An implementation of Facebook's Flux architecture.


The Facebook / React team has an introduction to Flux included with the React documentation. Distilled to its core, Flux reimagines the traditional MVC approach to client-side webapps, replacing it with the same concept of "one-way data flow" that powers React.

While the Facebook documentation (and accompanying video) do an excellent job of introducing the core concepts behind Flux, there's a few key details that are important to emphasize:

###Stores All application data is managed in Stores, which are Singleton objects focused on a specific set of business logic (e.g ArticleStore, UserStore). Views should interact with Stores as the single source of data truth. Store do not replace the React state system; rather, React components should use state to handle view-specific data state, and stores to handle application data state. Stores are event emitters that emit change events for underlying state changes.

###Actions While Views access data directly from Stores, they never mutate data directly. Rather, Views should trigger Actions, which in turn trigger broadcast notifications throughout the system. Broadcast notifications are identified via lookups in enum Constants, and include a payload object.

For example, clicking a button in a view to favorite an article would make a call to an ArticleFavorite action, which would then broadcast a FAVORITE_ARTICLE notification with an appropriate payload for any Stores that want to perform an optimistic update. The Action would then interact with a DAO or Service to perform the actual server-side favorite; depending on the result of this interaction, the Action would then broadcast a FAVORITE_ARTICLE_COMPLETED or FAVORITE_ARTICLE_FAILED notification.

###Dispatcher Each of the notifications broadcast via the Actions described above are marshalled through a central bus, the Dispatcher. The Dispatcher ensures that only one notification can be handled at a time; new notifications are queued until all action handlers have finished operating for the previous notification. Stores register their action handlers for specific notifications directly with the Dispatcher, and can specify action handler dependencies - that is, an action handler can have its invocation delayed until the completion of other action handlers.

###This Implementation Facebook has not yet released their own implementation of Flux, but the JS community has started the ideas behind the Flux architecture, most notably in @BinaryMuse's fluxxor. Fluxy is an implementation that includes some differentiating features:

  • View components should be completely separated from Flux. They can intereact with Flux Stores and Actions via direct API calls, without coupling them to the Flux implementation.

  • Promises (via Bluebird) drive the Dispatcher/Action system, along for the easy registration of async action handlers

  • Stores embrace immutable data, going so far as to be powered by Mori, which provides a light convenience API around ClojureScript's data structures.

##How it Works A Fluxy implementation will define and use three different types of service objects: Stores, Actions, and Constants. While these three types of objects operate in concert, their roles are very distinct from one another - in fact, a Store should never call an Action, and an Action should never call into a Store. Instead, the Dispatcher unifies Action and Store objects, with messages defined by Constants. (All of the details of the Dispatch queue are hidden from the actual application implementation, so you don't need to worry about how the communication is happening behind the scenes.)

In order to keep your application modular, there's just one rule to remember: Views issue COMMANDS to Actions, and QUERIES to Stores.

That means that any updates to application state should be routed through an Action, and never via a direct call from a view into a Store.

###Actions Actions can be considered command objects. They are called directly by view components. They trigger service updates directly, and send messages to Stores by calling this.dispatchAction(messageName, ...args).

An example Action can be seen below:

var Fluxy = require('fluxy');
var TodoConstants = require('../constants/TodoConstants');
var TodoService = require('../services/TodoService');
var TodoActions = Fluxy.createActions({
  serviceActions: {
    create: [TodoConstants.TODO_CREATE, function (text) {
      return TodoService.create(text); //returns promise 
    destroy: [TodoConstants.TODO_DESTROY, function (id) {
      return TodoService.destroy(id); //returns promise 
  toggleExpandedfunction (expandFlag) {
    this.dispatchAction(TodoConstants.TODO_TOGGLE_EXPANDED, expandFlag);
module.exports = TodoActions;

The most interesting aspect to Actions is the special definition of serviceActions. Because so many Fluxy behaviors amount to the same flow - Action Message leading to a service call, which either succeeds (MESSAGE_COMPLETED) or fails (MESSAGE_FAILED) - Fluxy defines a special type of action to reduce boilerplate. serviceActions automatically wire up a promise and message dispatch queue.

Let's take the first service action defined: create. That service action will define a create(text) method on the TodoActions singleton. Calling create(text) will AUTOMATICALLY dispatch a TODO_CREATE message. The service action will then call the service method, and wire up promise result and failure handlers.

All of that means, most simply, that this:

var TodoActions = Fluxy.createActions({
  serviceActions: {
    create: [TodoConstants.TODO_CREATE, function (text) {
      return TodoService.create(text); //returns promise 

Is actually the same as this:

var TodoActions = Fluxy.createActions({
  createfunction (text) {
    var self = this;
      .then(function (result) {
        self.dispatchAction(TodoConstants.TODO_CREATE_COMPLETED, result);
      .catch(function (err) {
        self.dispatchAction(TodoConstants.TODO_CREATE_FAILED, err);

See how much boilerplate that saved? It's like a poor dev's macro.

There's nothing preventing you from defining the create action in its longer form. Any methods can be defined on an Action. serviceActions just make your life a little easier.

###Constants Constants are exactly that: Enums of constants. While Constants are primarily used to define string message names for dispatching, they can also store any constant values used by the app.

An example:

  var Fluxy = require('fluxy');
  var TodoConstants = Fluxy.createConstants({
    serviceMessages: [
    messages: ['TODO_CHANGED'],
    values: {
      MAX_TODOS: 99
  module.exports = TodoConstants;

A Constant singleton is really just an Enum - in fact, the createConstants call is simply deferring to the Enum constructor. messages and servicesMessages have equivalent key and value pairs. values have a string key and a variable value.

serviceMessages perform the same helpful function as serviceActions above - they accomodate for the routing call/completed/failed service flow. So each string defined in the serviceMessages array actually creates three string constants. In the example above, TodoConstants.TODO_CREATE_COMPLETED and TodoConstants.TODO_CREATE_FAILED are created in addition to TODO_CREATE.

messages create string constants as well, they just don't add the COMPLETED and FAILED pairings.

values can have any type of value. In the above example, TodoConstants.MAX_TODOS.value will equal 99.

###Stores Stores are the heart of the Fluxy implementation. In addition to responding to message dispatches, they manage all application state via immutable mori data structures.

An example Store implementation:

var TodoConstants = require('../constants/TodoConstants');
var Fluxy = require('fluxy');
var $ = Fluxy.$;
var TodoStore = Fluxy.createStore({
  getInitialStatefunction () {
    return {
      todos: {}
  areAllCompletefunction () {
    return $.every(function (todo) {
      return $.get(todo, 'completed') === true;
    }, $.vals(this.get('todos')));
  actions: [
    [TodoConstants.TODO_CREATE_COMPLETED, function (todo) {
      this.set(['todos', todo.id.toString()], $.js_to_clj(todo));
    [TodoConstants.TODO_COMPLETE_ALL, function () {
      this.set(['todos'], function (todoMap) {
        return $.reduce_kv(
          function(acckeyval) {
            return $.assoc(acc, key, $.assoc(val, 'complete', true));

Stores are the most complex part of Fluxy, and require a much deeper explanation.

First, lets look at the actions definition. actions autowrite message handlers to the dispatcher. In the example above, a TodoStore.handleTodoCreateCompleted function is created. This function is automatically wired to the Fluxy Dispatcher and listens to the TodoConstants.TODO_CREATE_COMPLETED message. The actual handler behavior is defined as a function in the array definition.

Also note that it's possible to create dependencies in how Stores handle dispatches. In the below example, the defined handler is called AFTER the SessionStore handler has finished.

  actions: [
    [UserConstants.LOGIN_COMPLETED, {waitFor: [SessionStore]}, function (user) {
      this.set('user', user);

And if the SessionStore handler had returned a promise, the handler above wouldn't execute until after the promise had resolved.

You'll note the use of $ in the Store. That's not a jQuery reference! It's a reference to the mori API. You can access mori directly through Fluxy.$, or directly on the Store singleton - all of mori's functions are proxied through to the Store and prefixed with a $. (So Store.$equals will proxy to mori.equals).

getInitialState defines the initial state of the Store. The object that's returned by the getInitialState function will be cast into a ClojureScript data structure. It's accessible directly via Store.state, but most of the interactions should be made through the helper API functions get and set.

get(keyOrKeyArray) calls directly into Store.state. It can be provided with a string key or an array of keys. A string key lookup is the rough equivalent of state.<key>. The array lookup is much more powerful, as it allows for "looking into" the Store state, reaching deeply into a nested data structore. For example, get(['todos', 571, 'text']) is the rough equivalent of Store.state.todos[571].text. (Don't worry, ClojureScript won't throw an undefined error as it walks its data structure!)

set(keyOrArrayOfKeys, valOrFn) is the setter equivalent of get. In addition to performing the same flat or deep access that get provides, it can also update a value in one of two ways - either by assigning a value directly, or as the result of an update function. In the example below...

Store.set(['todos', 101, 'complete'], function (completeFlag) { return !completeFlag; });

...the state of todo 101's completion is being toggled by the update function. This notation allows for some nice functional composition.

There are a few other benefits to using the set helper. First, it not only updates the Store state - it handles the storage of the Store's existing state. That's right - the entire history of Store states is tracked in Store.state. That means you can always call Store.undo() to rollback state. After updating the Store's state and state history, set will trigger an event for any defined watchers with references to both the previous state and the new state.

In addition to set, there's a corresponding setFromJS method which automatically converts the set value into a ClojureScript data structure (e.g. a JS object to a CLJS map). It delegates to the underlying Store.set method.

Watchers are defined against the Store directly:

Store.addWatch(function (keysoldStatenewState) {...});

And removed just as easily:


Note the keys argument of the watcher handler. It's a reference to string or array of keys passed into the set function. Using keys, you can easily figure out whether your watcher handler should care about the updated state (or not).

Note that get by default returns the ClojureScript data structure. You can access the JS representation either by calling Store.getAsJS, which behaves the same as Store.get, or by calling Store.toJS(cljObj), which casts the provided structure to JavaScript. Using the ClojureScript objects allows for easy tests in your shouldComponentUpdate functions, but otherwise, ClojureScript objects should not be used extensively (if at all) in your view components, for risk of tying your components too tightly to the Fluxy implementation.

###"Harness" Stores, Actions, and Constants are all managed by a singleton Fluxy. In addition to allowing you to create these key components of a Fluxy app, it also allows you to easily access the global application state.

createStore, createActions, and createConstants are all described in the respective sections above.

start(initialState) is the call that triggers the instantiation and injection of all Fluxy components. If you pass a hash map into the start function, it will set the state of each store to reflect the injected map. For example, say you have a Store with a name of TodoStore and call start with {TodoStore: { todos: todosArr }}; making this call will start the app with the TodoStore having the todo key in its state set to the value of todosArr.

bootstrap(prop, context) is a convenience method for calling start with the value of a window prop (or a prop on a provided context). Bootstrap is most useful for, appropriately, bootstrapping an application's state with serialized JSON embedded in the page's HTML.

renderStateToString(serializer) serializes the entire graph of all Store states to a string, with the keys of the serialization corresponding to each Store's name. While this method will default to using a "safe" version of JSON.stringify, any serialization function can be provided - for example, the write function from cognitect/transit-js.

start, bootstrap, and renderStateToString can be easily combined to power server-side rendering. Just follow the below steps:

  1. Give your stores a unique name. Fluxy.createStore({name: 'TodoStore', //other config})

  2. Before calling React.renderComponentToString, make sure to call Fluxy.start on the server, passing it a hash of Store initial states (keyed by Store name)

  3. Make sure to bootstrap your HTML load with the state you passed to Fluxy with Fluxy.renderStateToString

  4. Finally, in your client side code, instead of calling Fluxy.start, call Fluxy.bootstrap(windowKeyOfBootstrappedData)

For further details, be sure to check out the examples directory and the test suite.

##Roadmap to 1.0

  • [x] Update generators to work with new API
  • [x] Update example app to work with new API
  • [x] Provide basic implementation steps in the README
  • [ ] Lock down API
  • [ ] Add code documentation
  • [ ] Cleanup internal implementation