This course aims to give you an understanding of back-end development and in it you will learn how to build an HTTP server in node.js and integrating it with a MongoDB database. The course will focus a lot on JavaScript and give you an understanding for how the language works so you will be able to solve issues in your code more readily.
You can find working code samples for each chapter along with some exercises in the course Github repository.
In order to follow along with this course you will want to have the following installed:- node.js v16 or higher
npm- An editor with proper JavaScript support (VSCode, Sublime, Vim, Emacs, …)
- Git
You can check that you have all necessary command line tools by running the following commands in your terminal:
node --version
npm --version
git --versionIf you don’t like to go through the hassle of setting up an account you can either 1) install MongoDB locally following the instructions in link above or 2) use the Docker image if you are comfortable with Docker. Personally I installed it locally.
MongoDB Compass offers a graphical interface to your database and often comes bundled when installing it locally. If you don’t have it installed you can follow these instructions.
As long as you have Homebrew installed it’s very easy to install both MongoDB and MongoDB Compass on your computer:brew tap mongodb/brew
brew install [email protected]
brew install mongodb-compassInstalling the mongodb-community package should give you two commands:
mongoshfor interacting with a MongoDB database through the terminal.mongodfor starting a MongoDB database process (the final d stands for daemon and means a long-running process).
When installing through Homebrew it seems that running mongod by itself doesn’t work. So instead you need to start MongoDB as a Homebrew service, which means that MongoDB will be running in the background. You do this by running:
brew services start mongodb/brew/mongodb-communityIf everything works, you should be able to run mongosh and be taken to a MongoDB prompt.
If this is not working you can list your brew services by running brew services list and see the status of mongodb-community. you can show detailed information about it by running:
brew services info mongodb-communitywinget:
winget install mongodb.server
winget install mongodb.compass.fullInstalling mongodb.server installs mongodb as a service that automatically starts when you start your computer.
You can change it with the services application (or sc if you rather like using a terminal).
The MongoDB Compass application is easily launched by pressing Start and typing mongodb.
To connect Compass to your server, you simply press “Connect” (no connection string required).
- The same language across the stack (front-end and back-end)
- Simplify the transition to full-stack for front-end developers
- The asynchronous nature of JavaScript makes it great for easily building high performance HTTP servers
In summary: familiarity and performance
Node.JS was created 12 years ago by creating a system interface to Chrome’s V8 JavaScript engine. That means that Node.JS is running the same version of JavaScript as Chrome and other Chromium-based browsers such as Microsoft Edge, Brave etc. Which V8 version Node uses dictates what JavaScript features it supports. If you are curious you can check which exact version of V8 your node.js installation is using by running the following command in a terminal:
node -p process.versions.v8- There’s no browser environment, that is you do not have access to the global
windowanddocumentobjects. - You instead have the global variable
globalto refer to the global scope. - You have the global variable
processfor reading environment variables etc. - You have access to built-in modules for doing things like reading and writing files and networking etc.
hello-node and move into it. Now create a file called index.js and write the following piece of code:
console.log("Hello node! \(>0<)/")Now you can run your program with the command node index.js and you should see Hello node! \(>0<)/ printed to your terminal. We have run JavaScript outside of the browser and successfully printed text, hooray!
fs to play around with files.
import fs from "fs";
const databases = [
{ name: 'MongoDB', type: 'document' },
{ name: 'PostgreSQL', type: 'relational' },
{ name: 'Neo4j', type: 'graph' },
{ name: 'Redis', type: 'in-memory' },
];
fs.writeFileSync("test.txt", JSON.stringify(databases, null, 2));
const contents = fs.readFileSync("test.txt").toString();
console.log(`File contents: ${contents}`);The difference between the module systems lies not only in cosmetics but also semantics, ES6 modules being a lot more restrictive in when and how you can import modules. Given the flexibility of CommonJS modules we might never see a full transition to ES6 modules.
Let’s create new module with a function that randomly picks an element from a list. And let’s call it fromindex.js.
export default function randomElement(xs) {
const randomIndex = Math.floor((Math.random() * 10) % xs.length)
return xs[randomIndex];
}import fs from "fs";
import randomElement from './random-element.js';
const databases = [
{ name: 'MongoDB', type: 'document' },
{ name: 'PostgreSQL', type: 'relational' },
{ name: 'Neo4j', type: 'graph' },
{ name: 'Redis', type: 'in-memory' },
];
// ...
const randomDatabase = randomElement(databases);
console.log('Got database:', randomDatabase);global variable.
let count = 0;
global.ourGlobalFunction = (source) => {
count++;
console.log(`Call count: ${count} (from ${source})`);
};import fs from "fs";
import randomElement from './random-element.js';
import './modifying-global-scope.js';
global.ourGlobalFunction(import.meta.url);
// Since the scope is global we can even call it directly as well
ourGlobalFunction(import.meta.url);
// ...Exercise Try calling ourGlobalFunction from randomElement.js. Try both within the function and outside. Is it working? If not, why not?
Finally, please do not modify ~global~ in /real/ code. it breaks encapsulation and makes it more difficult to understand what’s going on.
Another thing we can do in node.js that we can’t do in the browser is to get information about the current environment especially things like environment variables.We can access environment variables via the process variable:
console.log('USER:', process.env.USER); // Prints your username
console.log('MY_VARIABLE', process.env.MY_VARIABLE); // Prints undefined- An API is a set of exposed methods for interacting with a program or package.
- When you write a JavaScript module and export functions to interact with it you are designing an API.
- When you are interacting with a third-party package, for example
express, you are using its API. - Designing an API allows you to create a layer of abstraction which hides implementation details and simplifies using your service or package.
Often when we say API we actually mean an HTTP API to be specific, that is an API which is used over the internet using HTTP.
Express is by far the most popular NPM package for creating HTTP APIs in node.js and has been around almost as long as the language itself. Start by creating a new directory calledhello-express and initialize it using npm init (also don’t forget to update package.json if you want to use ES6 modules). Now let’s install Express:
npm install expressNow let’s create our first API by creating a new file called index.js in the project root directory and write the following code:
import express from 'express';
const app = express();
app.get('/hello', (req, res) => {
res.send('Hello there!').end();
});
const PORT = 8080;
app.listen(PORT, () => {
console.log(`Server running at http://localhost:${PORT}`)
});There is a lot to unpack here…
- We begin by creating an instance of an Express app.
- Then we register a handler on the
/helloendpoint which will respond with Hello there!. - Lastly we start a server listening on port 8080.
node index.js. The first thing you will notice is that your program never quits: you see the message Server running at http://localhost:8080 but you don’t get a new prompt. This is because your program is running a server which is meant to serve responses to requests from clients and your program needs to be kept alive and running to be able to do that.
A client is whatever uses, or consumes, the API served by your server and can be anything from a web browser, website, another server or a command-line tool etc. For now, let’s use our browser as the client and access the URL printed out by the program: http://localhost:8080. You should see an error message saying something like Cannot GET /.
This means that we tried to GET something at the endpoint /. We’ll get more into what GET actually means later when we talk about HTTP, but for now let’s try changing the endpoint and go to http://localhost:8080/hello instead. Now you should instead see the expected message Hello there!.
So what went wrong the first time? There are four pieces of information needed to interact with a server:
- The protocol the server expects (
http) - The machine the server is running on (our machine
localhostor127.0.0.1if we use its IP address). This is also called the host. - The port the server is listening on (
8080) - The endpoint we want to consume (
/hello)
A server only responds on the port it is listening on and only handles requests on endpoints which have been registered on it. When not specifying an endpoint, the browser will pick the default one which is / and since we never registered a handler for that endpoint the request failed. You can think of endpoints as file paths on your own computer.
// ...
app.get('/another-page', (req, res) => {
res.send('Another page!').end();
});
// ...If we add another endpoint and try to access it in the browser: http://localhost:8080/another-page we get the same error message as we did before.
The reason is that the server process is already running and changes made to the code will not be reflected until it is restarted. You can stop the server by selecting the terminal where it is running and press Ctrl-c (that means pressing the Ctrl button and the c key at the same time). This will terminate your server and get you back to the terminal prompt.
If you now run node index.js again you will be able to access http://localhost:8080/another-page.
npm install --save-dev nodemonNow we add a convenience script called dev in package.json to make it easy to use it:
{
// ...
"scripts": {
"dev": "nodemon index.js",
"test": "echo \"Error: no test specified\" && exit 1"
}
// ...
}By running npm run dev your server will be started up and nodemon will watch your files for changes and restart the server when necessary.
There is another tool I highly recommend you install and that is prettier. This tool formats your code automatically and you should be able to make your editor run it every time you save. Here is a VSCode plugin and here is one for Emacs.
Let’s make our new endpoint do something more interesting: let’s see what happens if we serve a string which looks like HTML.// ...
app.get("/another-page", (req, res) => {
res
.send(
`
<html>
<head>
<style>
body {
margin: 32px;
background: hotpink;
color: darkgreen;
font-family: arial;
}
</style>
</head>
<body>
<h1>Our beautiful page</h1>
<marquee>We're serving a string which is rendered as a web page!</marquee>
</body>
</html>
`
)
.end();
});
// ...And we can see that our browser interprets it as HTML! The secret is that the browser interprets EVERYTHING as HTML, so we shouldn’t be surprised.
While it’s pretty cool that we can serve web pages as plain strings, what you usually want to do is to serve HTML files instead. We move our HTML to a file which we can call beautiful-page.html.
<html>
<head>
<style>
body {
margin: 32px;
background: hotpink;
color: darkgreen;
font-family: arial;
}
</style>
</head>
<body>
<h1>Our beautiful page</h1>
<marquee>We're serving a string which is rendered as a web page!</marquee>
</body>
</html>And we change our handler to read that file and serve its contents.
import express from "express";
import fs from "fs";
// ...
app.get("/another-page", (req, res) => {
const contents = fs.readFileSync("beautiful-page.html").toString();
res.send(contents).end();
});
// ...The page should load like before but the code looks a lot nicer without the inline HTML.
A website made up from files like this is called a static website. This is how the whole web worked through-out the 90s and the beginning of the 00s until Single Page Applications (SPAs) became a thing. In this course we will assume you will write your website as a SPA (in React), so we won’t be serving static pages. In addition, the above code is highly inefficient and is just for illustrative purposes. First we are reading the HTML file for every request even though the contents doesn’t change, this will lead to a lot of file system access which impacts performance. Second, we send the page a single string all at once which also impacts performance. If you are interested in how to serve static web pages using Express you can have a look at this documentation.
MongoDB is a document (NoSQL) database and has a few important characteristics which makes it a suitable as a first database:- Flexible data schemas.
- Intuitive data models (basically looks like JSON).
- Simple yet powerful query language.
MongoDB, and document databases in general, are often used in MVPs and prototypes when you are still exploring and have yet to decide on the data models to use. This does not mean however that they are not production-ready: document databases are among the most scalable databases out there and allow for efficient horizontal scaling (this means running multiple connected instances in a database cluster).
While we discuss MongoDB specifically in this section many of the concepts are applicable to other document databases as well such as CouchDB and elasticsearch, though the terminology might be a bit different.
A MongoDB system consists of one or several databases, which each can have one or multiple collections and each collection contains documents. Documents are the central concept of a document database, naturally.
The main selling point of MongoDB compared to relational (SQL) databases (MySQL, Postgres, …) is the flexibility. In relational databases you have to define how your data is structured and the relationship between different kinds of data models. The structure of your data is called its schema or sometimes its data model and defines the properties it has and what data types these properties have. Here’s a made-up example of how a schema might look like:PersonSchema = {
"id": "string",
"name": "string",
"age": "integer",
"weight": "float",
}In a relational database a schema like the above ensures for instance that a Person’s name is a string and that its weight is a float. If you would try to store a Person with a string weight the operation would fail. This makes it difficult for bad and ill-structured data to enter the database.
In a document database schemas still exist, but they are just suggestions and are meant to improve performance when querying the data. As you most likely will see when you start to work with MongoDB yourself is that it will happily accept a float as the name, or even allow you to insert documents with a completely different set of properties in the same collection.
This flexibility is something to be mindful of and I recommend using MongoDB Compass to explore your data set from time to time to ensure that it looks like you expect it to.
Operations are ways of interacting with your database in the terms of data, the most general operations being:- Create data
- Read data
- Update data
- Delete data
These are often called CRUD operations for short.
The following sections describes what the common CRUD operations are in MongoDB and examples assume that you have a connected db database instance available:
const client = mongodb.MongoClient('mongodb://localhost:27017');
await client.connect();
const db = client.db('mongodb-intro');The code assumes that you have the mongodb package in scope and you are in an async context where you can use async.
await db.collection('languages').insertOne({
name: 'JavaScript',
family: 'C',
year: 1995
});const languages = [{
name: 'Haskell',
family: 'ML',
year: 1990
}, {
name: 'Rust',
family: 'ML',
year: 2010,
}, {
name: 'Java',
family: 'C',
year: 1995,
}, {
name: 'Common Lisp',
family: 'Lisp',
year: 1984,
}];
await db.collection('languages').insertMany(languages)const cursor = db.collection("languages").find({});
const results = cursor.toArray();
console.log(results);The find operation can potentially return a huge amount of documents depending on the size of your data set so it does not return the results directly, but a cursor pointing to the results. This allows you to either do further processing or return a subset of the results. You can get all of the matching results by calling its toArray() method as in the example above.
The simplest filter apart from an empty one is to match on properties exactly. In this example we are picking out allow of the programming languages related to C in our data set.
const filter = {
family: 'C' // Matching property exactly
}
const results = await db.collection('languages').find(filter).toArray();
console.log(results);The findOne operation will return the first document it finds which matches the filter.
const filter = {
type: 'ML'
}
const result = await db.collection('languages').findOne(filter);For more advanced filtering we use query operators, you can quickly identify them since they start with a $. Some common ones are $gte (greater-than-or-equal), $lte (less-than-or-equal) and $regex for matching against a regular expression.
const filter = {
name: { $regex: /Java/ }
}
const results = await db.collection('languages').find(filter).toArray();
console.log(results);We can also combine multiple operators to express more complex queries; the next example finds all of the languages created in the 90s.
const filter = {
year: {
$gte: 1990,
$lte: 1999
}
};You can sort your results with the cursor’s sort method by passing it an object containing the property you want to sort on and 1 for ascending results (low to high) or -1 for ascending (high to low).
const cursor = await db.collection('languages').find({});
const results = cursor.sort({ year: 1 }).toArray();
console.log(results);find or findOne methods with deleteMany or deleteOne, the methods use the same kind of filters.
await db.collection('languages').deleteOne({
name: 'Java'
});updateOne or updateMany to update multiple documents at the same time and these methods take two arguments: a filter object to specify which documents will be affected, and an update object defining the modification.
const filter = { name: 'JavaScript'};
const modification = { $set: { year: 2022 } };
await db.collection('languages').updateOne(filter, modification)
The browser has a main thread which is responsible for not only executing JavaScript but also rendering the HTML and CSS as well as handling user input like clicks and scrolling. If JavaScript allowed for synchronous HTTP requests the whole browser tab would stall while waiting for the response to come back. This is of course something we want to avoid at all costs and thus JavaScript does these sort of I/O operations asynchronously by default for us.
All asynchronous operations are put on the Event Queue so as to not block the main thread, and from there the operations are executed in other threads. It is important to understand that once a piece of code has gone on the event queue, there’s really no way of fully “getting back” to the main thread.
Even though the explanation above focused on JavaScript in the browser node.js works in very much the same way with a main thread and an event queue to handle asynchronous requests. It’s important to know understand that the node.js process will stay alive for as long as there is something on its event queue.
In the beginning there was the callback.fs.readdir(directory, (err, files) => {
if (err) {
console.log('Error finding files: ', err)
} else {
files.forEach((filename) => {
const filePath = `${directory}/${filename}`;
fs.stat(filePath, (err, fileStats) => {
if (err) {
console.log('Error checking file status: ', err)
} else {
if (fileStats.isFile())
console.log('Found file:', filePath);
}
})
})
}
})This was the only means of handling asynchronous operations and as you can see from the example above, it quickly lead to unreadable and nested spaghetti code. Adding proper error handling made things worse which meant you would often skip error handling for the sake of readability, leading to code that was broken and error prone. Promises was created to solve this issue and remove the nested spaghetti mess of the code above.
A Promise can be thought of as a promise of a future value, that is, we do not have the value yet, but we capture the promise of it in a variable that we can use in our code. Before we revisit the file listing example above, let’s look at the connection between callbacks and promises usingsetTimeout.
console.log('Before setTimeout')
setTimeout(() => {
console.log('Inside setTimeout')
}, 1000);
console.log('After setTimeout')The output of the above example should be:
Before setTimeout
After setTimeout
# After one second:
Inside setTimeoutThe node.js process is kept alive until the callback passed to setTimeout is finished and prints its output. The callback is put on the event queue and hence executed asynchronously. We see that since the last console.log statement in the code, which is executed by the main thread, is printed out before the one in setTimeout.
We can turn the setTimeout call into a Promise by using the Promise constructor which takes a callback function with two arguments a resolve function and a reject function. For now we’ll only focus on the resolve function which resolves the promise.
console.log('Before setTimeout')
new Promise((resolve, reject) => {
setTimeout(() => {
console.log('Inside setTimeout')
resolve();
}, 1000);
});
console.log('After setTimeout')The output should be the same as in the previous example. But why would we want to wrap an asynchronous call in a Promise like this? Because it allows us to untangle the nested horizontal callback hell pyramid of doom that we saw above. The weapon we have at our disposal is the then method of the promise:
// We wrap our functions in functions that return promises
const readdir = (dir) => {
return new Promise((resolve, reject) => {
fs.readdir(dir, (err, files) => {
if (err) {
// We can handle errors by passing them to the reject callback
reject('Error finding files: ', err)
} else {
// and pass on values to the next Promise in the chain by
// using the resolve callback
resolve(files);
}
})
})
}
const fileStats = (filePath) => {
return new Promise((resolve, reject) => {
fs.stat(filePath, (err, fileStats) => {
if (err) {
reject('Error checking file status: ', err)
} else {
resolve(fileStats);
}
})
})
}
readdir(directory)
.then((directoryContents) => {
directoryContents.forEach((name) => {
const filePath = `${directory}/${name}`;
fileStats(filePath).then((stats) => {
if (stats.isFile()) {
console.log('Found file:', filePath);
}
}).catch((err) => {
console.log(err);
});
})
}).catch((err) => {
console.log(err);
})If we ignore the boilerplate code for creating our promises, the code looks a little bit neater now. It’s still nested but we’ve been able to extract the error handling so it doesn’t pollute our core logic as much. We can do better however. The helper function Promise.all will allow us to pass a list of promises and get back the results of all promises in as a list:
readdir(directory)
.then((directoryContents) => directoryContents.map((name) => `${directory}/${name}`))
.then((filePaths) => {
// Collect all of the stats calls into a list of promises
const promises = filePaths.map((filePath) =>
fileStats(filePath).then((stats) => ({
filePath,
isFile: stats.isFile()
}))
);
// Use Promise.all to make this promise resolve when all promises in the list are resolved.
return Promise.all(promises);
})
.then((maybeFiles) => maybeFiles.filter((f) => f.isFile))
.then((files) => files.map((f) => f.filePath))
.then((paths) => console.log(paths))
// Now we only need one (1!) catch
.catch((err) => console.log(err))Now instead of having a horizontal callback pyramid of doom, we have a pillar of promises. Many of the operations we do in the then clauses are not asynchronous themselves, but once we enter Promiseland there’s no escape. You can only use the result of a promise in its then clause, and since then also returns a Promise we can’t get out.
The main benefit of this approach is that we reduce the nesting and our error handling is significantly simplified. However, we can do better. Enter async/await.
The final improvement we can do is to replace our pillar of promises with awaits, but first let’s look at the relationship betweenasync functions, await and promises.
You can think of await being similar to calling then on a promise, the main difference is that await can only be used in an async context.
// We can't do this:
await fetch('https://http.cat/500')
const foo = async () {
// This is OK since we're in an async arrow function context
await fetch('https://http.cat/200')
}
async function bar() {
// This is OK since we're in an async function context
await fetch('https://http.cat/200')
}In fact, if you log the unawaited return value of an async function you will see that it actually returns a Promise:
async function willReturnAPromise() {
return 42
}
console.log(willReturnAPromise()) // Prints: Promise { 42 }This means that you can await Promise:s and then async functions:
async function foo() {
return 10;
}
async function () {
const ten = await Promise.resolve(10);
console.log(ten);
await foo().then((result) => console.log(result + 32));
}Understanding this, we are now equipped to clean-up our file listing example above and make it really appear synchronous.
async function listFilesInDirectory(directory) {
try {
const directoryContents = await readdir(directory);
const filePaths = directoryContents.map((name) => `${directory}/${name}`);
const promises = filePaths.map((filePath) =>
fileStats(filePath).then((stats) => ({
filePath,
isFile: stats.isFile()
}))
);
const maybeFiles = await Promise.all(promises);
const files = maybeFiles.filter((f) => f.isFile);
const paths = files.map((f) => f.filePath);
console.log(paths);
} catch (err) {
console.log(err);
}
}Another consequence of functional programming is that you clearly separate the data and operations on the data.
Although most languages can be used in a more or less functional style, there are certain languages that are considered functional like Haskell, Elixir, Clojure and Elm.
Object oriented programming (OOP) is extremely common and you encounter this style a lot. Below are the four principles of OOP (taken from Object-oriented programming in C#):
- Abstraction
- Modeling the relevant attributes and interactions of entities as classes to define an abstract representation of a system.
- Encapsulation
- Hiding the internal state and functionality of an object and only allowing access through a public set of functions.
- Inheritance
- Ability to create new abstractions based on existing abstractions.
- Polymorphism
- Ability to implement inherited properties or methods in different ways across multiple abstractions.
The goal of OOP is to create modular and flexible code. Some common languages that more or less strictly follows OOP are Java, C# and Smalltalk.
A common critique of OOP is that it can lead to extremely complex code with an excess of abstraction layers which add very little to in form of functionality. This kind of over-engineering is not specific to OOP though and can also be seen in code-bases following a functional programming approach. The tendency to rely on mutation does however often lead to code that is hard to debug.
If the topic of programming languages interests you I can’t recommend Dan Grossman’s course on programming languages enough. It is by far the best programming course I have ever taken and will give you a thorough understanding of different programming paradigms.
We are going to have a look at the power of functional programming by implementing some of the JavaScript array methods ourselves, namely [].map(), [].filter() and [].reduce().
const map = (operation, list) => {
let results = [];
for (element of list) {
results.push(operation(element));
}
return results;
};Since we will apply an operation to each element in the list we need our operations to take only 1 argument. We can do that by turning a multi-argument function into a so-called higher-order function that only take one argument and returns a function which takes the next argument. This is called currying.
function add(x) {
return function (y) {
return x * y;
};
}
// The above can be shortened to this using arrow functions:
const subtract = (x) => (y) => x - y;
// Now we can "configure" our operation according to our needs. Let's
// create an operation that takes 1 argument and adds 5 to it:
const addFive = add(5);The addFive function is a function that takes one number as its argument and adds 5 to it, so we can readily pass it to our map function:
const result = map(addFive, [1, 2, 3, 4]);
console.log(result) // [6, 7, 8, 9]true or false when given an element of the list and filter will return a list of all elements for which the function returned true. This kind of function is usually called a predicate.
const filter = (predicate, list) => {
let results = [];
for (element of list) {
if (predicate(element)) {
results.push(element);
}
}
return results;
};
// Our predicate function which returns true if the passed in value is even
const isEven = (x) => x % 2 == 0;
const result = filter(isEven, [1, 2, 3, 4]);
console.log(result) // [2, 4]reduceList:
const reduceList = (operation, list) => {
let results = [];
for (element of list) {
// Since we don't know what the operation will do to the
// accumulated results list (append or not append), we need
// to be able to pass it to the operation function:
results = operation(results, element);
}
return results;
};reduceList is able to handle both mapping and filtering at the expense of the operation functions becoming more specific in that they need to update the results list.
let result = reduceList((results, x) => {
return [...results, addFive(x)];
}, [1, 2, 3, 4]);
console.log(result); // [6, 7, 8, 9]
result = reduceList((results, x) => {
if (isEven(x)) {
return [...results, x];
}
return results;
}, [1, 2, 3, 4]);
console.log(result); // [2, 4]const reduce = (operation, list, accumulator) => {
for (element of list) {
accumulator = operation(accumulator, element);
}
return accumulator;
};We can use reduce to not only work with lists, but we can actually use it do calculations as well if we pass in a number as the accumulator.
const result = reduce((sum, x) => sum + x, [1, 2, 3, 4], 0);
console.log(result); // 10Passing an object as the accumulator allows us to create more complex aggregations from for instance a list of objects. The name reduce might start to make sense now, we are taking a list of something and reducing it to a value of some sort.
// Or build an object from a list
const dogs = [
{ name: "Fido", breed: "Chihuahua" },
{ name: "Woofmeister", breed: "Poodle" },
{ name: "Puglifer", breed: "Pug" },
{ name: "Poddle McPoodleface", breed: "Poodle" },
];
const result = reduce((accumulator, dog) => ({
...accumulator,
[dog.breed]: (accumulator[dog.breed] || 0) + 1,
}), dogs, {});
console.log(result) // { 'Chihuahua': 1, 'Pug': 1, 'Poodle': 2 }Now we have implemented almost a full version of the JavaScript reduce array method [].reduce. I hope this illustrates the power of the concept of higher-order functions. Higher-order functions together with referential transparency (the fact that a function should always return the same result when passed the same arguments) are at the heart of functional programming and grasping the potential of them allows for very powerful abstractions.
For this section we will work on the WoofWoof API we developed in class and use it to illustrate various API-related topics such as REST, middlewares, query parameters and CORS. We begin by looking at what endpoints we have thus far:| HTTP Method | Endpoint | Action |
|---|---|---|
| POST | /dogs | Add a dog |
| GET | /dogs | Retrieve all dogs |
| GET | /dogs/:id | Retrieve a specific dog |
| PATCH | /dogs/:id | Update a specific dog |
| DELETE | /dogs/:id | Delete a specific dog |
We have route handlers for adding dogs, getting all dogs or a specific dogs, we can also update a dog entry or delete it. In fact, this API is a good example of a simple REST API.
REST stands for Representational State Transfer and is a convention for how to write an HTTP API. One of the most important aspect of a REST API is that it has a common interface regardless of the type of data resource it is exposing. A REST API is also meant to be client (/frontend/) agnostic and able to support different clients and use-cases at once. Another important aspect is that a REST API needs to be stateless, which means that the server should not hold any client-specific state and that everything needed to fulfill a query must be provided in the request (given the same input we should get identical output). REST APIs make sense when fronting a database where we have one or more collections of things and need to add, find, update and/or delete entries (see the section on MongoDB).The interface of our WoofWoof API is an example of a typical REST API: you operate on the whole dogs collection using the /dogs endpoint and we use different HTTP methods (or verbs) to either retrieve dogs (GET) or add a dog to the collection (POST). If we want to do something to a specific dog entry we append its ID: /dogs/:dogId, and use HTTP methods to specify the action: retrieve (GET), update (PATCH) and delete (DELETE).
So what are the benefits of using a REST API? First of all, having a uniform interface to an API greatly simplifies integrating it to frontends or other APIs. If your API is a unique snowflake it will most likely require more work for the people using your API. Another benefit is the ability to cache responses since all information required is provided by the request.
Currently our server is very quiet. Apart from an initial greeting message our server doesn’t give any information at all when we interact with it. What we really want is for the server to be a bit more talkative and for instance tell us when a new request comes in etc. This will help us when debugging and potentially provide us some statistics about our server.We could do this by adding console.log statements in our route handlers, but that quickly becomes repetitive and worse could be easy to forget to do. A better alternative is to use what is called middleware.
A middleware is a function which is applied to the request (or response) before the request reaches the route handlers. Think of it as something that comes in the in-between (or in the middle of) a request and the route handling of that request.
A middleware function can be used for anything from setting headers (as the cors middleware which you can read more about in the section on connecting a front-end client), logging, authenticating a user and much, much more. The express.json middleware for instance, which we used previously, reads and parses the request body as JSON if the Content-Type header of the request is set to application/json.
request, the response and a next function.
// ...
const helloMiddleware = (request, response, next) => {
console.log("hello");
next();
}
// ...
app.use(helloMiddleware);
// ...The first two arguments are pretty self-explanatory but the third argument is a bit more interesting. By calling the next function you signal that the middleware is done and the processing can move on to either the next middleware in the chain or the route handlers. Now we are touching on something important: in Express all middlewares and the matching route handlers are applied in the order they are used in the file. This means that it can be important where you put your app.use(middlewareFn) call.
The above middleware is pretty useless, but by picking out useful information from the request we can make the log statement more useful. Let’s rename it to requestLogger and pick out some useful information such as the HTTP verb and endpoint so we can track which endpoints are called.
const requestLogger = (request, response, next) => {
const timestamp = new Date().toISOString();
const method = request.method;
const url = request.url;
console.log(`${timestamp} ${method} ${url}`);
next();
}Now we should see something like 2022-04-01T16:52:45.392Z GET /dogs in our terminal console. If we want we can dig deeper and add information about the duration of the request, but that requires us to know when the request ended. Luckily we can hook into the request’s end event:
const requestLogger = (request, response, next) => {
const requestStartTimeMs = Date.now(); // Added current time in milliseconds for reference
const timestamp = new Date().toISOString();
const method = request.method;
const url = request.url;
console.log(`${timestamp} ${method} ${url}`);
request.on("end", () => {
const duration = Date.now() - requestStartTimeMs;
console.log(`${new Date().toISOString()} ${method} ${url} ${duration}ms`)
});
next();
}Each request will now generate at least two log statements, the latter one giving the duration of the request in milliseconds.
2022-04-01T16:52:45.372Z GET /dogs
2022-04-01T16:52:45.392Z GET /dogs 20msFor now it’s easy to enough to correlate the requests, but if you would have multiple clients making requests at once it becomes very difficult to see which request had what duration. This is where something like a requestId, a unique ID for every request, becomes useful. As an exercise write a middleware which adds a property requestId to the request which you then read in the requestLogger.
While writing logging middleware by hand like this is fun there are already great packages which help with logging. I recommend looking into packages such as Winston or Morgan and see what capabilities they offer. Having good logging for an API is extremely important and will potentially save you hours of debugging.
For now our endpoint/dogs retrieves ALL dogs in the database. Any filtering or sorting is left up to the client application to take care of. In order to alleviate the workload of the client we want to of course allow them to specify what kind of dogs they want to list from the get-go. The common way to solve this in REST is to introduce query parameters.
In a URL query parameters appear after a ? and are key value pairs separated by &. In our case it might look something like: http://localhost:4649/dogs?title=breed=chihuahua&containsPuppy=true where we ask for every Chihuahua puppy picture in our data set.
Of course this functionality does not implement itself, let’s see what we need to do in our /dogs route handler to support this.
app.get("/dogs", async (request, response) => {
const query = request.query;
let filter = {};
if (query.containsPuppy) {
filter.containsPuppy = query.containsPuppy === "true";
}
if (query.breed) {
// Case-insensitive substring matching using regular expressions
filter.breed = { $regex: new RegExp(query.breed, "i") };
}
const dogs = await collection.find(filter).toArray();
response.json(dogs);
});Express gives us the query parameters as request.query which is an object with string keys and string values. This last point is important since if we expect the value of a query parameter to be a number we need to turn it into a number ourselves.
On lines 4-11 we construct a filter object which we can use to query our database on line 13. Building the filter in this way allows us to handle each query parameter differently should we need to and it also makes it explicit (to us developers at least) what query parameters we expect and handle.
Line 6 illustrates the fact that the query parameter values are always strings and not parsed into their intended types implicitly, so we have to match query.containsPuppy against the string "true" and not the boolean value true.
Using query parameters allows the caller of your API to configure their request even if it is a GET and they are used prolifically on the web. Whenever you use Google or some other service and you in the URL see a question mark followed by a number of ampersand separated key value pairs, you’re seeing query parameters in use.
Assuming we have our server running on http://localhost:4649 I can retrieve the full list of dogs by using the fetch function which is built-in to the browser. Let’s try it out in a simple HTML page:
<script>
fetch('http://localhost:4649/dogs', {
headers: {
'Content-Type': 'application/json'
}
}).then((response) => {
return response.json()
}).then((dogs) => {
console.log(dogs);
}).catch((err) => {
console.error('Something went wrong:', err);
});
</script>We need to pass in the Content-Type header since that will make our API parse the JSON payload for us (this is what the app.use(express.json()) line on the server does).
When loading the page you might see the following error in the console, this is called a Cross-Origin Resource Sharing, or CORS, issue. The reason we are seeing this error is that we are trying to load resources from an origin (domain) that is different from the one our client is running on: http://localhost:4649 vs. http://localhost:3000.
In order to fix this issue we need to tell our server to tell the browser that it is OK to communicate with http://localhost:4649 and fortunately there is an NPM package to help us do that.
import cors from "cors";
// ...
app.use(
cors({
// We allow requests from our frontend, if you want to allow any client
// you can use "*" instead.
origin: "http://localhost:3000",
})
);
// ...If you refresh the page after making this change you should be able to make the request and receive the results. Congratulations you have successfully connected a front-end client to your server.
For the brief you are to write an HTTP API and some front-end client which uses your API. There are a few ways to go about doing this and in this section I am going to list some common approaches.First, it helps if you consider your API and your client as completely separate services, imagine that you are two teams: one working on the front-end and one on the back-end.
The first question you might face is one git repository or two repositories? What do I mean by this? In the single repository case you have both the back-end API and the front-end client in the same directory as in the listing below.my-service
├── .git
├── client
│ ├── node_modules
│ ├── src
│ │ └── index.js
│ ├── package-lock.json
│ └── package.json
├── server
│ ├── node_modules
│ ├── src
│ │ └── index.js
│ ├── package-lock.json
│ └── package.json
└── README.mdThe benefit of single repository is that it’s very easy to make correlated changes to the client and the server at the same time and in the same commit. This ensures that the current client and the current server are compatible, if you make sure to update both at the same time that is. This practice is also referred to as a mono-repo and its an approach that companies such as Google use and from what I have heard over 90% of their code resides in a giant mono-repo.
So what are the drawbacks of a mono-repo? Mainly it becomes a bit more complicated to deploy your application since you will need to support multiple means of deployment within the same repository. For instance Netlify or Vercel for the client and Heroku or Digital Ocean for the server. You need to make sure to restrict your deployments to sub-folders of your repository. There’s also a question of how flexible you can be with regards to deployments: will you be forced to use separate branches for deployment or will you be able to use git tags.
Another drawback of having a mono-repo is that git was not meant to handle huge code bases, so if your service would scale to the size of Google you will have to implement workarounds and/or even modify git to allow your team to work efficiently. You most likely will not reach that stage during this course.
The other common pattern is to split everything up into separate git repositories, one for your client and one for your server. This makes it easier to treat them as separate services and can simplify the deployment process. You will be able to chose the most appropriate production deployment procedure for each service. However, there is a bit of overhead that comes from handling multiple git repositories and their respective configurations etc.
The multi-repository approach has been dominant for the last five years or so in my experience, but lately there has been a hype around mono-repos, so it might be that this will become more dominant in the future. In summary, both approaches are widely used so you can’t really go wrong here.
Here there are SO many options that it’s almost a joke. The most common tools for writing front-ends in the industry currently though is by far React, followed by Vue and then Angular. While I do love using vanilla JavaScript, when making a more complex web page there are just so many benefits to using a framework like React, if you learn the basics you will be able to focus more on implementing the functionality than trying to figure out how to implement the functionality since you need to reinvent the wheel at every turn. MDN’s page on JavaScript frameworks gives a good overview of some common ones and what they bring to the table.I recommend you use the framework you are most comfortable with. If you don’t know any framework, I recommend you use this opportunity to get yourself acquainted with one. If you are not sure which one to choose go with React because it is basically an industry standard at the time of writing (2022).
If you feel that the mainstream JavaScript frameworks are boring and not for you, then you’ve come to the right place. There are so many other interesting and fascinating ways to build front-ends that do not follow down the much trodden path. Maybe the frameworks listed here are not for this project right now, but they might provide a source of inspiration later on your journey as a developer. The list here focuses on frameworks that choose to do things differently and some of them are not even in JavaScript (though they do transpile down to JavaScript in most cases). I do recommend you at one point give yourself the opportunity to try some of these out, not only because it’s a lot of fun, but also because they challenge the conventional way of doing things and provide a great learning experience.- Elm
- You think TypeScript does not provide you the amount of type-checking you need for your project? Say hi to Elm, a Haskell-inspired purely functional programming language made for the sole purpose of building web apps. The on-boarding experience is great and the error messages are AMAZING.
- ClojureScript
- ClojureScript (as well as its parent language Clojure) is a LISP which is a family of languages that has a few common traits: very simple syntax, lots of parentheses and the treatment of code as data. LISPs are really great for writing code that writes code. However, this is not the only thing ClojureScript does well, it is tailored towards functional programming with immutability at the center and interesting ways of handling concurrency. Getting used to LISPs is a real mind-bender and will really help broaden your horizons as a developer.
- PureScript
- You tried Elm and felt it did a bit too much hand-holding? Then PureScript is for you. It’s even more inspired by Haskell and comes with full support for esoteric features such as type classes, higher kinded types and higher rank polymorphism (not sure what the last one means, but sounds awesome). Prepare yourself for a steep learning curve, but you will most likely find yourself at the cusp of modern programming language theory after getting proficient in this language.
- Cycle.js
- What if instead of writing a web app that allows for human interactions, you flipped the whole premise on its head and instead put the stream of interactions at the center of your web app? Cycle.js is built around the concept of reactive streams. Everything are streams of events which you use to render you application and add interactivity. In the beginning it feels extremely backwards, but once you “get it” it starts to feel like a really natural way of thinking about apps and services.
- Yew
- Ready to tick off two buzzwords with a single framework? With Yew you write your code in Rust and it compiles down to WebAssembly, the new web language on the block. Even though you write your page in Rust, the overall experience is similar to React and Elm.
- Reason ML
- From what I heard the draft for this project was what lead to the development of React at Facebook. It is very similar to Elm, but uses a language called OCaml as its foundation.
- First comes the interface layer which is the only thing the client needs to know about.
- Next comes a controller layer which encapsulates the main server logic, for instance picking out data from the request and preparing it for the next layer, and decides the response to be sent back to the client.
- Next comes the service layer which encapsulates your database or third-party API etc.
There are multiple benefits to following this approach and separate each layer into its own module (file). The main benefit is separation of concerns: if you manage to confine your database logic and operations into a separate module it will allow you to switch the database with hardly any modifications necessary outside of the database module. It also allows you to have your own data models within your server that does not necessarily correspond to the database models. This too simplifies the process of switching out your database without any major modifications outside of the database module.
GeneralExpress
MongoDB