State management for the GraphQL heads
Clone or download
Ionut Achim
Latest commit a6414b9 Dec 12, 2017

README.md

Client for managing application state with GraphQL operations.

Build Status npm version npm downloads

Table of contents

The concept

Blips exposes a simple interface for managing your application state with GraphQL.

The state is contained inside a single store object. It can only be changed through mutations, you can read from it through queries and can also listen for changes with subscriptions.

The why?

I developed Blips because I wanted to use GraphQL with every project, regardless of what the API server looks like. So if my application is consuming a simple API (whatever kind), I can still have a store that manages the application state and write queries/mutations that would resolve with making requests to that API.

The basics

The following examples assume some familiarity with GraphQL. If you haven't used it before, or you have no idea about what it is, you should visit graphql.org and read about GraphQL in detail.

Installation

npm install blips graphql

The client instance

new BlipsClient({ typeDefs [, resolvers] } [, initialState] [, config] )

Creating a client requires type definitions, optional resolvers, optional initial state and an optional configuration object.

typeDefs
// types.js

export default `
  type Todo {
    id: String!
    label: String!
    completed: Boolean!
  }

  type Query {
    allTodos: [Todo]!
    todo(id: String!): Todo!
  }

  type Mutation {
    createTodo(id: String!, label: String!, completed: Boolean)
  }
`

The types.js file contains 3 definitions:

  • a basic Todo object type containing three fields: id, label, completed.
  • the root Query type
  • the root Mutation type

The ! marks the field as required. See graphql.org to read more about the type system.

resolvers
// resolvers.js

export default {
  Query: {
    allTodos: (obj, args, { store }) => store.state.todos || [],
  },
  Mutation: {
    createTodo: (
      obj,
      { id, label, completed = false },
      { store: { state: { todos } } }
    ) => {
      const newTodo = { id, label, completed }
      todos = [...todos, newTodo]
      return newTodo
    },
  },
}

The resolvers.js file provides definitions for allTodos query and the createTodo mutation. Each resolver function accepts three arguments:

  • obj: The previous object.
  • args: The arguments provided to the field in the GraphQL query.
  • context: Provides access to important information like the currently logged in user, or the store itself.

See graphql.org to read more about resolvers.

client
// index.js

import { BlipsClient } from 'blips'
import resolvers from './resolvers'
import typeDefs from './types'

const initialState = {
  todos: {
    '3c4a086e-2151-4b54-acb2-13044ea553c1': {
      id: '3c4a086e-2151-4b54-acb2-13044ea553c1',
      label: 'Buy milk',
      completed: false,
    },
  },
}
const schemaDef = { typeDefs, resolvers }

// new BlipsClient({ typeDefs [, resolvers] } [, initialState] [, config] )
const client = new BlipsClient(schemaDef, initialState)

You create a new client by providing your schema definitions ({ typeDefs, resolvers }) and an optional initial state. In addition to the schemaDefs and initialState, you can also privide a configuration object as the last and optional argument. The configuration parameter is used for tapping into the initial setup to define default variables or context properties and to configure the fetch method.

The client object has the following API:

  • state: getter for your entire state.
  • schema: getter for your generated schema. state-clerk.
  • query: method for executing queries.
  • mutate: method for executing mutations.
  • subscribe: method for registering subscriptions.
  • fetch: method for sending queries to a real GraphQL API.

Operations and Execution

You define operations in the form of GraphQL queries, mutations or subscriptions and use the client methods for executing them.

operations

// operations.js

export const todoQuery = `
  query todoQuery {
    todo {
      label
      completed
    }
  }
`

export const createTodoMutation = `
  mutation createTodoMutation($label: String!, $completed: Boolean) {
    createTodo(label: $label, completed: $completed) {
      id
      label
      completed
    }
  }
`

Read more about queries and mutations

execution

import { todoQuery, createTodoMutation } from './operations'

client
  .query(todoQuery, {
    variables: { id: '3c4a086e-2151-4b54-acb2-13044ea553c1' },
  })
  .then(res => {
    console.log(res)
    // {
    //   data: {
    //     todo: {
    //       label: 'Buy milk',
    //       completed: false
    //     }
    //   }
    // }
  })

client
  .mutate(createTodoMutation, {
    variables: {
      id: '4ecca858-67f8-491e-94cc-48b262061819',
      label: 'Learn Blips',
    },
  })
  .then(res => {
    console.log(res)
    // {
    //   data: {
    //     createTodo: {
    //       id: '4ecca858-67f8-491e-94cc-48b262061819',
    //       label: 'Learn Blips',
    //       completed: false
    //     }
    //   }
    // }
  })

If you've used Redux before you can think of execution methods as dispatchers. All executors may receive three arguments:

  • operation: can be in the form of source or documentAST
  • options: object containing:
    • variables: the operation variables
    • context: object containing any additional data that you want to pass to the resolvers, like the currently logged in user, tokens etc. (This will extend the default context)

All executors return promises that eventually resolve with an object containing a data prop which holds the requested data. If, for some reason, the operation was not successful, the data prop will hold an error object instead of the requested field.

Managing your state in an asynchronous way might seem scary but unless you're actually fetching data from a web service, hitting an API or performing database queries, your data will be available with the next tick. This behaviour also comes with a few benefits:

  • it forces you to introduce checks and error handlers which will eventually result in better code.
  • you can write abstractions or helpers for managing your data and they will apply everywhere.
  • it's much easier if you decide to switch from caching some of your state locally to exclusively fetching it from an external service.

Plus, using async/await will still make your code look synchronous and badass.

Subscriptions

  1. GraphQL subscriptions RFC
  2. Proposal for GraphQL Subscriptions by Apollo

In addition to polling queries or scheduling them to execute at different moments throughout your application life-cycle to keep your data up-to-date, you can also subscribe to changes through GraphQL subscriptions:

const allTodosSubscription = `
  subscription allTodosSubscription {
    allTodos {
      id
      label
      completed
    }
  }
`

const asyncIterator = await client.subscribe(allTodosSubscription)

/* Either use for-await-of */
try {
  for await (const tick of asyncIterator) {
    console.log(tick);
    // { data: { allTodos: [ ... ] } }
  }
} catch (e) {
  // { data: { error: { ... } } }
}


/* or transform the iterator into an observable and subscribe to it */
const sub = asyncIterator.toObservable().subscribe(
  next => {
    // { data: { allTodos: [ ... ] } }
  },
  error => {
    // { data: { error: { ... } } }
  }
)

// remember to dispose of any unused subscriptions
// sub.unsubscribe()

Blips uses Apollo's graphql-subscriptions PubSub implementation, where any query or mutation would publish data over a specific topic while your subscription resolvers subscribe to one or more topics.

In order to use subscriptions with Blips, you need to have access to the client's PubSub instance. We can achieve that by passing a resolvers function instead of an object when creating the client instance. This function accepts as first argument an object containing the PubSub instance and the withFilter method.

simple subscriptions

// resolvers.js

export default ({ pubsub, withFilter }) => ({
  Query: { ... },
  Subscription: {
    allTodos: {
      resolve: (obj, args, { store }) => store.state.todos || [],
      subscribe: () => pubsub.asyncIterator([ 'TODO_UPDATED', 'TODO_CREATED', 'TODO_DELETED', ]),
    }
  }
  Mutation: {
    createTodo: (obj, { id, label, completed = false }, { store: { state: { todos } } }) => {
      const newTodo = { id, label, completed }
      todos = [ ...todos, newTodo ]

      pubsub.publish('TODO_CREATED')
      return newTodo
    }
  }
})

In the above example, the createTodoMutation publishes over TODO_CREATED as soon as the todos get updated with the new entry. It doesn't need to publish any data because the allTodos subscription will return all the todos and it's resolve method will do just that.

filters

When publishing data to subscribers, we need to make sure that each subscribers get only the data it needs.

To do so, we can use withFilter helper from this package, which wraps AsyncIterator with a filter function, and let you control each publication for each user.

// types.js

export default {
  ...
  type Subscription {
    todo(id: String!): Todo!
  }
  ...
}

// resolvers.js

export default ({pubsub, withFilter}) => ({
  Query: { ... },
  Subscription: {
    todo: {
      subscribe: withFilter(
        () => pubsub.asyncIterator('TODO_UPDATED'),
        (payload, variables) => payload && payload.todo.id === variables.id
      )
    }
  },
  Mutation: {
    updateTodo: (obj, { id, label, completed }, { store }) => {
      const modified = store.patch(id, {
        label,
        completed,
      })('todos')
      pubsub.publish('TODO_UPDATED', { todo: modified })
      return modified
    }
  }
})

The updateTodoMutation will publish the updated todo over the TODO_UPDATED topic. If we have subscriptions registered for each todo, only those who's ID variable match to the updated todo's ID will emit.

Head over to graphql-subscriptions for a more detailed documentation on how to use this PubSub implementation.

Fetching data

Your access is not limited to the default context, nor limited to working with only the local state. Resolvers can completely ignore any context and just make async requests to external API servers.

REST API

const resolvers = {
  Mutation: {
    createTodo: async () =>
      await fetch('api/v1/todos', {
        method: 'post',
        body: JSON.stringify(args)
      }).then(res => res.json())
  }
}

const createTodoMutation = `
  mutation createTodoMutation($label: String!, completed: Boolean) {
    createTodo(label: $label, completed) {
      id
      label
      completed
    }
  }
`

const newTodo = await client.mutate(createTodoMutation, { variables: { label: 'Buy milk' } })

GraphQL API

import { BlipsClient } from 'blips'
import resolvers from './resolvers'
import typeDefs from './types'

const initialState = {}

const client = new BlipsClient({ typeDefs, resolvers }, initialState, { fetch: { uri: 'http://localhost:3000/graphql' } })

const allTodosQuery = `
  query allTodosQuery($first: Int) {
    allTodos(first: $first) {
      id
      label
      completed
    }
  }
`
// client.fetch(query [, options] [, operationName] )
const todos = await client.fetch(allTodosQuery, { variables: { first: 10 } })

Extending the context

You can extend the context provided to the resolvers in two ways:

1. When creating the client instance, by passing a context object through the options argument. This will extend the default context will be available to all your resolvers:

const client = new BlipsClient({ typeDefs, resolvers }, initialState, {
  context: { foo: 'bar' },
})

2. Through the options argument of every executor, which will make it available only for that execution context:

client.query(allBookmarksQuery, { context: { user: loggedUser } })

example:

// types
const typeDefs = `
  type Bookmarks {
    id: Int!,
    post: Post!,
    user: User!
  }

  type Query {
    allBookmarks: [Bookmark]!
  }
`

// resolvers
const resolvers = {
  Query: {
    allBookmarks: (object, args, { store, user }) => {
      // using identifier { user_id: user.id } to return only the current user's bookmarks
      return store.get('bookmarks', { user_id: user.id })
      // Returns:
      // [
      //   { id: 1, post_id: 11, user_id: 23 },
      //   { id: 3, post_id: 155, user_id: 23 }
      // ]
    },
  },
}

// initial state
const initialState = {
  bookmarks: [
    { id: 1, post_id: 11, user_id: 23 },
    { id: 2, post_id: 356, user_id: 77 },
    { id: 3, post_id: 155, user_id: 23 },
  ],
}

const client = new BlipsClient({ typeDefs, resolvers }, initialState)

const allBookmarksQuery = `
  query allBookmarksQuery {
    allBookmarks {
      id
      post
      user
    }
  }
`
const myBookmarks = client.query(allBookmarksQuery, {
  context: { user: loggedUser },
})

Use with React

This basics guide contains examples of using Blips with vanilla JS. The preferred way to use with React is with react-blips. Read the documentation there and/or check out the *-react-blips examples to get an idea of how it works.

The tips

dataloader

Since this is GraphQL, some of your operations may result in multiple queries for the same resource. Imagine the following scenario:

const state = {
  comments: {
    '3c4a086e-2151-4b54-acb2-13044ea553c1': {
      id: '3c4a086e-2151-4b54-acb2-13044ea553c1',
      message: 'Nice job!',
      created: 1510737448000,
      user_id: 'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0'
    },
    'f0ff08d2-9ea4-48b3-a77f-6b1517b5c827': {
      id: 'f0ff08d2-9ea4-48b3-a77f-6b1517b5c827',
      message: 'Great read!',
      created: 1510737617221,
      user_id: 'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0'
    },
  },
  users: {
    'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0'; {
      id: 'b1e7ed4d-7baa-4209-8ef3-0ccea2b420b0',
      firstName: 'John',
      lastName: 'Doe',
      comments: ['3c4a086e-2151-4b54-acb2-13044ea553c1', 'f0ff08d2-9ea4-48b3-a77f-6b1517b5c827']
      // ...
    }
  }
}

const types = `
  ...
  type Comment {
    id: String!
    message: String!
    user: User!
    created: Int!
  }
  ...
`

const resolvers = {
  // ...
  Query: {
    userCommentsQuery: (obj, { uid }, { store }) => store.get('comments', { user_id: uid }),
  },

  Comment: {
    user: ({ user_id }, args, { store }) => store.get('users', user_id),
  },
  // ...
}

const userCommentsQuery = `
  query userCommentsQuery($uid: String!) {
    userComments(uid: $uid) {
      idea
      message
      created
      user {
        id
        firstName
        photo
      }
    }
  }
`

We have a comments collection where each of the comments contain a user_id field that is a reference to it's poster. When userCommentsQuery is executed, the user field resolver for the comment will be called twice, even though it will return the same user. That is bad design for a graphql server (querying the database multiple times for the same resource), but it doesn't apply when managing the client state. If you were to not add a resolver for the Comment's user field, you would get the list of comments and then map over it to expand the user data. That would still result in querying the store for the same resource multiple times.

Now, because Blips manages all state asynchronously, we can use dataloader to batch and cache our resolvers, and that is a nice win and another benefit of managing your state asynchronously!

graphql-tag

With graphql-tag you can write your queries and type definitions in .graphql|gql files rather than .js that export template literals. All you need to do is add the loader to your webpack config.

mergers

You can further split your queries and/or type definitions in multiple modules and use a package that will merge them before creating the client instance. See merge-graphql-schemas or gql-merge

The kudos

  • Props to the awesome developers at Facebook for giving us GraphQL
  • Props to the guys at Apollo for their awesome work. Using apollo-client is what inspired me to start working on this library, and they also provide some of the tools it depends on.