Mastering Node

Node is an exciting new platform developed by Ryan Dahl, allowing JavaScript developers to create extremely high performance servers by leveraging Google's V8 JavaScript engine, and asynchronous I/O. In Mastering Node we will discover how to write high concurrency web servers, utilizing the CommonJS module system, node's core libraries, third party modules, high level web development and more.

Installing Node

In this chapter we will be looking at the installation and compilation of node. Although there are several ways we may install node, we will be looking at homebrew, nDistro, and the most flexible method, of course - compiling from source.

Homebrew

Homebrew is a package management system for OSX written in Ruby, is extremely well adopted, and easy to use. To install node via the brew executable simply run:

$ brew install node.js

nDistro

nDistro is a distribution toolkit for node, which allows creation and installation of node distros within seconds. An nDistro is simply a dotfile named .ndistro which defines module and node binary version dependencies. In the example below we specify the node binary version 0.1.102, as well as several 3rd party modules.

node 0.1.102
module senchalabs connect
module visionmedia express 1.0.0beta2
module visionmedia connect-form
module visionmedia connect-redis
module visionmedia jade
module visionmedia ejs

Any machine that can run a shell script can install distributions, and keeps dependencies defined to a single directory structure, making it easy to maintain an deploy. nDistro uses pre-compiled node binaries making them extremely fast to install, and module tarballs which are fetched from GitHub via wget or curl (auto detected).

To get started we first need to install nDistro itself, below we cd to our bin directory of choice, curl the shell script, and pipe the response to sh which will install nDistro to the current directory:

$ cd /usr/local/bin && curl http://github.com/visionmedia/ndistro/raw/master/install | sh

Next we can place the contents of our example in ./.ndistro, and execute ndistro with no arguments, prompting the program to load the config, and start installing:

$ ndistro

Installation of the example took less than 17 seconds on my machine, and outputs the following stdout indicating success. Not bad for an entire stack!

... installing node-0.1.102-i386
... installing connect
... installing express 1.0.0beta2
... installing bin/express
... installing connect-form
... installing connect-redis
... installing jade
... installing bin/jade
... installing ejs
... installation complete

Building From Source

To build and install node from source, we first need to obtain the code. The first method of doing so is via git, if you have git installed you can execute:

$ git clone http://github.com/ry/node.git && cd node

For those without git, or who prefer not to use it, we can also download the source via curl, wget, or similar:

$ curl -# http://nodejs.org/dist/node-v0.1.99.tar.gz > node.tar.gz
$ tar -zxf node.tar.gz

Now that we have the source on our machine, we can run ./configure which discovers which libraries are available for node to utilize such as OpenSSL for transport security support, C and C++ compilers, etc. make which builds node, and finally make install which will install node.

$ ./configure && make && make install

CommonJS Module System

CommonJS is a community driven effort to standardize packaging of JavaScript libraries, known as modules. Modules written which comply to this standard provide portability between other compliant frameworks such as narwhal, and in some cases even browsers.

Although this is ideal, in practice modules are often not portable due to relying on apis that are currently only provided by, or are tailored to node specifically. As the framework matures, and additional standards emerge our modules will become more portable.

Creating Modules

Let's create a utility module named utils, which will contain a merge() function to copy the properties of one object to another. Typically in a browser, or environment without CommonJS module support, this may look similar to below, where utils is a global variable.

var utils = {};
  utils.merge = function(obj, other) {};

Although namespacing can lower the chance of collisions, it can still become an issue, and when further namespacing is applied it can look flat-out silly. CommonJS modules aid in removing this issue by "wrapping" the contents of a JavaScript file with a closure similar to what is shown below, however more pseudo globals are available to the module in addition to exports, require, and module. The exports object is then returned when a user invokes require('utils').

var module = { exports: {}};
  (function(module, exports){
      function merge(){};
      exports.merge = merge;
  })(module, module.exports);

First create the file ./utils.js, and define the merge() function as seen below. The implied anonymous wrapper function shown above allows us to seemingly define globals, however these are not accessible until exported.

  function merge(obj, other) {
      var keys = Object.keys(other);
      for (var i = 0, len = keys.length; i < len; ++i) {
          var key = keys[i];
          obj[key] = other[key];
      }
      return obj;
  };

  exports.merge = merge;

The typical pattern for public properties is to simply define them on the exports object like so:

exports.merge = function(obj, other) {
    var keys = Object.keys(other);
    for (var i = 0, len = keys.length; i < len; ++i) {
        var key = keys[i];
        obj[key] = other[key];
    }
    return obj;
};

Next we will look at utilizing out new module in other libraries.

Requiring Modules

There are four main ways to require a module in node, first is the synchronous method, which simply returns the module's exports, second is the asynchronous method which accepts a callback, third is the asynchronous http method which can load remote modules, and lastly is requiring of shared libraries or "node addons" which we will cover later.

To get started create a second file named ./app.js with the code shown below. The first line require('./utils') fetches the contents of ./utils.js and returns the exports of which we later utilize our merge() method and display the results of our merged object using console.dir().

var utils = require('./utils');

var a = { one: 1 };
var b = { two: 2 };
utils.merge(a, b);
console.dir(a);

Core modules such as the sys which are bundled with node can be required without a path, such as require('sys'), however 3rd-party modules will iterate the require.paths array in search of a module matching the given path. By default require.paths includes ~/.node_libraries, so if ~/.node_libraries/utils.js_ exists we may simply require('utils'), instead of our relative example require('./utils') shown above.

Node also supports the concept of index JavaScript files. To illustrate this example lets create a math module that will provide the math.add(), and math.sub() methods. For organizational purposes we will keep each method in their respective ./math/add.js and ./math/sub.js files. So where does index.js come into play? we can populate ./math/index.js with the code shown below, which is used when require('./math') is invoked, which is conceptually identical to invoking require('./math/index').

module.exports = {
    add: require('./add'),
    sub: require('./sub')
};

The contents of ./math/add.js show us a new technique, here we use module.exports instead of exports. Previously mentioned was the fact that exports is not the only object exposed to the module file when evaluated, we also have access to __dirname, __filename, and module which represents the current module. Here we simply define the module export object to a new object, which happens to be a function.

module.exports = function add(a, b){
    return a + b;
};

This technique is usually only helpful when your module has one aspect that it wishes to expose, be it a single function, constructor, string, etc. Below is an example of how we could provide the Animal constructor:

exports.Animal = function Animal(){};

which can then be utilized as shown:

var Animal = require('./animal').Animal;

if we change our module slightly, we can remove .Animal:

module.exports = function Animal(){};

which can now be used without the property:

var Animal = require('./animal');

Require Paths

We talked about require.paths, the Array utilized by node's module system in order to discover modules. By default node checks the following directories for modules:

The NODE_PATH environment variable is much like PATH, as it allows several paths delimited by the colon (:) character.

Runtime Manipulation

Since require.paths is just an array, we can manipulate it at runtime in order to expose libraries. In our previous example we defined the libraries ./math/{add,sub}.js, in which we would typically require('./math') or require('./math/add') etc. Another approach is to prepend or "unshift" a directory onto require.paths as shown below, after which we can simply require('add') since node will iterate the paths in order to try and locate the module.

require.paths.unshift(__dirname + '/math');

var add = require('add'),
    sub = require('sub');

console.log(add(1,2));
console.log(sub(1,2));

Pseudo Globals

As mentioned above, modules have several pseudo globals available to them, these are as follows:

require()

Although not obvious at first glance, the require() function is actually re-defined for the current module, and calls an internal function loadModule with a reference to the current Module to resolve relative paths and to populate module.parent.

module

When we require() a module, typically we only deal with the module's exports, however the module variable references the current module's Module instance. This is why the following is valid, as we may re-assign the module's exports to any object, even something trivial like a string:

// css.js
module.exports = 'body { background: blue; }';

To obtain this string we would simply require('./css'). The module object also contains these useful properties:

Asynchronous Require

Node provides us with an asynchronous version of require(), aptly named require.async(). Below is the sample example previously shown for our utils module, however non blocking. require.async() accepts a callback of which the first parameter err is null or an instanceof Error, and then the module exports. Passing the error (if there is one) as the first argument is an extremely common idiom in node for async routines.

require.async('./utils', function(err, utils){
    console.dir(utils.merge({ foo: 'bar' }, { bar: 'baz' }));
});

Requiring Over HTTP

Asynchronous requires in node also have the added bonus of allowing module loading via HTTP and HTTPS. To require a module via http all we have to do is pass a valid url as shown in the sass to css compilation example below:

var sassUrl = 'http://github.com/visionmedia/sass.js/raw/master/lib/sass.js',
    sassStr = ''
        + 'body\n'
        + '  a\n'
        + '    :color #eee';

require.async(sassUrl, function(err, sass){
    var str = sass.render(sassStr);
    console.log(str);
});

Outputs:

body a {
  color: #eee;}

Registering Module Compilers

Another cool feature that node provides us is the ability to register compilers for a specific file extension. A good example of this is the CoffeeScript language, which is a ruby/python inspired language compiling to vanilla JavaScript. By using require.registerExtension() we can have node compile CoffeeScript to JavaScript in an automated fashion.

To illustrate its usage, let's create a small (and useless) Extended JavaScript language, or "ejs" for our example which will live at ./compiler/example.ejs, its syntax will look like this:

::min(a, b) a < b ? a : b
::max(a, b) a > b ? a : b

which will be compiled to:

exports.min = function min(a, b) { return a < b ? a : b }
exports.max = function max(a, b) { return a > b ? a : b }

First let's create the module that will actually be doing the ejs to JavaScript compilation. In this example it is located at ./compiler/extended.js, and exports a single method named compile(). This method accepts a string, which is the raw contents of what node is requiring, transformed to vanilla JavaScript via regular expressions.

exports.compile = function(str){
    return str
        .replace(/(\w+)\(/g, '$1 = function $1(')
        .replace(/\)(.+?)\n/g, '){ return $1 }\n')
        .replace(/::/g, 'exports.');
};

Next we have to "register" the extension to assign out compiler. As previously mentioned our compiler lives at ./compiler/extended.js so we are requiring it in, and passing the compile() method to require.registerExtension() which simply expects a function accepting a string, and returning a string of JavaScript.

require.registerExtension('.ejs', require('./compiler/extended').compile);

Now when we require our example, the ".ejs" extension is detected, and will pass the contents through our compiler, and everything works as expected.

var example = require('./compiler/example');
console.dir(example)
console.log(example.min(2, 3));
console.log(example.max(10, 8));

// => { min: [Function], max: [Function] }
// => 2
// => 10

Globals

As we have learnt node's module system discourages the use of globals, however node provides a few important globals for use to utilize. The first and most important is the process global which exposes process manipulation such as signalling, exiting, the process id (pid), and more. Other globals help drive to be similar to other familiar JavaScript environments such as the browser, by providing a console object.

console

The console object contains several methods which are used to output information to stdout or stderr. Let's take a look at what each method does.

console.log()

The most frequently used console method is console.log() simply writing to stdout with a line feed (\n). Currently aliased as console.info().

console.log('wahoo');
// => wahoo

console.log({ foo: 'bar' });
// => [object Object]

console.error()

Identical to console.log(), however writes to stderr. Aliased as console.warn() as well.

console.error('database connection failed');

console.dir()

Utilizes the sys module's inspect() method to pretty-print the object to stdout.

console.dir({ foo: 'bar' });
// => { foo: 'bar' } 

console.assert()

Asserts that the given expression is truthy, or throws an exception.

console.assert(connected, 'Database connection failed');

process

The process object is plastered with goodies, first we will take a look at some properties that provide information about the node process itself.

process.version

The version property contains the node version string, for example "v0.1.103".

process.installPrefix

Exposes the installation prefix, in my case "/usr/local", as node's binary was installed to "/usr/local/bin/node".

process.execPath

Path to the executable itself "/usr/local/bin/node".

process.platform

Exposes a string indicating the platform you are running on, for example "darwin".

process.pid

The process id.

process.cwd()

Returns the current working directory, for example:

cd ~ && node
node> process.cwd()
"/Users/tj"

process.chdir()

Changes the current working directory to the path passed.

process.chdir('/foo');

process.getuid()

Returns the numerical user id of the running process.

process.setuid()

Sets the effective user id for the running process. This method accepts both a numerical id, as well as a string. For example both process.setuid(501), and process.setuid('tj') are valid.

process.getgid()

Returns the numerical group id of the running process.

process.setgid()

Similar to process.setuid() however operates on the group, also accepting a numerical value or string representation. For example process.setgid(20) or process.setgid('www').

process.env

An object containing the user's environment variables, for example:

{ PATH: '/Users/tj/.gem/ruby/1.8/bin:/Users/tj/.nvm/current/bin:/usr/bin:/bin:/usr/sbin:/sbin:/usr/local/bin:/usr/X11/bin'
, PWD: '/Users/tj/ebooks/masteringnode'
, EDITOR: 'mate'
, LANG: 'en_CA.UTF-8'
, SHLVL: '1'
, HOME: '/Users/tj'
, LOGNAME: 'tj'
, DISPLAY: '/tmp/launch-YCkT03/org.x:0'
, _: '/usr/local/bin/node'
, OLDPWD: '/Users/tj'
}

process.argv

When executing a file with the node executable process.argv provides access to the argument vector, the first value being the node executable, second being the filename, and remaining values being the arguments passed.

For example our source file ./src/process/misc.js can be executed by running:

$ node src/process/misc.js foo bar baz

in which we call console.dir(process.argv), outputting the following:

[ 'node'
, '/Users/tj/EBooks/masteringnode/src/process/misc.js'
, 'foo'
, 'bar'
, 'baz'
]

process.exit()

The process.exit() method is synonymous with the C function exit(), in which a exit code > 0 is passed indicating failure, or 0 to indicate success. When invoked the exit event is emitted, allowing a short time for arbitrary processing to occur before process.reallyExit() is called with the given status code.

process.on()

The process itself is an EventEmitter, allowing you to do things like listen for uncaught exceptions, via the uncaughtException event:

process.on('uncaughtException', function(err){
    console.log('got an error: %s', err.message);
    process.exit(1);
});

setTimeout(function(){
    throw new Error('fail');
}, 100);

process.kill()

process.kill() method sends the signal passed to the given pid, defaulting to SIGINT. In our example below we send the SIGTERM signal to the same node process to illustrate signal trapping, after which we output "terminating" and exit. Note that our second timeout of 1000 milliseconds is never reached.

process.on('SIGTERM', function(){
    console.log('terminating');
    process.exit(1);
});

setTimeout(function(){
    console.log('sending SIGTERM to process %d', process.pid);
    process.kill(process.pid, 'SIGTERM');
}, 500);

setTimeout(function(){
    console.log('never called');
}, 1000);

errno

The process object is host of the error numbers, these reference what you would find in C-land, for example process.EPERM represents a permission based error, while process.ENOENT represents a missing file or directory. Typically these are used within bindings to bridge the gap between C++ and JavaScript, however useful for handling exceptions as well:

if (err.errno === process.ENOENT) {
    // Display a 404 "Not Found" page
} else {
    // Display a 500 "Internal Server Error" page
}

Events

The concept of an "event" is crucial to node, and used greatly throughout core and 3rd-party modules. Node's core module events supplies us with a single constructor, EventEmitter.

Emitting Events

Typically an object inherits from EventEmitter, however our small example below illustrates the api. First we create an emitter, after which we can define any number of callbacks using the emitter.on() method which accepts the name of the event, and arbitrary objects passed as data. When emitter.emit() is called we are only required to pass the event name, followed by any number of arguments, in this case the first and last name strings.

var EventEmitter = require('events').EventEmitter;

var emitter = new EventEmitter;

emitter.on('name', function(first, last){
    console.log(first + ', ' + last);
});

emitter.emit('name', 'tj', 'holowaychuk');
emitter.emit('name', 'simon', 'holowaychuk');

Inheriting From EventEmitter

A perhaps more practical use of EventEmitter, and commonly used throughout node is to inherit from it. This means we can leave EventEmitter's prototype untouched, while utilizing its api for our own means of world domination!

To do so we begin by defining the Dog constructor, which of course will bark from time to time, also known as an event. Our Dog constructor accepts a name, followed by EventEmitter.call(this), which invokes the EventEmitter function in context to the given argument. Doing this is essentially the same as a "super" or "parent" call in languages that support classes. This is a crucial step, as it allows EventEmitter to set up the _events property which it utilizes internally to manage callbacks.

var EventEmitter = require('events').EventEmitter;

function Dog(name) {
    this.name = name;
    EventEmitter.call(this);
}

Here we inherit from EventEmitter, so that we may use the methods provided such as EventEmitter#on() and EventEmitter#emit(). If the __proto__ property is throwing you off, no worries! we will be touching on this later.

Dog.prototype.__proto__ = EventEmitter.prototype;

Now that we have our Dog set up, we can create .... simon! When simon barks we can let stdout know by calling console.log() within the callback. The callback it-self is called in context to the object, aka this.

var simon = new Dog('simon');

simon.on('bark', function(){
    console.log(this.name + ' barked');
});

Bark twice a second:

setInterval(function(){
    simon.emit('bark');
}, 500);

Removing Event Listeners

As we have seen event listeners are simply functions which are called when we emit() an event. Although not seen often we can remove these listeners by calling the removeListener(type, callback) method. In the example below we emit the message "foo bar" every 300 milliseconds, which has the callback of console.log(). After 1000 milliseconds we call removeListener() with the same arguments that we passed to on() originally. To compliment this method is removeAllListeners(type) which removes all listeners associated to the given type.

var EventEmitter = require('events').EventEmitter;

var emitter = new EventEmitter;

emitter.on('message', console.log);

setInterval(function(){
    emitter.emit('message', 'foo bar');
}, 300);

setTimeout(function(){
    emitter.removeListener('message', console.log);
}, 1000);

Buffers

To handle binary data, node provides us with the global Buffer object. Buffer instances represent memory allocated independently to that of V8's heap. There are several ways to construct a Buffer instance, and many ways you can manipulate it's data.

The simplest way to construct a Buffer from a string is to simply pass a string as the first argument. As you can see by the log output, we now have a buffer object containing 5 bytes of data represented in hexadecimal.

var hello = new Buffer('Hello');

console.log(hello);
// => <Buffer 48 65 6c 6c 6f>

console.log(hello.toString());
// => "Hello"

By default the encoding is "utf8", however this can be specified by passing as string as the second argument. The ellipsis below for example will be printed to stdout as the '&' character when in "ascii" encoding.

var buf = new Buffer('…');
console.log(buf.toString());
// => …

var buf = new Buffer('…', 'ascii');
console.log(buf.toString());
// => &

An alternative method is to pass an array of integers representing the octet stream, however in this case functionality equivalent.

var hello = new Buffer([0x48, 0x65, 0x6c, 0x6c, 0x6f]);

Buffers can also be created with an integer representing the number of bytes allocated, after which we may call the write() method, providing an optional offset and encoding. As shown below we provide the offset of 2 bytes to our second call to write(), buffering "Hel", and then we continue on to write another two bytes with an offset of 3, completing "Hello".

var buf = new Buffer(5);
buf.write('He');
buf.write('l', 2);
buf.write('lo', 3);
console.log(buf.toString());
// => "Hello"

The .length property of a buffer instance contains the byte length of the stream, opposed to JavaScript strings which will simply return the number of characters. For example the ellipsis character '…' consists of three bytes, however the buffer will respond with the byte length, and not the character length.

var ellipsis = new Buffer('…', 'utf8');

console.log('… string length: %d', '…'.length);
// => … string length: 1

console.log('… byte length: %d', ellipsis.length);
// => … byte length: 3

console.log(ellipsis);
// => <Buffer e2 80 a6>

When dealing with JavaScript strings, we may pass it to the Buffer.byteLength() method to determine it's byte length.

The api is written in such a way that it is String-like, so for example we can work with "slices" of a Buffer by passing offsets to the slice() method:

var chunk = buf.slice(4, 9);
console.log(chunk.toString());
// => "some"

Alternatively when expecting a string we can pass offsets to Buffer#toString():

var buf = new Buffer('just some data');
console.log(buf.toString('ascii', 4, 9));
// => "some"

Streams

Streams are an important concept in node. The stream api is a unified way to handle stream-like data, for example data can be streamed to a file, streamed to a socket to respond to an HTTP request, or a stream can be read-only such as reading from stdin. However since we will be touching on stream specifics in later chapters, for now we will concentrate on the api.

Readable Streams

Readable streams such as an HTTP request inherit from EventEmitter in order to expose incoming data through events. The first of these events is the data event, which is an arbitrary chunk of data passed to the event handler as a Buffer instance.

req.on('data', function(buf){
    // Do something with the Buffer
});

As we know, we can call toString() a buffer to return a string representation of the binary data, however in the case of streams if desired we may call setEncoding() on the stream, after which the data event will emit strings.

req.setEncoding('utf8');
req.on('data', function(str){
    // Do something with the String
});

Another import event is the end event, which represents the ending of data events. For example below we define an HTTP echo server, simply "pumping" the request body data through to the response. So if we POST "hello world", our response will be "hello world".

var http = require('http');

http.createServer(function(req, res){
    res.writeHead(200);
    req.on('data', function(data){
        res.write(data);
    });
    req.on('end', function(){
        res.end();
    });
}).listen(3000);

The sys module actually has a function designed specifically for this "pumping" action, aptly named sys.pump(), which accepts a read stream as the first argument, and write stream as the second.

var http = require('http'),
    sys = require('sys');

http.createServer(function(req, res){
    res.writeHead(200);
    sys.pump(req, res);
}).listen(3000);

File System

...

TCP

...

TCP Servers

...

TCP Clients

...

HTTP

...

HTTP Servers

...

HTTP Clients

...

Connect

Connect is a ...

Express

Express is a ...

Testing

...

Expresso

...

Vows

...

Deployment

...