Node.js Patterns for Discerning Developers

Node.js Patterns
For the Discerning Developer
C. Aaron Cois, Ph.D. :: Carnegie Mellon University, SEI

Me
@aaroncois
www.codehenge.net
github.com/cacois
Disclaimer: Though I am an employee of the Software Engineering Institute at Carnegie Mellon University, this work was
not funded by the SEI and does not reflect the work or opinions of the SEI or its customers.

Node.js Basics
• JavaScript
• Asynchronous
• Non-blocking I/O
• Event-driven

Prototype-based Programming
• JavaScript has no classes
• Instead, functions define objects
function Person() {}
var p = new Person();
Image: http://tech2.in.com/features/gaming/five-wacky-gaming-hardware-to-look-forward-to/315742
Prototype

Classless Programming
What do classes do for us?
• Define local scope / namespace
• Allow private attributes / methods
• Encapsulate code
• Organize applications in an object-oriented
way

Prototype-based Programming
function Person(firstname, lastname){
this.firstname = firstname;
this.lastname = lastname;
}
var p = new Person(“Philip”, “Fry”);
What else can do that?

Prototype Inheritance
}
// Create new class
Employee = Person;//Inherit from superclass
Employee.prototype = {
marital_status: 'single',
salute: function() {
return 'My name is ' + this.firstname;
}
}
var p = new Employee (“Philip”, “Fry”);

Watch out!
}
// Create new class
}
}
var p = new Employee (“Philip”, “Fry”);
The ‘new’ is very important!
If you forget, your new object will
have global scope internally

Another option
}
}
}
var p = Object.create(Employee);
p.firstname = 'Philip';
p.lastname = 'Fry';
Works, but you can’t initialize
attributes in constructor

Anti-Pattern: JavaScript Imports
• Spread code around files
• Link libraries
• No way to maintain private local
scope/state/namespace
• Leads to:
– Name collisions
– Unnecessary access

Pattern: Modules
• An elegant way of encapsulating and
reusing code
• Adapted from YUI, a few years before
Node.js
• Takes advantage of the anonymous
closure features of JavaScript
Image: http://wallpapersus.com/

Modules in the Wild
var http = require('http'),
io = require('socket.io'),
_ = require('underscore');
If you’ve programmed in Node, this looks
familiar

Anatomy of a module
var privateVal = 'I am Private!';
module.exports = {
answer: 42,
add: function(x, y) {
return x + y;
}
}
mymodule.js

Usage
mod = require('./mymodule');
console.log('The answer: '+ mod.answer);
var sum = mod.add(4,5);
console.log('Sum: ' + sum);

Modules are used everywhere
// User model
var mongoose = require('mongoose')
, Schema = mongoose.Schema;
var userSchema = new Schema({
name: {type: String, required: true},
email: {type: String, required: true},
githubid: String,
twitterid: String,
dateCreated: {type: Date, default: Date.now}
});
userSchema.methods.validPassword = function validPass(pass) {
// validate password…
}
module.exports = mongoose.model('User', userSchema);

My config files? Modules.
var config = require('config.js');
console.log('Configured user is: ' + config.user);
module.exports = {
user: 'maurice.moss'
}
config.js
app.js

Asynchronous Programming
• Node is entirely asynchronous
• You have to think a bit differently
• Failure to understand the event loop and
I/O model can lead to anti-patterns

Event Loop
Node.js
Event Loop
Node app

Event Loop
Node.js
Event Loop
Node apps pass async
tasks to the event loop,
along with a callback
(function, callback)
Node app

Event Loop
Node.js
Event Loop
The event loop efficiently
manages a thread pool and
executes tasks efficiently…
Thread
1
Thread
2
Thread
n
…
Task 1
Task 2
Task 3
Task 4
Return 1
Callback1()
…and executes each callback as
tasks complete
Node app

Async I/O
The following tasks should be done
asynchronously, using the event loop:
• I/O operations
• Heavy computation
• Anything requiring blocking

Your Node app is single-threaded

Anti-pattern: Synchronous Code
for (var i = 0; i < 100000; i++){
// Do anything
}
Your app only has one thread, so:
…will bring your app to a grinding halt

But why would you do that?
Good question.
But in other languages (Python), you may do this:
for file in files:
f = open(file, ‘r’)
print f.readline()

The Node.js equivalent is:
Based on examples from: https://github.com/nodebits/distilled-patterns/
var fs = require('fs');
for (var i = 0; i < files.length; i++){
data = fs.readFileSync(files[i]);
console.log(data);
}
…and it will cause severe performance
problems

Pattern: Async I/O
fs = require('fs');
fs.readFile('f1.txt','utf8',function(err,data){
if (err) {
// handle error
}
console.log(data);
});

Async I/O
fs = require('fs');
fs.readFile('f1.txt','utf8',function(err,data){
if (err) {
// handle error
}
console.log(data);
});
Anonymous, inline callback

Async I/O
fs = require('fs');
fs.readFile('f1.txt','utf8',
function(err,data){
if (err) {
// handle error
}
console.log(data);
}
);
Equivalent
syntax

Callback Hell
When working with callbacks, nesting can
get quite out of hand…

Callback Hell
var db = require('somedatabaseprovider');
//get recent posts
http.get('/recentposts', function(req, res) {
// open database connection
db.openConnection('host', creds,function(err, conn){
res.param['posts'].forEach(post) {
conn.query('select * from users where
id='+post['user'],function(err,users){
conn.close();
res.send(users[0]);
});
}
});
});

Callback Hell
//get recent posts
conn.close();
res.send(users[0]);
});
}
});
});
Get recent posts from web
service API

Callback Hell
//get recent posts
conn.close();
res.send(users[0]);
});
}
});
});
Open connection to DB

Callback Hell
//get recent posts
conn.close();
res.send(users[0]);
});
}
});
});
Get user from DB for each
post

Callback Hell
//get recent posts
conn.close();
res.send(users[0]);
});
}
});
});
Return users

Anti-Pattern: Callback Hell
fs.readdir(source, function(err, files) {
if (err) {
console.log('Error finding files: ' + err)
} else {
files.forEach(function(filename, fileIndex) {
console.log(filename)
gm(source + filename).size(function(err, values) {
if (err) {
console.log('Error identifying file size: ' + err)
} else {
console.log(filename + ' : ' + values)
aspect = (values.width / values.height)
widths.forEach(function(width, widthIndex) {
height = Math.round(width / aspect)
console.log('resizing ' + filename + 'to ' + height + 'x' + height)
this.resize(width, height).write(destination+'w’+width+'_’+filename, function(err){
if (err) console.log('Error writing file: ' + err)
})
}.bind(this))
}
})
})
}
})
http://callbackhell.com/

Solutions
• Separate anonymous callback functions
(cosmetic)
• Async.js
• Promises
• Generators

Pattern:
Separate Callbacks
fs = require('fs');
callback = function(err,data){
if (err) {
// handle error
}
console.log(data);
}
fs.readFile('f1.txt','utf8',callback);

Can Turn This
http.get('/recentposts', function(req, res){
db.openConnection('host', creds, function(err,
conn){
conn.query('select * from users where id=' +
post['user'],function(err,results){
conn.close();
res.send(results[0]);
});
}
});
});

Into This
http.get('/recentposts', afterRecentPosts);
function afterRecentPosts(req, res) {
db.openConnection('host', creds, function(err, conn) {
afterDBConnected(res, conn);
});
}
function afterDBConnected(err, conn) {
conn.query('select * from users where id='+post['user'],afterQuery);
}
}
function afterQuery(err, results) {
conn.close();
res.send(results[0]);
}

This is really a Control Flow issue

Pattern: Async.js
Async.js provides common patterns for
async code control flow
https://github.com/caolan/async
Also provides some common functional
programming paradigms

Serial/Parallel Functions
• Sometimes you have linear serial/parallel
computations to run, without branching
callback growth
Function
1
Function
2
Function
3
Function
4
Function
1
Function
2
Function
3
Function
4

async.parallel([
function(){ ... },
function(){ ... }
], callback);
async.series([
function(){ ... },
function(){ ... }
]);

async.parallel([
function(){ ... },
function(){ ... }
], callback);
async.series([
function(){ ... },
function(){ ... }
], callback);
Single Callback

Waterfall
Async.waterfall([
function(callback){ ... },
function(input,callback){ ... },
function(input,callback){ ... },
], callback);

Map
var arr = ['file1','file2','file3'];
async.map(arr, fs.stat, function(err, results){
// results is an array of stats for each file
console.log('File stats: ' +
JSON.stringify(results));
});

Filter
var arr = ['file1','file2','file3'];
async.filter(arr, fs.exists, function(results){
// results is a list of the existing files
console.log('Existing files: ' + results);
});

Carefree
for (var i = 0; i < 10000; i++) {
fs.readFileSync(filename);
}
With synchronous code, you can loop as much
as you want:
The file is opened once each iteration.
This works, but is slow and defeats the point of
Node.

Synchronous Doesn’t Scale
What if we want to scale to 10,000+
concurrent users?
File I/O becomes
the bottleneck
Users get in a
long line

Async to the Rescue
function onRead(err, file) {
if (err) throw err;
}
for (var i = 0; i < 10000; i++) {
fs.readFile(filename, onRead);
}
What happens if I do this asyncronously?

Ruh Roh
The event loop is fast
This will open the file 10,000 times at once
This is unnecessary…and on most systems,
you will run out of file descriptors!

Pattern:
The Request Batch
• One solution is to batch requests
• Piggyback on existing requests for the
same file
• Each file then only has one open request
at a time, regardless of requesting clients

// Batching wrapper for fs.readFile()
var requestBatches = {};
function batchedReadFile(filename, callback) {
// Is there already a batch for this file?
if (filename in requestBatches) {
// if so, push callback into batch
requestBatches[filename].push(callback);
return;
}
// If not, start a new request
var callbacks = requestBatches[filename] = [callback];
// Flush out the batch on complete
delete requestBatches[filename];
for(var i = 0;i < callbacks.length; i++) {
// execute callback, passing arguments along
callbacks[i](err, file);
}
}
}

return;
}
}
}
}
Is this file already being read?

return;
}
}
}
}
If not, start a new file read
operation

return;
}
}
}
}
When read finished, return to
all requests

Usage
//Request the resource 10,000 times at once
for (var i = 0; i < 10000; i++) {
batchedReadFile(file, onComplete);
}
function onComplete(err, file) {
if (err) throw err;
else console.log('File contents: ' + file);
}

Pattern:
The Request Batch
This pattern is effective on many read-type
operations, not just file reads
Example: also good for web service API
calls

Shortcomings
Batching requests is great for high request
spikes
Often, you are more likely to see steady
requests for the same resource
This begs for a caching solution

Pattern:
Request Cache
Let’s try a simple cache
Persist the result forever and check for new
requests for same resource

// Caching wrapper around fs.readFile()
var requestCache = {};
function cachingReadFile(filename, callback) {
//Do we have resource in cache?
if (filename in requestCache) {
var value = requestCache[filename];
// Async behavior: delay result till next tick
process.nextTick(function () { callback(null, value); });
return;
}
// Cache the result if there is no error
function onRead(err, contents) {
if (!err) requestCache[filename] = contents;
callback(err, contents);
}
}

Usage
// Request the file 10,000 times in series
// Note: for serial requests we need to iterate
// with callbacks, rather than within a loop
var its = 10000;
cachingReadFile(file, next);
function next(err, contents) {
console.log('File contents: ' + contents);
if (!(its--)) return;
cachingReadFile(file, next);
}

Almost There!
You’ll notice two issues with the Request
Cache as presented:
• Concurrent requests are an issue again
• Cache invalidation not handled
Let’s combine cache and batch strategies:

// Wrapper for both caching and batching of requests
var requestBatches = {}, requestCache = {};
function readFile(filename, callback) {
if (filename in requestCache) { // Do we have resource in cache?
var value = requestCache[filename];
// Delay result till next tick to act async
process.nextTick(function () { callback(null, value); });
return;
}
if (filename in requestBatches) {// Else, does file have a batch?
return;
}
// If neither, create new batch and request
// Cache the result and flush batch
if (!err) requestCache[filename] = file;
for (var i=0;i<callbacks.length;i++) { callbacks[i](err, file); }
}
}

scale-fs
I wrote a module for scalable File I/O
https://www.npmjs.org/package/scale-fs
Usage:
var fs = require(’scale-fs');
for (var i = 0; i < 10000; i++) {
fs.readFile(filename);
}

Final Thoughts
Most anti-patterns in Node.js come from:
• Sketchy JavaScript heritage
• Inexperience with Asynchronous Thinking
Remember, let the Event Loop do the heavy
lifting!

Thanks
Code samples from this talk at:
https://github.com/cacois/node-
patterns-discerning

Disclaimer
Though I am an employee of the Software
Engineering Institute at Carnegie Mellon
University, this wok was not funded by the
SEI and does not reflect the work or
opinions of the SEI or its customers.

Let’s chat
@aaroncois
www.codehenge.net
github.com/cacois

Node.js
Event Loop
The event loop efficiently
manages a thread pool and
executes tasks efficiently…
Thread
1
Thread
2
Thread
n
…
Task 1
Task 2
Task 3
Task 4
Return 1
Callback1()
…and executes each callback as
tasks complete
Node.js
app
Node apps pass async
tasks to the event loop,
along with a callback
(function, callback)
1 2
3

Node.js Patterns for Discerning Developers

More Related Content

What's hot

Viewers also liked

Similar to Node.js Patterns for Discerning Developers

More from cacois

Recently uploaded

Node.js Patterns for Discerning Developers

Editor's Notes