bigpipe

Bigpipe a radical new web framework for Node.js that's inspired by Facebook's bigpipe concept.

BigPipe

BigPipe is a radical new web framework for Node.JS. The general idea is to decompose web pages into small re-usable chunks of functionality called Pagelets and pipeline them through several execution stages inside web servers and browsers. This allows progressive rendering at the front-end and results in exceptional front-end performance.

Most web frameworks are based on a request and response pattern, a request comes in, we process the data and output a template. But before we can output the template we have to wait until all data has been received in order for the template to be processed. This doesn't make any sense for Node.js applications where everything is done asynchronously. When receiving your first batch of data, why not send it directly to the browser so it can start downloading the required CSS, JavaScript and render it.

BigPipe is distributed through the node package manager (npm) and is written against Node.js 0.10.x.

npm install --save bigpipe

To keep track of cross module compatibility, the imported components will be synced on minor releases. For example, bigpipe@0.5.0 will always be compatible with pagelet@0.5.0 and pipe.js@0.5.0.

Got stuck? Or can't wrap your head around a concept or just want some feedback, we got a dedicated IRC channel for that on Freenode:

  • IRC Server: irc.freenode.net
  • IRC Room: #bigpipe

Still stuck? Create an issue. Every question you have is a bug in our documentation and that should be corrected. So please, don't hesitate to create issues, many of them.

BigPipe

In all of these example we assume that your file is setup as:

'use strict';
 
var BigPipe = require('bigpipe');

public, returns BigPipe.

To create a BigPipe powered server can simply call the createServer method. This creates an HTTP or HTTPS server based on the options provided.

var bigpipe = BigPipe.createServer(8080, {
  pages: __dirname +'/pages',
  dist:  __dirname +'/dist'
});

The first argument in the function call is port number you want the server the listen on. The second argument is an object with the configuration/options of the BigPipe server. The following options are supported:

  • cache A cache which is used for storing URL lookups. This cache instance should have a .get(key) and .set(key, value) method. Defaults to false
  • dist The location of our folder where we can store our compiled CSS and JavaScript to disk. If the path or folder does not exist it will be automatically created. Defaults to working dir/dist.
  • pages A directory that contains your Page definitions or an array of Page constructors. Defaults to working dir/dist. If you don't provide Pages it will serve a 404 page for every request.
  • parser The message parser we should use for our real-time communication. See Primus for the available parsers. Defaults to JSON.
  • pathname The root path of an URL that we can use our real-time communication. This path should not be used by your Pages. Defaults to /pagelet
  • transformer The transformer or real-time framework we should for the real-time communication. We're bundling and using ws by default. See Primus for the supported transformers. Please note that you do need to add the transformer dependency to your package.json when you choose something else than ws.
  • redirect When creating a HTTPS server you could automatically start a HTTP server which redirects all traffic to the HTTPS equiv. The value is the port number on which this server should be started. Defaults to false.

In addition to the options above, all options of a HTTPS server are also supported. When you provide the server with cert and key files or set the port number to 443 it assumes you want to setup up a HTTPS server instead.

var bigpipe = BigPipe.createServer(443, {
  key: fs.readFileSync(__dirname +'/ssl.key', 'utf-8'),
  cert: fs.readFileSync(__dirname +'/ssl.cert', 'utf-8')
});

When you're creating an HTTPS server you got to option to also setup a simple HTTP server which redirects all content to HTTPS instead. This is done by supplying the redirect property in the options. The value of this property should be the port number you want this HTTP server to listen on:

var bigpipe = BigPipe.createServer(443, {
  ..
 
  key: fs.readFileSync(__dirname +'/ssl.key', 'utf-8'),
  cert: fs.readFileSync(__dirname +'/ssl.cert', 'utf-8'),
  redirect: 80
});

public, returns BigPipe.

If you want more control over the server creation process you can manually create a HTTP or HTTPS server and supply it to the BigPipe constructor.

'use strict';
 
var server = require('http').createServer()
  , BigPipe = require('bigpipe');
 
var bigpipe = new BigPipe(server, { options });

If you are using this pattern to create a BigPipe server instance you need to use the bigpipe.listen method to listen to the server. When this is called we will start our compiling all assets, attach the correct listeners to the supplied server, attach event listeners and finally listen on the server. The first argument of this method is the port number you want to listen on, the second argument is an optional callback function that should be called when server is listening for requests.

bigpipe.listen(8080, function listening() {
  console.log('hurray, we are listening on port 8080');
});

public, returns string.

bigpipe.version;

The current version of the BigPipe framework that is running.

public, returns BigPipe.

bigpipe.define(pages, callback);

Merge page or pages in the collection of existing pages. If given a string it will search that directly for the available Page files. After all dependencies have been compiled the supplied callback is called.

bigpipe.define('../pages', function done(err) {
 
});
 
bigpipe.define([Page, Page, Page], function done(err) {
 
}).define('../more/pages', function done(err) {
 
});

public, returns BigPipe.

bigpipe.before(name, fn, options);

BigPipe has two ways of extending it's build-in functionality, we have plugins but also middleware layers. The important difference between these is that middleware layers allow you modify the incoming requests before they are used by BigPipe.

There are 2 different kinds of middleware layers, async and sync. The main difference is that the sync middleware doesn't require a callback. It's completely optional and ideal for just introducing or modifying the properties on a request or response object.

All middleware layers need to be named, this allows you to enable, disable or remove the middleware layers. The supplied middleware function can either be a pre-configured function that is ready to modify the request and responses:

bigpipe.before('foo', function (reqres) {
  req.foo = 'bar';
});

Or an unconfigured function. We assume that a function is unconfigured if the supplied function has less than 2 arguments. When we detect these function we automatically call the function with the context that is set to BigPipe and the supplied options object and assume that it returns a configured middleware layer.

bigpipe.before('foo', function (configure) {
  return function (reqres) {
    res.foo = configure.foo;
  };
}, { foo: 'bar' });

If you're building async middleware layers, you simply need to make sure that your function accepts 3 arguments:

  • req The incoming HTTP request.
  • req The outgoing HTTP response.
  • next The continuation callback function. This function follows the error first callback pattern.
bigpipe.before('foo', function (reqresnext) {
  asyncthings(function (errdata) {
    req.foo = data;
    next(err);
  });
});

public, returns BigPipe.

bigpipe.remove(name);

Removes a middleware layer from the stack based on the given name.

bigpipe.before('layer', function () {});
bigpipe.remove('layer');

public, returns BigPipe.

bigpipe.disable(name);

Temporarily disable a middleware layer, it's not removed from the stack but it's just skipped when we iterate over the middleware layers. When a middleware layer has been disabled you can re-enable it.

bigpipe.before('layer', function () {});
bigpipe.disable('layer');

public, returns BigPipe.

bigpipe.enable(name);

Re-Enable a previously disabled module.

bigpipe.disable('layer');
bigpipe.enable('layer');

public, returns BigPipe.

bigpipe.use(name, plugin);

Plugins can be used to extend the functionality of BigPipe it self. You can control the client code as well as the server side code of BigPipe using the plugin interface.

bigpipe.use('ack', {
  // 
  // Only ran on the server. 
  // 
  serverfunction (bigpipeoptions) {
     // do stuff 
  },
 
  // 
  // Runs on the client, it's automatically bundled. 
  // 
  clientfunction (bigpipeoptions) {
     // do client stuff 
  },
 
  // 
  // Optional library that needs to be bundled on the client (should be a string) 
  // 
  library: '',
 
  // 
  // Optional plugin specific options, will be merged with Bigpipe.options 
  // 
  options: {}
});

required: writable, string

The HTTP pathname or URL we that we should respond to. The routing make use of the routable module for all of it's path matching. This allows you to use:

  • plain strings: '/foo/bar'
  • capturing strings: '/foo/:bar'
  • regexp: /^\/foo\/bar$/
  • capturing regexp: /^\/(foo|bar)\/bar$/
  • xRegExp: '/^\\/(?<named>[\\d\\.]+)\\/foo/'
Page.extend({
  path: '/'
}).on(module);

required: writable, string

The location of the base template which will be flushed to the browser as the first chunk of content. In your template we will introduce a bootstrap variable that needs to be placed in the <head> of HTML. Make sure when you're outputting this bootstrap variable is that it's not escaping the HTML tags.

<!doctype html>
<html class="no-js">
<head>
  <%- bootstrap %>
</head>

The template engine that you use should be supported by the temper project. The path the template is relative to the location of the Page. So you don't need to the nasty path.join(__dirname, 'folder') "hack" to set the correct template.

Page.extend({
  view: '../views/index.ejs'
}).on(module);

optional: writable, string

The meta character set for the Page. We add a <meta charset=""> to the bootstrap code by default so the browser doesn't have to do any HTML buffering in order to figure out what charset it should render the HTML in.

When you set this to null it will not include the meta charset, but this it not advised.

Default value: utf-8

Page.extend({
  charset: 'UTF-8'
}).on(module);

optional: writable, string

The Content-Type of the response. This defaults to text/html with a charset preset. The charset does not inherit it's value from the charset option.

Default value: text/html; charset=UTF-8

Page.extend({
  contentType: 'text/html; charset=UTF-7'
}).on(module);

optional: writable, string or array

Which HTTP methods should this page accept. It can be string with comma separated values or an Array with all the individual methods.

Default value: GET

Page.extend({
  method: ['GET', 'POST', 'HEAD']`
}).on(module);

Or using a string:

Page.extend({
  method: 'GET, POST, HEAD'
}).on(module);

optional: writable, number

The default status code that we should send back to the response.

Default value: 200

Page.extend({
  statusCode: 416
}).on(module);

optional: writable, object

Optional data that is passed through to the template renderer when rendering the view associated with this page.

Page.extend({
  data: {
    timestamp: Date.now()
  }
}).on(module);

optional: writable, function

An authorization handler to see if the request is authorized to interact with this page. This is set to null by default as there isn't any authorization in place. The authorization function will receive 2 arguments:

  • req: the http request that initialized the pagelet
  • done: a callback function that needs to be called with only a boolean.
Page.extend({
  authorize: function authorize(reqdone) {
    done(true); // True indicates that the request is authorized for access. 
  }
}).on(module);

optional: writable, string

What kind of generation mode should we render the pagelets. There are 3 different render modes available:

  • sync: Render all the pagelets in a single flush so we don't rely on JavaScript to be active in the browser to put the rendered pagelets in their correct positions again. This is set by default if your browser is not supporting JavaScript or doesn't support HTTP 1.1 chunking.
  • async: Render all pagelets as fast as possible and flush them to the response once they are done with rendering. The client side would then place the pagelets in their correct place holders. There is no pre-defined order when we are rendering.
  • pipeline: Almost the same as async rendering but the main difference is that the pagelets are flushed in the order that we're defined on the pagelet object.

Default value: async

Page.extend({
  mode: 'async'
}).on(module);

Everything in BigPipe is build upon the EventEmitter interface. It's either a plain EventEmitter or a proper stream. This a summary of the events we emit:

EventUsageLocationDescription
logpublicserverA new log message
transform::pageletpublicserverTransform a Pagelet
transform::pagepublicserverTransform a Page
listeningpublicserverThe server is listening
errorpublicserverThe HTTP server received an error
pagelet::configurepublicserverA new pagelet has been configured
page::configurepublicserverA new page has been configured

The library makes use the debug module and has all it's internals namespaced to bigpipe:. These debug messages can be trigged by starting your application with the DEBUG= env variable. In order to filter out all messages except BigPipe's message run your server with the following command:

DEBUG=bigpipe:* node <server.js>

The following DEBUG namespaces are available:

  • bigpipe:server The part that handles the request dispatching, page / pagelet transformation and more.
  • bigpipe:page Page generation.
  • bigpipe:compiler Asset compilation.
  • bigpipe:primus BigPipe Primus setup.
  • pagelet:primus Pagelet and Primus interactions
  • pagelet Pagelet interactions

Tests are automatically run on Travis CI to ensure that everything is functioning as intended. For local development we automatically install a pre-commit hook that runs the npm test command every time you commit changes. This ensures that we don't push any broken code in to this project.

Bigpipe is inspired by the concept behind Facebook's BigPipe. For more details read their blog post: Pipelining web pages for high performance.

BigPipe is released under MIT.