pagination, I chose to use cursor-based pagination, you can see in the messages query in message schema, that query accepts a cursor as an argument, yes, we can pass arguments to GraphQL queries, the cursor value is the ID of the last message returned.

Now we have one last thing to do, and that is to define the application entry point (`src/index.js`): 

```js
//src/index.js

import { ApolloServer, PubSub } from 'apollo-server'

import { schema } from './schema'
import models from './model/index'

const pubsub = new PubSub()

const getUser = async (req, connection) => {
  let user = null

  if (req && req.headers.authorization) {
    const token = req.headers.authorization.replace('Bearer ', '')
    user = await models.user.getUserByToken(token)
  } else if (connection && connection.context.Authorization) {
    const token = connection.context.Authorization.replace('Bearer ', '')
    user = await models.user.getUserByToken(token)
  }

  return user
}

const server = new ApolloServer({
  schema,
  context: async ({ req, res, connection }) => {
    return {
      models,
      pubsub,
      user: await getUser(req, connection)
    }
  }
})

server.listen().then(({ url }) => {
  console.log(`🚀  Server ready at ${url}`)
})


```

Here we create an instance of ApolloServer with the schema we defined before, in the option `context` we set which resources will be available for the resolvers in the context argument before we return these resources we check if there is a logged in user using the token that we will receive from the request if you use `express` you can put the logic of fetching a user by a token in a middleware like in this [example](https://github.com/ricardo93borges/graphql-chat-server/blob/express/src/index.js)

The server will run in the default URL `http://localhost:4000/`, where you can test the application making some queries in GraphQL playground, you can learn more about [here](https://graphql.org/learn/queries/).

In the [part two](/learning-graphql-by-building-a-chat-application-part-2) we will develop the front-end using Apollo Client and ReactJS.


Learning some GraphQL concepts by building a chat application.

Ricardo Borges

Learning GraphQL by building a chat application - Part 1