Sunday, May 15, 2011

A Not Very Short Introduction To Node.js

Node.js is a set of asynchronous libraries, built on top of the Google V8 Javascript Engine. Node is used for server side development in Javascript. Do you feel the rush of the 90's coming through your head. It is not the revival of LiveWire, Node is a different beast. Node is a single threaded process, focused on doing networking right. Right, in this case, means without blocking I/O. All the libraries built for Node use non-blocking I/O. This is a really cool feature, which allows the single thread in Node to serve thousands of request per second. It even lets you run multiple servers in the same thread. Check out the performance characteristics of Nginx and Apache that utilize the same technique.

Concurrency x Requests

The graph for memory usage is even better.

Concurrency x Memory

Read more about it at the Web Faction Blog

OK, so what's the catch? The catch is that all code that does I/O, or anything slow at all, has to be called in an asynchronous style.

// Synchronous 
var result = db.query("select * from T"); 
// Use result 
 
// Asynchronous 
db.query("select * from T", function (result) { 
    // Use result 
}); 

So, all libraries that deal with IO has to be re-implemented with this style of programming. The good news is that even though Node has only been around for a couple of years, there are more than 1800 libraries available. The libraries are of varying quality but the popularity of Node shows good promise to deliver high-quality libraries for anything that you can imagine.

History

Node is definitely not the first of its kind. The non-blocking select() loop, that is at the heart of Node, dates back to 1983.

Twisted appeared in Python (2002) and EventMachine in Ruby (2003).

This year a couple of newcomers appeared.

Goliath, which builds on EventMachine, and uses fibers to allow us to program in an synchronous style even though it is asynchronous under the hood.

And, the Async Framework in .Net, which enhances the compiler with the keywords async and await to allow for very elegant asynchronous programming.

Get Started

This example uses OSX as an example platform, if you use something else you will have to google for instructions.

# Install Node using Homebrew 
$ brew install node
==> Downloading http://nodejs.org/dist/node-v0.4.7.tar.gz
######################################################################## 100.0% 
==> ./configure --prefix=/usr/local/Cellar/node/0.4.7 
==> make install
==> Caveats
Please add /usr/local/lib/node to your NODE_PATH environment variable to have node libraries picked up. 
==> Summary
/usr/local/Cellar/node/0.4.7: 72 files, 7.5M, built in 1.2 minutes

When installed you have access to the node command-line command. When invoked without arguments, it start a REPL.

$ node
> function hello(name) {
... return 'hello ' + name; 
... }
> hello('tapir') 
'hello tapir' 
> 

When invoked with a script it runs the script.

// hello.js 
setTimeout(function() { 
    console.log('Tapir'); 
}, 2000); 
console.log('Hello'); 
 
$ node hello.js
Hello
... 
Tapir

Networking

As I mentioned above, Node is focused on networking. That means it should be easy to write networking code. Here is a simple echo server.

// Echo Server 
var net = require('net'); 
 
var server = net.createServer(function(socket) { 
    socket.on('data', function(data) { 
        console.log(data.toString()); 
        socket.write(data); 
    }); 
}); 
server.listen(4000); 
 

And here is a simple HTTP server.

// HTTP Server 
var http = require('http'); 
var web = http.createServer(function(request, response) { 
  response.writeHead(200, { 
    'Content-Type': 'text/plain' 
  }); 
  response.end('Tapirs are beautiful!\n'); 
}); 
web.listen(4001); 
 

Quite similar. A cool thing is that the servers can be started from the same file and node will, happily, serve both HTTP and echo requests from the same thread without any problems. Let's try them out!

# curl the http service 
$ curl localhost:4001 
Tapirs are beautiful! 
 
# use netcat to send the string to the echo server 
$ echo 'Hello beautiful tapir' | nc localhost 4000 
Hello beautiful tapir
 

Modules

Node comes with a selection of built in modules. Ryan Dahl says that they try to keep the core small, but even so the built-in modules cover a lot of useful functionality.

  • net - contains tcp/ip related networking functionality.
  • http - contains functionality for dealing with the HTTP protocol.
  • util - holds common utility functions, such as log, inherits, pump, ...
  • fs - contains filesystem related functionality, remember that everything should be asynchronous.
  • events - contains the EventEmitter that is used for dealing with events in a consistent way. It is used internally but it can be used externally too.

An example

Here is an example of a simple module.

// module tapir.js 
 
// require another module 
var util = require('util'); 
 
function eat(food) { 
  util.log('eating '+ food); 
} 
 
// export a function 
exports.eat = eat; 
 

As you can see it looks like a normal Javascript file and it even looks like it has global variables. It doesn't. When a module is loaded it is wrapped in code, similar to this.

var module = { exports: {}}; 
(function(module, exports){ 
  // module code from file 
  ... 
})(module, module.exports); 
 

As you can see the code is wrapped in a function and an empty object with an export property is sent into it. This is used by the file to export only the functions that it want to publish.

The require function works in symphony with the module and it returns the exported functions to the caller.

Node Package Manager, npm

To allow simple handling of third-party packages, Node uses npm. It can be installed like this:

$ curl http://npmjs.org/install.sh | sh
... 
 

And used like this:

$ npm install -g express
mime@1.2.1 /usr/local/lib/node_modules/express/node_modules/mime
connect@1.4.0 /usr/local/lib/node_modules/express/node_modules/connect
qs@0.1.0 /usr/local/lib/node_modules/express/node_modules/qs
/usr/local/bin/express -> /usr/local/lib/node_modules/express/bin/express
express@2.3.2 /usr/local/lib/node_modules/express
 

As you can see, installing a module also installs its dependencies. This works because a module can be package with meta-data, like so:

// express/package.json 
{ 
  "name": "express", 
  "description": "Sinatra inspired web development framework", 
  "version": "2.3.2", 
  "author": "TJ Holowaychuk <tj@vision-media.ca>", 
  "contributors": [ 
    { "name": "TJ Holowaychuk", "email": "tj@vision-media.ca" }, 
    { "name": "Guillermo Rauch", "email": "rauchg@gmail.com" } 
  ], 
  "dependencies": { 
    "connect": ">= 1.4.0 < 2.0.0", 
    "mime": ">= 0.0.1", 
    "qs": ">= 0.0.6" 
  }, 
  "keywords": ["framework", "sinatra", "web", "rest", "restful"], 
  "repository": "git://github.com/visionmedia/express", 
  "main": "index", 
  "bin": { "express": "./bin/express" }, 
  "engines": { "node": ">= 0.4.1 < 0.5.0" } 
} 
 

The package.json contains information about who made the module, its dependencies, along with some additional information to enable better searching facilities.

Npm installs the modules from a common respository, which contains more than 1800 modules.

Noteworthy Modules

Express is probably the most used of all third-party modules. It is a Sinatra clone and it is very good, just like Sinatra.

// Create a server 
var app = express.createServer(); 
app.listen(4000); 
 
// Mount the root (/) and redirect to index 
app.get('/', function(req, res) { 
  res.redirect('/index.html'); 
}); 
 
// Handle a post to /quiz 
app.post('/quiz', function(req, res) { 
  res.send(quiz.create().id.toString()); 
}); 

Express uses Connect to handle middleware. Middleware is like Rack but for Node (No wonder that Node is nice to work with when it borrows its ideas from Ruby :)

connect( 
      // Add a logger 
      connect.logger() 
      // Serve static file from the current directory 
    , connect.static(__dirname) 
      // Compile Sass and Coffescript files, on the fly 
    , connect.compiler({enable: ['sass', 'coffeescript']}) 
      // Profile all requests 
    , connect.profiler() 
  ).listen(3000); 

Another popular library is Socket.IO. It handles the usual socket variants, such as WebSocket, Comet, Flash Sockets, etc...

var http = require('http'); 
var io = require('socket.io'); 
 
server = http.createServer(function(req, res){...}); 
server.listen(80); 
 
// socket.io attaches to an existing server 
var socket = io.listen(server); 
socket.on('connection', function(client){ 
  // new client is here! 
  client.on('message', function(){ ... }) 
  client.on('disconnect', function(){ ... }) 
}); 

MySql has a library for Node.

client.query( 
  'SELECT * FROM ' + TEST_TABLE, 
  // Note the callback style 
  function(err, results, fields) { 
    if (err) { throw err; } 
 
    console.log(results); 
    console.log(fields); 
    client.end(); 
  } 
); 

And Mongoose can be used for accessing MongoDB.

// Declare the schema 
var Schema = mongoose.Schema
  , ObjectId = Schema.ObjectId; 
 
var BlogPost = new Schema({ 
    author    : ObjectId
  , title     : String
  , body      : String
  , date      : Date
}); 
 
// Use it 
var BlogPost = mongoose.model('BlogPost'); 
 
// Save 
var post = new BlogPost(); 
post.author = 'Stravinsky'; 
instance.save(function (err) { 
  // 
}); 
 
// Find 
BlogPost.find({}, function (err, docs) { 
  // docs.forEach 
}); 

Templating Engines

Everytime a new platform makes its presence, it brings along a couple of new templating languages and Node is no different. Along with the popular ones from the Ruby world, like Haml and Erb (EJS in Node), comes some new ones like Jade and some browser templating languages like Mustache and jQuery templates. I'll show examples of Jade and Mu (Mustache for Node).

I like Jade, because it is a Javascript dialect of Haml and it seems appropriate to use if I'm using Javascript on the server side.

!!! 5 
html(lang="en") 
  head
    title= pageTitle
    script(type='text/javascript') 
      if (foo) { 
        bar() 
      } 
  body
    h1 Jade - node template engine
    #container 
      - if (youAreUsingJade) 
        p You are amazing
      - else 
        p Get on it! 

I'm not really sure if I like Mustache or not, but I can surely see the value of having a templating language which works both on the server side and in the browser.

<h1>{{header}}</h1> 
{{#bug}}
{{/bug}}
 
{{#items}}
  {{#first}}
    <li><strong>{{name}}</strong></li> 
  {{/first}}
  {{#link}}
    <li><a href="{{url}}">{{name}}</a></li> 
  {{/link}}
{{/items}}
 
{{#empty}}
  <p>The list is empty.</p> 
{{/empty}}

Testing

Node comes with assertions built in, and all testing frameworks build on the Assert module, so it is good to know.

assert.ok(value, [message]); 
assert.equal(actual, expected, [message]) 
assert.notEqual(actual, expected, [message]) 
assert.deepEqual(actual, expected, [message]) 
assert.strictEqual(actual, expected, [message]) 
assert.throws(block, [error], [message]) 
assert.doesNotThrow(block, [error], [message]) 
assert.ifError(value) 
assert.fail(actual, expected, message, operator) 
 
// Example 
// assert.throws(function, regexp) 
assert.throws( 
  function() { throw new Error("Wrong value"); }, 
  /value/ 
); 

Apart from that there are at least 30 different testing frameworks to use. I have chosen to use NodeUnit since I find that it handles asynchronous testing well, and it has a nice UTF-8 output that looks good in the terminal,

// ./test/test-doubled.js 
var doubled = require('../lib/doubled'); 
 
// Exported functions are run by the test runner 
exports['calculate'] = function (test) { 
    test.equal(doubled.calculate(2), 4); 
    test.done(); 
}; 
 
// An asynchronous test 
exports['read a number'] = function (test) { 
    test.expect(1); // Make sure the assertion is run 
 
    var ev = new events.EventEmitter(); 
    process.openStdin = function () { return ev; }; 
    process.exit = test.done; 
 
    console.log = function (str) { 
        test.equal(str, 'Doubled: 24'); 
    }; 
 
    doubled.read(); 
    ev.emit('data', '12'); 
}; 
 

Deployment

There are already a lot of platforms providing Node as a service (PaaS , Platform as a Service). Most of them are using Heroku style deployment by pushing to a Git remote. I'll show three alternatives that all provide free Node hosting.

Joyent (no.de)

Joyent, the employers of Ryan Dahl, give you ssh access so that you can install the modules you need. Deployment is done by pushing to a Git remote.

$ ssh node@my-machine.no.de
$ nmp install express
$ git remote add node node@andersjanmyr.no.de:repo
$ git push node master
Counting objects: 5, done. 
Delta compression using up to 2 threads. 
Compressing objects: 100% (3/3), done. 
Writing objects: 100% (3/3), 321 bytes, done. 
Total 3 (delta 2), reused 0 (delta 0) 
remote: Starting node v0.4.7... 
remote: Successful
To node@andersjanmyr.no.de:repo
  8f59169..c1177b0  master -> master

Nodester

Nodester, gives you a command line tool, nodester, that you use to install modules. Deployment by pushing to a Git remote.

$ nodester npm install express
$ git push nodester master
Counting objects: 5, done. 
Delta compression using up to 2 threads. 
Compressing objects: 100% (3/3), done. 
Writing objects: 100% (3/3), 341 bytes, done. 
Total 3 (delta 2), reused 0 (delta 0) 
remote: Syncing repo with chroot
remote: From /node/hosted_apps/andersjanmyr/1346-7856c14e6a5d92a6b5374ec4772a6da0.git/. 
remote:    38f4e6e..8f59169  master     -> origin/master
remote: Updating 38f4e6e..8f59169
remote: Fast-forward
remote:  Gemfile.lock |   10 ++++------
remote:  1 files changed, 4 insertions(+), 6 deletions(-) 
remote: Checking ./.git/hooks/post-receive
remote: Attempting to restart your app: 1346-7856c14e6a5d92a6b5374ec4772a6da0
remote: App restarted.. 
remote: 
remote: 
remote:     \m/ Nodester out \m/ 
remote: 
remote: 
To ec2-user@nodester.com:/node/hosted_apps/andersjanmyr/1346-7856c14e6a5d92a6b5374ec4772a6da0.git
   38f4e6e..8f59169  master -> master

Cloud Foundry

Cloud Foundry is one of the most interesting platforms in the cloud. It was genius by VM Ware to open source the platform, allowing anyone to set up their own cloud if they wish. If you don't want to setup your own Cloud Foundry Cloud, you can use the service hosted at cloundfoundry.com.

With Cloud Foundry, you install the modules locally and then they are automatically deployed as part of the vmc push. Push in this case does not mean git push, but instead, copy all the files from my local machine to the server.

$ npm install express  # Install locally 
mime@1.2.1 ./node_modules/express/node_modules/mime
connect@1.4.0 ./node_modules/express/node_modules/connect
qs@0.1.0 ./node_modules/express/node_modules/qs
express@2.3.0 ./node_modules/express
 
$ vmc push
Would you like to deploy from the current directory? [Yn]: Y
Application Name: snake
Application Deployed URL: 'snake.cloudfoundry.com'? 
Detected a Node.js Application, is this correct? [Yn]: 
Memory Reservation [Default:64M] (64M, 128M, 256M, 512M, 1G or 2G) 
Creating Application: OK
Would you like to bind any services to 'snake'? [yN]: 
Uploading Application: 
  Checking for available resources: OK
  Packing application: OK
  Uploading (1K): OK
Push Status: OK
Staging Application: OK
Starting Application: ........OK

Tools

There are of course a bunch of tools that come with a new platform, Jake, is a Javascript version of Rake, but I am happy with Rake and I don't see the need to switch. But, there are some tools that I cannot live without when using Node.

Reloaders

If you use the vanilla node command then you have to restart it every time you make a change to a file. That is awfully annoying and there are already a number of solutions to the problem.

# Nodemon watches the files in your directory and reloads them if necessary 
$ npm install nodemon
nodemon@0.3.2 ../node_modules/nodemon
 
$ nodemon server.js 
30 Apr 08:21:23 - [nodemon] running server.js 
... 
# Saving the file 
30 Apr 08:22:01 - [nodemon] restarting due to changes... 
 
# Alternative 
$ npm install supervisor
$ supervisor server.js 
DEBUG: Watching directory '/evented-programming-with-nodejs/.  

Debuggers

Another tool that it is hard to live without is a debugger. Node comes with one built in. It has a gdb flavor to it and it is kind of rough.

$ node debug server.js
debug> run
debugger listening on port 5858 
connecting...ok
break in #<Socket> ./server.js:9 
    debugger; 
 
debug> p data.toString(); 
tapir
 
// Javascript 
var echo = net.createServer(function(socket) { 
  socket.on('data', function(data) { 
      debugger; // <= break into debugger 
      socket.write(data); 
  }); 
}); 

If you want a GUI debugger, it is possible to use the one that comes with Chrome by installing the node-inspector. It is started similarly to the built in debugger, but the --debug is an option instead of a subcommand.

$ node-inspector & 
visit http://0.0.0.0:8080/debug?port=5858 to start debugging
 
$ node --debug server.js debugger listening on port 5858 

After that you can just fire up Chrome on the URL, http://0.0.0.0:8080/debug?port=5858 and you can debug the node process just as if it was running in the browser.

Idioms

Idioms, patterns, techniques, call it what you like. Javascript code is littered with callbacks, and event more so with Node. Here are some tips on how to write good asynchronous code with Node.

Return on Callbacks

It is easy to forget to escape from the function after a callback has been called. An easy way to remedy this problem is to call return before every call to a callback. Even though the value is never used by the caller, it is an easy pattern to recognize and it prevents bugs.

 
function doSomething(response, callback) { 
  doAsyncCall('tapir', function(err, result) { 
    if (err) { 
      // return on the callback 
      return callback(err); 
    } 
    // return on the callback 
    return callback(null, result); 
  }); 
} 

Exceptions in Callbacks

Exceptions that occur in callbacks cannot be handled the way we are used to, since the context is different. The solution to this is to pass along the exception as a parameter to the callback. In Node the convetion is to pass the error as the first parameter into the callback.

function insertIntoTable(row, function(err, data) { 
  if (err) return callback(err); 
  ... 
 
  // Everything is OK 
  return callback(null, 'row inserted'); 
} 

Parallel Execution

If you have multiple tasks that need to be finished before you take some new action, this can be handled with a simple counter. Here is an example of a simple function that starts up a bunch of functions in parallel and waits for all of them to finish before calling the callback.

// Do all in parallel 
function doAll(collection, callback) { 
  var left = collection.length; 
  collection.forEach(function(fun) { 
    fun(function() { 
      if (--left == 0) callback(); 
    }); 
  }); 
}; 
 
// Use it 
var result = []; 
doAll([ 
  function(callback) { 
    setTimeout(function() {result.push(1); callback();}, 2000 )}, 
  function(callback) { 
    setTimeout(function() {result.push(2); callback();}, 3000 )}, 
  function(callback) { 
    setTimeout(function() {result.push(3); callback();}, 1000 )} 
  ], function() { return result; } 
 
// returns [3, 1, 2] 

Sequential Execution

Sometimes the ordering is important. Here is a simple function that makes sure that the calls are executed in sequence. It uses recursion to to make sure that the calls are handled in the correct order. It also uses the Node function process.nextTick() to prevent the stack from getting to large for large collections. Similar results can be obtained with setTimeout() in browser Javascript. It can be seen as a simple trick to achieve tail recursion.

function doInSequence(collection, callback) { 
    var queue = collection.slice(0); // Duplicate 
 
    function iterate() { 
      if (queue.length === 0) return callback(); 
      // Take the first element 
      var fun = queue.splice(0, 1)[0]; 
      fun(function(err) { 
        if (err) throw err; 
        // Call it without building up the stack 
        process.nextTick(iterate); 
      }); 
    } 
    iterate(); 
} 
 
 
var result = []; 
doInSequence([ 
  function(callback) { 
    setTimeout(function() {result.push(1); callback();}, 2000 )}, 
  function(callback) { 
    setTimeout(function() {result.push(2); callback();}, 3000 )}, 
  function(callback) { 
    setTimeout(function() {result.push(3); callback();}, 1000 )} 
  ], function() { return result; }); 
 
// Returns [1, 2, 3] 

Library Support for Asynchronous Programming

If you don't want to write these functions yourself, there are a few libraries that can help you out. I'll show two version that I like.

Fibers

Fibers are also called co-routines. Fibers provide two functions, suspend and resume, which allows us to write code in a synchronous looking style. In the Node version of fibers, node-fibers, suspend and resume are called yield() and run() instead.

require('fibers'); 
var print = require('util').print; 
 
function sleep(ms) { 
    var fiber = Fiber.current; 
    setTimeout(function() { fiber.run(); }, ms); 
    yield(); 
} 
 
Fiber(function() { 
    print('wait... ' + new Date + '\n'); 
    sleep(1000); 
    print('ok... ' + new Date + '\n'); 
}).run(); 
print('back in main\n'); 

Fibers are a very nice way of writing asynchronous code but, in Node, they have one drawback. They are not supported without patching the V8 virtual machine. The patching is done when you install node-fibers and you have to run the command node-fibers instead of node to use it.

The async Library

If you don't want to use the patched version of V8, I can recommend the async library. Async provides around 20 functions that include the usual 'functional' suspects (map, reduce, filter, forEach...) as well as some common patterns for asynchronous flow control (parallel, series, waterfall...). All these functions assume you follow the Node convention of providing a single callback as the last argument of your async function.

async.map(['file1','file2','file3'], fs.stat, function(err, results){ 
    // results is now an array of stats for each file 
}); 
 
async.filter(['file1','file2','file3'], path.exists, function(results){ 
    // results now equals an array of the existing files 
}); 
 
async.parallel([ 
    function(){ ... }, 
    function(){ ... } 
], callback); 
 
async.series([ 
    function(){ ... }, 
    function(){ ... } 
], callback); 

Conclusion

Node is definitely an interesting platform. The possibility to have Javascript running through the whole stack, from the browser all the way down into the database (if you use something like CouchDB or MongoDB) really appeals to me. The easy way to deploy code to multiple, different cloud providers is also a good argument for Node.

22 comments:

  1. Should the second 'req' in:

    app.get('/', function(req, req) {
    res.redirect('/index.html');
    });

    be 'res'?:

    app.get('/', function(req, res) {
    res.redirect('/index.html');
    });

    Great article though!

    ReplyDelete
  2. Yes, you're right, fixed it, thanks!

    ReplyDelete
  3. This comment has been removed by the author.

    ReplyDelete
  4. In the "EXCEPTIONS IN CALLBACKS" section, you have "callback(err, data) { ... }". I think you meant "function", not "callback".

    Great article, I learned a few things about Node I didn't already :)

    ReplyDelete
  5. @Jonathan, thanks! I fixed it.

    ReplyDelete
  6. In the example on sequential execution, I think you meant process.nextTick(iterate) instead of process.nextTick(iterate()).

    ReplyDelete
  7. @Adrian, you're right. Thanks!

    ReplyDelete
  8. Thanks for this post, it was very useful for me. I don't think I 'got' node.js before this, because all the tutorials that I had seen before now stopped shortly after introducing npm and didn't go as far as the whole stack.

    ReplyDelete
  9. @jamuraa, I'm glad it helped you!

    ReplyDelete
  10. the mongoose example isn't async - should put the find() call in the save() function callback. i think

    ReplyDelete
  11. @obowah In this case it is meant as separate examples and not as a sequence of instructions. If it was you are correct!

    ReplyDelete
  12. Best thing so far for me... Truly thanks

    ReplyDelete
  13. Hey Anders,

    This was a really good read. I particularly like the examples at the end of how to write JS that is to execute in series or parallel.

    Node's biggest barrier to entry for non-functional programmers is the shift to working with callbacks and context.

    More clear, practical examples like this will surely demystify it for a lot of people.

    Keep up the good work!

    ReplyDelete
  14. @Andrew, thanks for the feedback, I'm glad you liked it.

    ReplyDelete
  15. Thats really very useful piece of code for me.Thanks

    ReplyDelete
  16. Fantastic post. A section on Node IDE's would be most helpful.

    ReplyDelete
  17. @Anand, I'm glad you liked it. I am not that much into IDEs myself, I use Vim, but to my knowledge the best Javascript IDE is the one that comes with the Jetbrains tools.

    ReplyDelete
  18. Excellent article, very nice overview based on your practical experience that can serve as a roadmap for folks starting out.

    Any chance you'll do a 'ANVSITN.js: Revisited'? You could do a brief update on any new tools you've found that complement or replace the ones you outlined in the original article.

    Either way, thanks for sharing.

    ReplyDelete
  19. @Bullfrog, I'm glad you liked it. I wont have time to do a follow up for a while now. But, I'll keep it in mind.

    ReplyDelete
  20. A great post, One question though, When you say 'parallel execution', it is not really parallel, right? The node event loop still handles only one event or callback at a time.

    ReplyDelete
  21. @sandy You are right, the execution will not be parallel from the point of view of Node. But, if the execution takes place outside of node, the execution may be parallel if there are multiple cores on the executing computer. The callbacks will be handled one at a time by Node.

    ReplyDelete