Skip to content

ReactRelayNetworkLayer with middlewares and query batching for Relay Classic.

License

Notifications You must be signed in to change notification settings

relay-tools/react-relay-network-layer

Folders and files

NameName
Last commit message
Last commit date
Aug 5, 2016
Nov 3, 2017
Nov 3, 2018
Sep 6, 2018
Sep 6, 2018
Sep 6, 2018
May 23, 2018
Mar 4, 2017
Apr 22, 2016
Nov 6, 2017
Apr 22, 2016
Mar 4, 2017
Apr 22, 2016
Sep 6, 2018
Sep 6, 2018
Sep 6, 2018

Repository files navigation

ReactRelayNetworkLayer (for Relay Classic)

npm Travis Commitizen friendly semantic-release FlowType compatible

For Relay Modern please use react-relay-network-modern package.

The ReactRelayNetworkLayer is a Relay Network Layer with various middlewares which can manipulate requests/responses on the fly (change auth headers, request url or perform some fallback if request fails), batch several relay request by timeout into one http request.

ReactRelayNetworkLayer can be used in browser, react-native or node server for rendering. Under the hood this module uses global fetch method. So if your client is too old, please import explicitly proper polyfill to your code (eg. whatwg-fetch, node-fetch or fetch-everywhere).

yarn add react-relay-network-layer
OR
npm install react-relay-network-layer --save

Migrating from v1 to v2

Changes in v2.0.0:

  • completely rewritten batch logic as middleware, added additional cool options to it batchTimeout, maxBatchSize
  • throw Error object on non-200 response (before thrown response)
  • much more tests

All other parts stay unaffected. So if you use request batching, you should change your config:

import Relay from 'react-relay';
import {
  RelayNetworkLayer,
  urlMiddleware,
+  batchMiddleware,
} from 'react-relay-network-layer';

Relay.injectNetworkLayer(new RelayNetworkLayer([
+  batchMiddleware({
+    batchUrl: (req) => '/graphql/batch',
+  }),
  urlMiddleware({
    url: (req) => '/graphql',
-    batchUrl: (req) => '/graphql/batch',
  }),
- ], { disableBatchQuery: false }));
+ ]));

Big thanks to @brad-decker and @jeanregisser in helping to done this release.

Previous documentation for version 1.x.x can be found here.

Middlewares

Available middlewares:

  • your custom inline middleware - see example below where added credentials and headers to the fetch method.
    • next => req => { /* your modification of 'req' object */ return next(req); }
  • urlMiddleware - for manipulating fetch url on fly via thunk.
    • url - string or function(req) for single request (default: /graphql)
  • batchMiddleware - gather some period of time relay-requests and sends it as one http-request. You server must support batch request, how to setup your server
    • batchUrl - string or function(requestMap). Url of the server endpoint for batch request execution (default: /graphql/batch)
    • batchTimeout - integer in milliseconds, period of time for gathering multiple requests before sending them to the server. Will delay sending of the requests on specified in this option period of time, so be careful and keep this value small. (default: 0)
    • maxBatchSize - integer representing maximum size of request to be sent in a single batch. Once a request hits the provided size in length a new batch request is ran. Actual for hardcoded limit in 100kb per request in express-graphql module. (default: 102400 characters, roughly 100kb for 1-byte characters or 200kb for 2-byte characters)
    • allowMutations - by default batching disabled for mutations, you may enable it passing true (default: false)
  • retryMiddleware - for request retry if the initial request fails.
    • fetchTimeout - number in milliseconds that defines in how much time will request timeout after it has been sent to the server (default: 15000).
    • retryDelays - array of millisecond that defines the values on which retries are based on (default: [1000, 3000]).
    • statusCodes - array of response status codes which will fire up retryMiddleware (default: status < 200 or status > 300).
    • allowMutations - by default retries disabled for mutations, you may allow process retries for them passing true (default: false)
    • forceRetry - function(cb, delay), when request is delayed for next retry, middleware will call this function and pass to it a callback and delay time. When you call this callback, middleware will proceed request immediately (default: false).
  • authMiddleware - for adding auth token, and refreshing it if gets 401 response from server.
    • token - string or function(req) which returns token. If function is provided, then it will be called for every request (so you may change tokens on fly).
    • tokenRefreshPromise: - function(req, err) which must return promise or regular value with a new token. This function is called when server returns 401 status code. After receiving a new token, middleware re-run query to the server with it seamlessly for Relay.
    • allowEmptyToken - allow made a request without Authorization header if token is empty (default: false).
    • prefix - prefix before token (default: 'Bearer ').
    • header - name of the HTTP header to pass the token in (default: 'Authorization').
    • If you use auth middleware with retry, retry must be used before auth. Eg. if token expired when retries apply, then retry can call auth middleware again.
  • loggerMiddleware - for logging requests and responses.
    • logger - log function (default: console.log.bind(console, '[RELAY-NETWORK]'))
    • If you use Relay@^0.9.0 you may turn on relay's internal extended mutation debugger. For this you should open browser console and type __DEV__=true. With webpack you may use webpack.BannerPlugin('__DEV__=true;', {raw: true}) or webpack.DefinePlugin({__DEV__: true}).
    • If you use Relay@^0.8.0 you may turn on internal Relay requests debugger: import RelayNetworkDebug from 'react-relay/lib/RelayNetworkDebug'; RelayNetworkDebug.init();
  • perfMiddleware - simple time measure for network request.
    • logger - log function (default: console.log.bind(console, '[RELAY-NETWORK]'))
  • gqErrorsMiddleware - display errors data to console from graphql response. If you want see stackTrace for errors, you should provide formatError to express-graphql (see example below where graphqlServer accept formatError function).
    • logger - log function (default: console.error.bind(console))
    • prefix - prefix message (default: [RELAY-NETWORK] GRAPHQL SERVER ERROR:)
  • deferMiddleware - experimental Right now deferMiddleware() just set defer as supported option for Relay. So this middleware allow to community play with defer() in cases, which was described by @wincent.

Advanced options (2nd argument after middlewares)

RelayNetworkLayer may accept additional options:

const middlewares = []; // array of middlewares
const options = {}; // optional advanced options
const network = new RelayNetworkLayer(middlewares, options);

Available options:

  • noThrow - EXPERIMENTAL (May be deprecated in the future) set true to not throw when an error response is given by the server, and to instead handle errors in your app code.

Example of injecting NetworkLayer with middlewares on the client side.

import Relay from 'react-relay';
import {
  RelayNetworkLayer,
  urlMiddleware,
  batchMiddleware,
  loggerMiddleware,
  gqErrorsMiddleware,
  perfMiddleware,
  retryMiddleware,
  authMiddleware,
} from 'react-relay-network-layer';

Relay.injectNetworkLayer(new RelayNetworkLayer([
  urlMiddleware({
    url: (req) => '/graphql',
  }),
  batchMiddleware({
    batchUrl: (reqestMap) => '/graphql/batch',
    batchTimeout: 10,
  }),
  loggerMiddleware(),
  gqErrorsMiddleware(),
  perfMiddleware(),
  retryMiddleware({
    fetchTimeout: 15000,
    retryDelays: (attempt) => Math.pow(2, attempt + 4) * 100, // or simple array [3200, 6400, 12800, 25600, 51200, 102400, 204800, 409600],
    forceRetry: (cb, delay) => { window.forceRelayRetry = cb; console.log('call `forceRelayRetry()` for immediately retry! Or wait ' + delay + ' ms.'); },
    statusCodes: [500, 503, 504]
  }),
  authMiddleware({
    token: () => store.get('jwt'),
    tokenRefreshPromise: (req) => {
      console.log('[client.js] resolve token refresh', req);
      return fetch('/jwt/refresh')
        .then(res => res.json())
        .then(json => {
          const token = json.token;
          store.set('jwt', token);
          return token;
        })
        .catch(err => console.log('[client.js] ERROR can not refresh token', err));
    },
  }),

  // example of the custom inline middleware
  next => req => {
    // `req` is an object with settings for `fetch` function. It's not an express request object.
    // Internally works following code:
    //    let { url, ...opts } = req;
    //    fetch(url, opts)
    // So `req` is a fetch options. And into this options, I added `url` prop, which will be extracted as shown above.
    // You have fully control under `fetch` via `req` object.

    req.method = 'GET'; // change default POST request method to GET
    req.headers['X-Request-ID'] = uuid.v4(); // add `X-Request-ID` to request headers
    req.credentials = 'same-origin'; // provide CORS policy to XHR request in fetch method
    return next(req);
  }
]));

How middlewares work internally

Middlewares on bottom layer use fetch method. So req is compliant with a fetch() options. And res can be obtained via resPromise.then(res => ...), which returned by fetch().

Middlewares have 3 phases:

  • setup phase, which runs only once, when middleware added to the NetworkLayer
  • capturing phase, when you may change request object, and pass it down via next(req)
  • bubbling phase, when you may change response promise, made re-request or pass it up unchanged

Basic skeleton of middleware:

export default function skeletonMiddleware(opts = {}) {
  // [SETUP PHASE]: here you can process `opts`, when you create Middleware

  return next => req => {
    // [CAPTURING PHASE]: here you can change `req` object, before it will pass to following middlewares.
    // ...some code which modify `req`

    const resPromise = next(req); // pass request to following middleware and get response promise from it

    // [BUBBLING PHASE]: here you may change response of underlying middlewares, via promise syntax
    // ...some code, which may add `then()` or `catch()` to response promise
    //    resPromise.then(res => { console.log(res); return res; })

    return resPromise; // return response promise to upper middleware
  };
}

Middlewares use LIFO (last in, first out) stack. Or simply put - use compose function. So if you pass such middlewares [M1(opts), M2(opts)] to NetworkLayer it will be work such way:

  • call setup phase of M1 with its opts
  • call setup phase of M2 with its opts
  • for each request
  • call capture phase of M1
  • call capture phase of M2
  • call fetch method
  • call bubbling phase of M2
  • call bubbling phase of M1
  • chain to resPromise.then(res => res.json()) and pass this promise for resolving/rejecting Relay requests.

Batching several requests into one

Joseph Savona wrote: For legacy reasons, Relay splits "plural" root queries into individual queries. In general we want to diff each root value separately, since different fields may be missing for different root values.

Also if you use react-relay-router and have multiple root queries in one route pass, you may notice that default network layer will produce several http requests.

So for avoiding multiple http-requests, the ReactRelayNetworkLayer is the right way to combine it in single http-request.

Example how to enable batching

...on server

Firstly, you should prepare server to proceed the batch request:

import express from 'express';
import graphqlHTTP from 'express-graphql';
import { graphqlBatchHTTPWrapper } from 'react-relay-network-layer';
import bodyParser from 'body-parser';
import myGraphqlSchema from './graphqlSchema';

const port = 3000;
const server = express();

// setup standart `graphqlHTTP` express-middleware
const graphqlServer = graphqlHTTP({
  schema: myGraphqlSchema,
  formatError: (error) => ({ // better errors for development. `stack` used in `gqErrors` middleware
    message: error.message,
    stack: process.env.NODE_ENV === 'development' ? error.stack.split('\n') : null,
  }),
});

// declare route for batch query
server.use('/graphql/batch',
  bodyParser.json(),
  graphqlBatchHTTPWrapper(graphqlServer)
);

// declare standard graphql route
server.use('/graphql',
  graphqlServer
);

server.listen(port, () => {
  console.log(`The server is running at http://localhost:${port}/`);
});

More complex example of how you can use a single DataLoader for all (batched) queries within a one HTTP-request.

If you are on Koa@2, koa-graphql-batch provides the same functionality as graphqlBatchHTTPWrapper (see its docs for usage example).

...on client

And right after server side ready to accept batch queries, you may enable batching on the client:

Relay.injectNetworkLayer(new RelayNetworkLayer([
  batchMiddleware({
    batchUrl: '/graphql/batch', // <--- route for batch queries
  }),
]));

How batching works internally

Internally batching in NetworkLayer prepare list of queries [ {id, query, variables}, ...] sends it to server. And server returns list of responces [ {id, payload}, ...], (where id is the same value as client requested for identifying which data goes with which query, and payload is standard response of GraphQL server: { data, error }).

Recommended modules

  • babel-plugin-transform-relay-hot - Babel 6 plugin for transforming Relay.QL tagged templates via the GraphQL json schema file. Each time when schema file was changed, the wrapper updates instance of standard babelRelayPlugin with new schema without completely restarting dev server.

Contribute

I actively welcome pull requests with code and doc fixes. Also if you made great middleware and want share it within this module, please feel free to open PR.

CHANGELOG

License

MIT