Skip to main content

Scraping with Puppeteer Scraper

This scraping tutorial will go into the nitty gritty details of extracting data from https://apify.com/store using Puppeteer Scraper (apify/puppeteer-scraper). If you arrived here from the Getting started with Apify scrapers, tutorial, great! You are ready to continue where we left off. If you haven't seen the Getting started yet, check it out, it will help you learn about Apify and scraping in general and set you up for this tutorial, because this one builds on topics and code examples discussed there.

Getting to know our tools

In the Getting started with Apify scrapers tutorial, we've confirmed that the scraper works as expected, so now it's time to add more data to the results.

To do that, we'll be using the Puppeteer library. Puppeteer is a browser automation library that allows you to control a browser using JavaScript. That is, simulate a real human sitting in front of a computer, using a mouse and a keyboard. It gives you almost unlimited possibilities, but you need to learn quite a lot before you'll be able to use all of its features. We'll walk you through some of the basics of Puppeteer, so that you can start using it for some of the most typical scraping tasks, but if you really want to master it, you'll need to visit its documentation and really dive deep into its intricacies.

The purpose of Puppeteer Scraper is to remove some of the difficulty faced when using Puppeteer by wrapping it in a nice, manageable UI. It provides almost all of its features in a format that is much easier to grasp when first trying to scrape using Puppeteer.

Web Scraper differences

At first glance, it may seem like Web Scraper (apify/web-scraper) and Puppeteer Scraper are almost the same. Well, they are. In fact, Web Scraper uses Puppeteer underneath. The difference is the amount of control they give you. Where Web Scraper only gives you access to in-browser JavaScript and the pageFunction is executed in the browser context, Puppeteer Scraper's pageFunction is executed in Node.js context, giving you much more freedom to bend the browser to your will. You're the puppeteer and the browser is your puppet. It's also much easier to work with external APIs, databases or the Apify SDK in the Node.js context. The tradeoff is simple. Power vs simplicity. Web Scraper is simple, Puppeteer Scraper is powerful (and the Apify SDK is super-powerful).

Simply put, Web Scraper's pageFunction is just a single page.evaluate() call.

Now that's out of the way, let's open one of the actor detail pages in the Store, for example the Web Scraper page and use our DevTools-Fu to scrape some data.

If you're wondering why we're using Web Scraper as an example instead of Puppeteer Scraper, it's only because we didn't want to triple the number of screenshots we needed to make. Lazy developers!

Building our Page function

Before we start, let's do a quick recap of the data we chose to scrape:

  1. URL - The URL that goes directly to the actor's detail page.
  2. Unique identifier - Such as apify/web-scraper.
  3. Title - The title visible in the actor's detail page.
  4. Description - The actor's description.
  5. Last modification date - When the actor was last modified.
  6. Number of runs - How many times the actor was run.

$1

We've already scraped numbers 1 and 2 in the Getting started with Apify scrapers tutorial, so let's get to the next one on the list: title.

Title

$1

By using the element selector tool, we find out that the title is there under an <h1> tag, as titles should be. Maybe surprisingly, we find that there are actually two <h1> tags on the detail page. This should get us thinking. Is there any parent element that includes our <h1> tag, but not the other ones? Yes, there is! A <header> element that we can use to select only the heading we're interested in.

Remember that you can press CTRL+F (CMD+F) in the Elements tab of DevTools to open the search bar where you can quickly search for elements using their selectors. And always make sure to use the DevTools to verify your scraping process and assumptions. It's faster than changing the crawler code all the time.

To get the title we just need to find it using a header h1 selector, which selects all <h1> elements that have a <header> ancestor. And as we already know, there's only one.

// Using Puppeteer
async function pageFunction(context) {
const { page } = context;
const title = await page.$eval(
'header h1',
((el) => el.textContent),
);

return {
title,
};
}

The page.$eval function allows you to run a function in the browser, with the selected element as the first argument. Here we use it to extract the text content of a h1 element that's in the page. The return value of the function is automatically passed back to the Node.js context, so we receive an actual string with the element's text.

Description

Getting the actor's description is a little more involved, but still pretty straightforward. We can't just simply search for a <p> tag, because there's a lot of them in the page. We need to narrow our search down a little. Using the DevTools we find that the actor description is nested within the <header> element too, same as the title. Moreover, the actual description is nested inside a <span> tag with a class actor-description.

$1

async function pageFunction(context) {
const { page } = context;
const title = await page.$eval(
'header h1',
((el) => el.textContent),
);
const description = await page.$eval(
'header span.actor-description',
((el) => el.textContent),
);

return {
title,
description,
};
}

Modified date

The DevTools tell us that the modifiedDate can be found in a <time> element.

$1

async function pageFunction(context) {
const { page } = context;
const title = await page.$eval(
'header h1',
((el) => el.textContent),
);
const description = await page.$eval(
'header span.actor-description',
((el) => el.textContent),
);

const modifiedTimestamp = await page.$eval(
'ul.ActorHeader-stats time',
(el) => el.getAttribute('datetime'),
);
const modifiedDate = new Date(Number(modifiedTimestamp));

return {
title,
description,
modifiedDate,
};
}

Similarly to page.$eval, the page.$$eval function runs a function in the browser, only this time, it does not provide you with a single Element as the function's argument, but rather with an Array of Elements. Once again, the return value of the function will be passed back to the Node.js context.

It might look a little too complex at first glance, but let us walk you through it. We find all the <time> elements. Then, we read its datetime attribute, because that's where a unix timestamp is stored as a string.

But we would much rather see a readable date in our results, not a unix timestamp, so we need to convert it. Unfortunately, the new Date() constructor will not accept a string, so we cast the string to a number using the Number() function before actually calling new Date(). Phew!

Run count

And so we're finishing up with the runCount. There's no specific element like <time>, so we need to create a complex selector and then do a transformation on the result.

async function pageFunction(context) {
const { page } = context;
const title = await page.$eval(
'header h1',
((el) => el.textContent),
);
const description = await page.$eval(
'header span.actor-description',
((el) => el.textContent),
);

const modifiedTimestamp = await page.$eval(
'ul.ActorHeader-stats time',
(el) => el.getAttribute('datetime'),
);
const modifiedDate = new Date(Number(modifiedTimestamp));

const runCountText = await page.$eval(
'ul.ActorHeader-stats > li:nth-of-type(3)',
((el) => el.textContent),
);
const runCount = Number(runCountText.match(/[\d,]+/)[0].replace(',', ''));

return {
title,
description,
modifiedDate,
runCount,
};
}

The ul.ActorHeader-stats > li:nth-of-type(3) looks complicated, but it only reads that we're looking for a <ul class="ActorHeader-stats ..."> element and within that element we're looking for the third <li> element. We grab its text, but we're only interested in the number of runs. We parse the number out using a regular expression, but its type is still a string, so we finally convert the result to a number by wrapping it with a Number() call.

The numbers are formatted with commas as thousands separators (e.g. '1,234,567'), so to extract it, we first use regular expression /[\d,]+/ - it will search for consecutive number or comma characters. Then we extract the match via .match(/[\d,]+/)[0] and finally remove all the commas by calling .replace(/,/g, ''). We need to use /,/g with the global modifier to support large numbers with multiple separators, without it we would replace only the very first occurrence.

This will give us a string (e.g. '1234567') that can be converted via Number function.

Wrapping it up

And there we have it! All the data we needed in a single object. For the sake of completeness, let's add the properties we parsed from the URL earlier and we're good to go.

async function pageFunction(context) {
const { page, request } = context;
const { url } = request;

// ...

const uniqueIdentifier = url
.split('/')
.slice(-2)
.join('/');

const title = await page.$eval(
'header h1',
((el) => el.textContent),
);
const description = await page.$eval(
'header span.actor-description',
((el) => el.textContent),
);

const modifiedTimestamp = await page.$eval(
'ul.ActorHeader-stats time',
(el) => el.getAttribute('datetime'),
);
const modifiedDate = new Date(Number(modifiedTimestamp));

const runCountText = await page.$eval(
'ul.ActorHeader-stats > li:nth-of-type(3)',
((el) => el.textContent),
);
const runCount = Number(runCountText.match(/[\d,]+/)[0].replace(',', ''));

return {
url,
uniqueIdentifier,
title,
description,
modifiedDate,
runCount,
};
}

All we need to do now is add this to our pageFunction:

async function pageFunction(context) {
// page is Puppeteer's page
const { request, log, skipLinks, page } = context;

if (request.userData.label === 'START') {
log.info('Store opened!');
// Do some stuff later.
}
if (request.userData.label === 'DETAIL') {
const { url } = request;
log.info(`Scraping ${url}`);
await skipLinks();

// Do some scraping.
const uniqueIdentifier = url
.split('/')
.slice(-2)
.join('/');

// Get attributes in parallel to speed up the process.
const titleP = page.$eval(
'header h1',
(el) => el.textContent,
);
const descriptionP = page.$eval(
'header span.actor-description',
(el) => el.textContent,
);
const modifiedTimestampP = page.$eval(
'ul.ActorHeader-stats time',
(el) => el.getAttribute('datetime'),
);
const runCountTextP = page.$eval(
'ul.ActorHeader-stats > li:nth-of-type(3)',
(el) => el.textContent,
);

const [
title,
description,
modifiedTimestamp,
runCountText,
] = await Promise.all([
titleP,
descriptionP,
modifiedTimestampP,
runCountTextP,
]);

const modifiedDate = new Date(Number(modifiedTimestamp));
const runCount = Number(runCountText.match(/[\d,]+/)[0].replace(',', ''));

return {
url,
uniqueIdentifier,
title,
description,
modifiedDate,
runCount,
};
}
}

You have definitely noticed that we changed up the code a little bit. This is because the back and forth communication between Node.js and browser takes some time and it slows down the scraper. To limit the effect of this, we changed all the functions to start at the same time and only wait for all of them to finish at the end. This is called concurrency or parallelism. Unless the functions need to be executed in a specific order, it's often a good idea to run them concurrently to speed things up.

Test run

As always, try hitting that Save & Run button and visit the Dataset preview of clean items. You should see a nice table of all the attributes correctly scraped. You nailed it!

Pagination

Pagination is just a term that represents "going to the next page of results". You may have noticed that we did not actually scrape all the actors, just the first page of results. That's because to load the rest of the actors, one needs to click the Show more button at the very bottom of the list. This is pagination.

This is a typical form of JavaScript pagination, sometimes called infinite scroll. Other pages may just use links that take you to the next page. If you encounter those, just make a Pseudo URL for those links and they will be automatically enqueued to the request queue. Use a label to let the scraper know what kind of URL it's processing.

Waiting for dynamic content

Before we talk about paginating, we need to have a quick look at dynamic content. Since Apify Store is a JavaScript application (as many, if not most, modern websites are), the button might not exist in the page when the scraper runs the pageFunction.

How is this possible? Because the scraper only waits with executing the pageFunction for the page to load its HTML. If there's additional JavaScript that modifies the DOM afterwards, the pageFunction may execute before this JavaScript had the time to run.

At first, you may think that the scraper is broken, but it just cannot wait for all the JavaScript in the page to finish executing. For a lot of pages, there's always some JavaScript executing or some network requests being made. It would never stop waiting. It is therefore up to you, the programmer, to wait for the elements you need. Fortunately, we have an easy solution.

The context.page.waitFor() function

waitFor() is a function that's available on the Puppeteer page object that's in turn available on the context argument of the pageFunction (as you already know from previous chapters). It helps you with, well, waiting for stuff. It accepts either a number of milliseconds to wait, a selector to await in the page, or a function to execute. It will stop waiting once the time elapses, the selector appears or the provided function returns true.

See page.waitFor() in the Puppeteer documentation.

// Waits for 2 seconds.
await page.waitFor(2000);
// Waits until an element with id "my-id" appears in the page.
await page.waitFor('#my-id');
// Waits until a "myObject" variable appears
// on the window object.
await page.waitFor(() => !!window.myObject);

The selector may never be found and the function might never return true, so the page.waitFor() function also has a timeout. The default is 30 seconds. You can override it by providing an options object as the second parameter, with a timeout property.

await page.waitFor('.bad-class', { timeout: 5000 });

With those tools, you should be able to handle any dynamic content the website throws at you.

How to paginate

With the theory out of the way, this should be pretty easy. The algorithm is a loop:

  1. Wait for the Show more button.
  2. Click it.
  3. Is there another Show more button?
    • Yes? Repeat the above. (loop)
    • No? We're done. We have all the actors.

Waiting for the button

Before we can wait for the button, we need to know its unique selector. A quick look in the DevTools tells us that the button's class is some weird randomly generated string, but fortunately, there's an enclosing <div> with a class of show-more. Great! Our unique selector:

div.show-more > button

Don't forget to confirm our assumption in the DevTools finder tool (CTRL/CMD + F).

$1

Now that we know what to wait for, we just plug it into the waitFor() function.

await page.waitFor('div.show-more > button');

Clicking the button

We have a unique selector for the button and we know that it's already rendered in the page. Clicking it is a piece of cake. We'll use the Puppeteer page again to issue the click. Puppeteer will actually simulate dragging the mouse and making a left mouse click in the element.

await page.click('div.show-more > button');

This will show the next page of actors.

Repeating the process

We've shown two function calls, but how do we make this work together in the pageFunction?

async function pageFunction(context) {

// ...

let timeout; // undefined
const buttonSelector = 'div.show-more > button';
for (;;) {
log.info('Waiting for the "Show more" button.');
try {
// Default timeout first time.
await page.waitFor(buttonSelector, { timeout });
// 2 sec timeout after the first.
timeout = 2000;
} catch (err) {
// Ignore the timeout error.
log.info('Could not find the "Show more button", '
+ 'we\'ve reached the end.');
break;
}
log.info('Clicking the "Show more" button.');
await page.click(buttonSelector);
}

// ...

}

We want to run this until the waitFor() function throws, so that's why we use a while(true) loop. We're also not interested in the error, because we're expecting it, so we just ignore it and print a log message instead.

You might be wondering what's up with the timeout. Well, for the first page load, we want to wait longer, so that all the page's JavaScript has had a chance to execute, but for the other iterations, the JavaScript is already loaded and we're just waiting for the page to re-render so waiting for 2 seconds is enough to confirm that the button is not there. We don't want to stall the scraper for 30 seconds just to make sure that there's no button.

Plugging it into the Page function

We've got the general algorithm ready, so all that's left is to integrate it into our earlier pageFunction. Remember the // Do some stuff later comment? Let's replace it.

async function pageFunction(context) {
const { request, log, skipLinks, page } = context;
if (request.userData.label === 'START') {
log.info('Store opened!');
let timeout; // undefined
const buttonSelector = 'div.show-more > button';
for (;;) {
log.info('Waiting for the "Show more" button.');
try {
// Default timeout first time.
await page.waitFor(buttonSelector, { timeout });
// 2 sec timeout after the first.
timeout = 2000;
} catch (err) {
// Ignore the timeout error.
log.info('Could not find the "Show more button", '
+ 'we\'ve reached the end.');
break;
}
log.info('Clicking the "Show more" button.');
await page.click(buttonSelector);
}
}

if (request.userData.label === 'DETAIL') {
const { url } = request;
log.info(`Scraping ${url}`);
await skipLinks();

// Do some scraping.
const uniqueIdentifier = url
.split('/')
.slice(-2)
.join('/');

// Get attributes in parallel to speed up the process.
const titleP = page.$eval(
'header h1',
(el) => el.textContent,
);
const descriptionP = page.$eval(
'header span.actor-description',
(el) => el.textContent,
);
const modifiedTimestampP = page.$eval(
'ul.ActorHeader-stats time',
(el) => el.getAttribute('datetime'),
);
const runCountTextP = page.$eval(
'ul.ActorHeader-stats > li:nth-of-type(3)',
(el) => el.textContent,
);

const [
title,
description,
modifiedTimestamp,
runCountText,
] = await Promise.all([
titleP,
descriptionP,
modifiedTimestampP,
runCountTextP,
]);

const modifiedDate = new Date(Number(modifiedTimestamp));
const runCount = Number(runCountText.match(/[\d,]+/)[0].replace(',', ''));

return {
url,
uniqueIdentifier,
title,
description,
modifiedDate,
runCount,
};
}
}

That's it! You can now remove the Max pages per run limit, Save & Run your task and watch the scraper paginate through all the actors and then scrape all of their data. After it succeeds, open the Dataset tab again and click on Preview**. You should have a table of all the actor's details in front of you. If you do, great job! You've successfully scraped Apify Store. And if not, no worries, just go through the code examples again, it's probably just some typo.

$1

Downloading the scraped data

You already know the Dataset tab of the run console since this is where we've always previewed our data. Notice the row of data formats such as JSON, CSV, and Excel. Below it are options for viewing and downloading the data. Go ahead and try it.

If you prefer working with an API, you can find the example endpoint under the API tab: Get dataset items.

Clean items

You can view and download your data without modifications, or you can choose to only get clean items. Data that aren't cleaned include a record for each pageFunction invocation, even if you did not return any results. The record also includes hidden fields such as #debug, where you can find a variety of information that can help you with debugging your scrapers.

Clean items, on the other hand, include only the data you returned from the pageFunction. If you're only interested in the data you scraped, this format is what you will be using most of the time.

To control this, open the Advanced options view on the Dataset tab.

Bonus: Making your code neater

You may have noticed that the pageFunction gets quite bulky. To make better sense of your code and have an easier time maintaining or extending your task, feel free to define other functions inside the pageFunction that encapsulate all the different logic. You can, for example, define a function for each of the different pages:

async function pageFunction(context) {
switch (context.request.userData.label) {
case 'START': return handleStart(context);
case 'DETAIL': return handleDetail(context);
default: throw new Error('Unknown request label.');
}

async function handleStart({ log, page }) {
log.info('Store opened!');
let timeout; // undefined
const buttonSelector = 'div.show-more > button';
for (;;) {
log.info('Waiting for the "Show more" button.');
try {
// Default timeout first time.
await page.waitFor(buttonSelector, { timeout });
// 2 sec timeout after the first.
timeout = 2000;
} catch (err) {
// Ignore the timeout error.
log.info('Could not find the "Show more button", '
+ 'we\'ve reached the end.');
break;
}
log.info('Clicking the "Show more" button.');
await page.click(buttonSelector);
}
}

async function handleDetail({
request,
log,
skipLinks,
page,
}) {
const { url } = request;
log.info(`Scraping ${url}`);
await skipLinks();

// Do some scraping.
const uniqueIdentifier = url
.split('/')
.slice(-2)
.join('/');

// Get attributes in parallel to speed up the process.
const titleP = page.$eval(
'header h1',
(el) => el.textContent,
);
const descriptionP = page.$eval(
'header span.actor-description',
(el) => el.textContent,
);
const modifiedTimestampP = page.$eval(
'ul.ActorHeader-stats time',
(el) => el.getAttribute('datetime'),
);
const runCountTextP = page.$eval(
'ul.ActorHeader-stats > li:nth-of-type(3)',
(el) => el.textContent,
);

const [
title,
description,
modifiedTimestamp,
runCountText,
] = await Promise.all([
titleP,
descriptionP,
modifiedTimestampP,
runCountTextP,
]);

const modifiedDate = new Date(Number(modifiedTimestamp));
const runCount = Number(runCountText.match(/[\d,]+/)[0].replace(',', ''));

return {
url,
uniqueIdentifier,
title,
description,
modifiedDate,
runCount,
};
}
}

If you're confused by the functions being declared below their executions, it's called hoisting and it's a feature of JavaScript. It helps you put what matters on top, if you so desire.

Bonus 2: Using jQuery with Puppeteer Scraper

If you're familiar with the jQuery library, you may have looked at the scraping code and thought that it's unnecessarily complicated. That's probably up to everyone to decide on their own, but the good news is, you can easily use jQuery with Puppeteer Scraper too.

Injecting jQuery

To be able to use jQuery, we first need to introduce it to the browser. Fortunately, we have a helper function to do just that: Apify.utils.puppeteer.injectJQuery

Just a friendly warning. Injecting jQuery into a page may break the page itself, if it expects a specific version of jQuery to be available and you override it with an incompatible one. Be careful.

You can either call this function directly in your pageFunction, or you can set up jQuery injection in the Pre goto function in the Input and options section.

async function pageFunction(context) {
const { Apify, page } = context;
await Apify.utils.puppeteer.injectJQuery(page);

// your code ...
}
async function preGotoFunction({ page, Apify }) {
await Apify.utils.puppeteer.injectJQuery(page);
}

The implementations are almost equal in effect. That means that in some cases, you may see performance differences, or one might work while the other does not. Depending on the target website.

Let's try refactoring the Bonus 1 version of the pageFunction to use jQuery.

async function pageFunction(context) {
switch (context.request.userData.label) {
case 'START': return handleStart(context);
case 'DETAIL': return handleDetail(context);
default: throw new Error(`Unknown label: ${context.request.userData.label}`);
}

async function handleStart({ log, page }) {
log.info('Store opened!');
let timeout; // undefined
const buttonSelector = 'div.show-more > button';
for (;;) {
log.info('Waiting for the "Show more" button.');
try {
await page.waitFor(buttonSelector, { timeout });
timeout = 2000;
} catch (err) {
log.info('Could not find the "Show more button", '
+ 'we\'ve reached the end.');
break;
}
log.info('Clicking the "Show more" button.');
await page.click(buttonSelector);
}
}

async function handleDetail(contextInner) {
const {
request,
log,
skipLinks,
page,
Apify,
} = contextInner;

// Inject jQuery
await Apify.utils.puppeteer.injectJQuery(page);

const { url } = request;
log.info(`Scraping ${url}`);
await skipLinks();

// Do some scraping.
const uniqueIdentifier = url
.split('/')
.slice(-2)
.join('/');

// Use jQuery only inside page.evaluate (inside browser)
const results = await page.evaluate(() => {
return {
title: $('header h1').text(),
description: $('header span.actor-description').text(),
modifiedDate: new Date(
Number(
$('ul.ActorHeader-stats time').attr('datetime'),
),
).toISOString(),
runCount: Number(
$('ul.ActorHeader-stats > li:nth-of-type(3)')
.text()
.match(/[\d,]+/)[0]
.replace(/,/g, ''),
),
};
});

return {
url,
uniqueIdentifier,
// Add results from browser to output
...results,
};
}
}

There's an important takeaway from the example code. You can only use jQuery in the browser scope, even though you're injecting it outside of the browser. We're using the page.evaluate() function to run the script in the context of the browser and the return value is passed back to Node.js. Keep this in mind.

Final word

Thank you for reading this whole tutorial! Really! It's important to us that our users have the best information available to them so that they can use Apify easily and effectively. We're glad that you made it all the way here and congratulations on creating your first scraping task. We hope that you liked the tutorial and if there's anything you'd like to ask, join us on Discord!

What's next?

Learn how to scrape a website using Apify's Puppeteer Scraper. Build an actor's page function, extract information from a web page and download your data.