isomorphic-react-with-alt

Preface

This is an example showcasing how to build isomorphic react application with Alt which utilizes the same codebase on both server and browser and able to correctly handle multiple requests at one time.

Feel free to ask questions, chat or send over pull requests.

Features Highlights

There are already a bunch of isomorphic examples out there, what's special about this one?

• it showcased client-side routing using normal flux store (see RouteStore.js)

• it reuses data-fetching logics, no need to dupe them on the server and trying to keep both in sync in the future

• it uses Alt in a multiple-instance way instead of singleton, which is easier to maintain in the long run

• it's implementation is extremely simple and easy to understand

Why Isomorphic

• shorter time-to-first page for better user experience

• deep linking for better usability (ie: bookmarking)

• SEO

Goals

• reuse same codebase on both browser and server without any modification

• must be able to handle multiple requests on the server at the same time without polluting each other's data stores

• the approach must be easily understandable and simple to develop and maintain

• applicable to both isomorphic or browser-only applications

Key problems solved with this approach

• how to handle multiple requests at the same time since

  • flux store is singleton
  • data-fetching is async

• reuse router and routing table on both ends

• passing same data between browser and server to save a round trip

Why Alt

• it adheres to original flux implementation

• it adds just enough abstractions to avoid boilerplate code (ie: constants)

• it has nice auto-mappings between actionCreators and stores which saves time

• it supports isomorphic out of the box, with both singleton and multi-instance stores

• it provides Iso util to help with data marshalling

• in short, it provides a nice balance between simplicity and efficiency, which helps developer to do more and faster

How to run the sample

$ npm install --global babel
$ npm install
$ npm start

5000ft overview

Here's a 5000ft view of how a web application works on the server.

When a request hits the server, we figure out what data it needs, fetch those, feed them to react app, let the app populates all stores and renders the views, when it's all done we invoke React.renderToString to get a html string which can be returned to the client request.

One key goal here is we want to use the same codebase on both server and browser, without any modification.

Server

• see ./server.js

• although we are using Express here, you are free to use koa or any other stacks, as long as it supports javascript

• index.html

  • notice there's no physical index file

  • index page is concatenated on the fly from template strings because there are variables needed to be replaced

• routing

  • in order to avoid writing routing rules twice on both server and browser, we share the routing table (a file with plain old javascript inside) on both ends

  • when server.js first run, it reads in the routing table and create routing rules using express router

  • one added benefit of doing it this way is we get to handle 404 pages on the server, instead of delegating everything blindly to the client's router using a catch all rule

• handling requests

  • when a request hits the server, it will trigger one of the routing rule defined earlier

  • then our app will create a new instance of flux which contains standalone actionCreators and stores

  • each request is handled by a standalone flux hence no two requests can interfere with each other

  • then it passes in the requesting url to RouteStore and let it fetch the data needed and do the following job inside, by doing so we just need to write data fetching logic once and use it on both sides

  • one thing to note is RouteStore returns a promise so that we know when react app finishes it's work inside and it's safe for us to invoke React.renderToString to generate the final html strings

  • once we get the rendered html string, we augments it with special tags and serialize fetched data with the help from Iso (an utility provided by Alt which serialize and de-serialize data on both ends)

  • eventually we return the rendered string to the client, completes the request and respond cycle

Client

• flux instance on the client side

  • see flux.js

  • as mentioned earlier in Server section, we create a new flux instance for every incoming request to handle actionCreators and stores, we do the same on the client side by creating one flux instance to hold data and boot the app

  • inside each flux object it contains a standalone set of actionCreators and stores needed by the react app

  • this way we make sure data won't be mixed between different stores, hence ensuring user's privacy and security

  • flux.js handles all the heavy lifting for us, we just need to add in all needed actionCreators and stores there

  • one important thing to note though, we need to add actionCreators first so that it can be accessed later by stores

• Booting the app

  • all boot scripts lives inside boot.js, it's the entry point to our app

  • just like on the server, we create a new instance of flux first

  • then restore the app states with Iso.bootstrap() which de-serialize data passed from server and populate the client app's stores

  • once stores are populated and ready, we kick start the react app by rendering it's root component along with the flux instance created earlier

     React.render( React.createElement(App, {flux: flux}), container);

  • the benefit of doing so is we get to reuse data fetched by the server in the browser, hence saving an extra API call from client to the server to fetch those data again

• About the Store

  • each store module exposes a class definition by

     module.exports = ProductStore;

  • Unlike non-isomorphic alt implementation, which exposes a store instance by

     module.exports = alt.createStore(ProductStore, 'ProductStore');`

  • if you need to access actionCreators inside the stores, you can do that by:

     ActionCreators = this.alt.getActions('ActionCreators');

    normally we do that once inside the store's constructor and keep the reference for later use

  • this.alt is the flux instance we created earlier inside flux.js, remember that?

• About the view

  • remember when kicking start the react app, we passed in the flux instance which contains all actionCreators and stores needed through out the entire app?

  • We need to pass that flux instance down to all react components so that they can access actionCreators and stores

  • we use an undocumented react feature called context to help passing down the object (just be rest assured that context will stick around and it's safe to use)

  • in order to utilize context, we need to declare some variables in our root view, see js/app.js

     childContextTypes: {
          flux: React.PropTypes.object.isRequired
     },
    
     getChildContext: function() {
          return { flux: this.props.flux || new Error('flux not found!') };
     },

  • then in child components, we access that piece of information from the root view like this:

     componentWillMount() {
     	// retrieve the flux instance passed down from root view and keep the reference
     	flux = this.context.flux;
    
     	// then we get hold of actionCreators and stores needed in this child component
     	ActionCreators = flux.getActions('ActionCreators');
    
     	ProductStore = flux.getStore('ProductStore');
     }

  • we demonstrated writing react components in both ES5 (ProductItemContainer.js) and ES6 (ProductsContainer.js`), there are minor syntax and code structure differences you need to pay attention to

• About the router

  • We showcased a complete client-side routing solution too

    • the design goal here is to treat routing just like any other Stores, and make sure it's cleanly decoupled from the view

    • RouteStore provides client-side routing capabilities

    • it encapsulate page.js as the inner router (which can be swap out anytime)

    • provides a set of APIs (public methods) which can be invoked on both server and browser

    • drives the app via routing rules and handlers

  • Notice client-side routing rules are initialized inside a setTimeout call and it only gets run when in a browser environment

  • In short, we don't intend to directly reuse the same router on both server and browser, instead we share the routing table to avoid duped work

FAQ

• why avoid using react-router?

  IMHO router should not be tightly coupled with the view in any regard, wrapping it in a RouteStore is more flexible and decoupled, also we get to switch out different routers in the future (we used page.js in this example, but we tried a bunch of other routers which all worked pretty well)

• is this battle-tested?

  Yes, it's currently in production for some of the sites we are running

• need help?

  • First try the github issue section

  • Second, ask in the chat room

  • Lastly, shoot me an email

Notice

• currently we are depending on a custom-built flux dispatcher using npm-shrinkwrap.json due to the version available on npm is outdated, once that's updated to the latest one we will remove the dependency.

License

MIT