request.js

A ~1kB wrapper around fetch with convenient error handling, automatic JSON transforms, and support for request timeouts and exponential backoff.

It works in the browser, on the server (Node.js), and mobile apps (like React Native), has no dependencies, and ships with TypeScript declarations.

Installation

npm i request-dot-js or yarn add request-dot-js.

Usage

API: async request(url[, options])

• url: a string
• options: basically the fetch init object

import request from 'request-dot-js'

// in an async function...

const { data, type, headers, status, statusText, url } = await request('https://httpbin.org/get', {
  headers: { 'Custom-Header': 'abcd' },
  params: { a: 'b', c: 'd' }, // query params
})

if (type === 'success') handleSuccess(data)

if (type === 'error') handleError(data)

if (type === 'exception') handleException(data)

For all responses returned by request.js, type is either 'success', 'error' or 'exception'.

If no exception is thrown, data is read from the response stream. If status < 300 type is 'success'. If status >= 300, type is 'error'.

If there's a connection error, timeout, or other exception, type is 'exception', and data is the exception thrown by fetch.

The other properties come from the Response object returned by fetch:

• status: 200, 204, etc...
• statusText: 'OK', 'CREATED', etc...
• url: url after redirect(s)
• headers: object literal instead of Headers instance; header names are lowercased

If type is 'exception', these properties are undefined.

JSON by default

request.js is built for easy interaction with JSON APIs, the de facto standard for data exchange on the web.

If you pass an object literal or an array for options.body, request adds the 'content-type': 'application/json' request header and JSON stringifies the body.

By default, request also adds the 'accept': 'application/json' request header and tries to return parsed JSON for data. If you don't want it to do this, pass jsonOut: false in options, and it will return the fetch Response object for data instead.

const { data, type, ...rest } = await request('https://httpbin.org/post', {
  method: 'POST', // default method is 'GET'
  body: { a: 'b', c: 'd' }, // object body automatically passed to JSON.stringify
})
if (type === 'success') console.log(data.url) // data is parsed JSON by default

const { data, type } = await request('https://httpbin.org/post', { jsonOut: false })
if (type === 'success') {
  const blob = await data.blob()
  console.log(blob)
}

Retries with exponential backoff

import request from 'request-dot-js'

// retry request up to 5 times, with 1s, 2s, 4s, 8s and 16s delays between retries
const { data, type, ...rest } = await request('https://httpbin.org/get', {
  params: { a: 'b', c: 'd' },
  retry: { retries: 5, delay: 1000 },
})

// shouldRetry function
const { data, type, ...rest } = await request('https://httpbin.org/get', {
  retry: {
    shouldRetry: (response, { retries, delay }) => {
      console.log(retries, delay) // do something with current retries and delay if you want
      return response.type === 'exception' || response.status === 500
    },
  },
})

The options parameter has a special retry key that can point to an object with retry options:

• retries
• delay (in ms)
• multiplier (default 2)
• shouldRetry (default response => response.type === 'exception')

If you pass a retry object, and your request throws an exception, request.js retries it up to retries times and backs off exponentially.

If on any retry you regain connectivity and type is 'success' or 'error' instead of 'exception', the request method stops retrying your request and returns the usual { data, type, ...rest }.

If you want to set a custom condition for when to retry a request, pass your own shouldRetry function. It receives the usual response, { data, type, ...rest }, and the current backoff values, { retries, delay }. If it returns a falsy value request.js stops retrying your request.

The shouldRetry function also lets you react to individual retries before request is done executing all of them, if you want to do that. See the example above.

Request timeout / aborting a request

Little known fact, but this is actually easy to do with fetch, it's just not supported on older browsers.

The options parameter has a special timeout key. If a timeout is specified, request.js aborts the request after timeout ms, and returns an exception response.

const { data, type } = await request('https://httpbin.org/get', { timeout: 1 })
// time out after 1ms -- type will be 'exception'

Convenience methods

request.js has convenience methods for the following HTTP methods: DELETE, GET, HEAD, OPTIONS, PATCH, POST, and PUT.

import request from 'request-dot-js'

const { data, type } = await request.delete('https://httpbin.org/delete')

const { data, type } = await request.put('https://httpbin.org/put', { body: { a: 'b' } })

These allow you to send non-GET requests without passing the method key in options.

Query stringification

request.js comes with a function that converts a query params object to a query string. If you want a fancier stringify function, like the one in qs, you can pass your own in options.

import qs from 'qs'

const { data, type } = await request('https://httpbin.org/get', { params: { a: 'b' }, stringify: qs.stringify })

TypeScript

request.js works great with TypeScript (it's written in TypeScript). Check out examples here.

Make sure you enable esModuleInterop if you're using TypeScript to compile your application. This option is enabled by default if you run tsc --init.

Regarding convenience methods, you can't do this:

import request from 'request-dot-js'

request.put(...)

Do this instead:

import { put } from 'request-dot-js'

put(...)

Also, you can't do this, because delete is a restricted keyword:

import { delete as del } from 'request-dot-js'

Do this instead:

import { del } from 'request-dot-js'

On the server

request.js ships with a module that works in the browser, index.js, and a module that works on the server, server.js. If you're using it on the server, make sure to import everything from server.js.

import request from 'request-dot-js/server'

Why request.js?

Why not axios, or just fetch? Unlike fetch, request.js is very convenient to use:

• automatic JSON transforms
• query params and response headers as object literals
• no double await to read data
• automatically removes body for GET and HEAD requests

And unlike either of them, it doesn't require try / catch to handle exceptions. const { data, type } = await request(...) has all the info you need to handle successful requests, request errors, connection errors, and request timeouts.

This design leads to another feature the others don't have: flexible, built-in support for retries with exponential backoff.

Dependencies

For modern browsers, or React Native, request.js has no dependencies.

If you're targeting older browsers, like Internet Explorer, you need to polyfill fetch and Promise, and use Babel to transpile your code.

If you use request.js on the server, node-fetch and abort-controller are the only dependencies.

Development and tests

Clone the repo, then yarn, then yarn test.