Upgrading to v1

Summary

After 3.5 years of rapid development and a lot of breaking changes and deprecations, here comes the result - Apify SDK v1. There were two goals for this release. Stability and adding support for more browsers - Firefox and Webkit (Safari).

The SDK has grown quite popular over the years, powering thousands of web scraping and automation projects. We think our developers deserve a stable environment to work in and by releasing SDK v1, we commit to only make breaking changes once a year, with a new major release.

We added support for more browsers by replacing PuppeteerPool with browser-pool. A new library that we created specifically for this purpose. It builds on the ideas from PuppeteerPool and extends them to support Playwright. Playwright is a browser automation library similar to Puppeteer. It works with all well known browsers and uses almost the same interface as Puppeteer, while adding useful features and simplifying common tasks. Don't worry, you can still use Puppeteer with the new BrowserPool.

A large breaking change is that neither puppeteer nor playwright are bundled with the SDK v1. To make the choice of a library easier and installs faster, users will have to install the selected modules and versions themselves. This allows us to add support for even more libraries in the future.

Thanks to the addition of Playwright we now have a PlaywrightCrawler. It is very similar to PuppeteerCrawler and you can pick the one you prefer. It also means we needed to make some interface changes. The launchPuppeteerFunction option of PuppeteerCrawler is gone and launchPuppeteerOptions were replaced by launchContext. We also moved things around in the handlePageFunction arguments. See the migration guide for more detailed explanation and migration examples.

What's in store for SDK v2? We want to split the SDK into smaller libraries, so that everyone can install only the things they need. We plan a TypeScript migration to make crawler development faster and safer. Finally, we will take a good look at the interface of the whole SDK and update it to improve the developer experience. Bug fixes and scraping features will of course keep landing in versions 1.X as well.

Migration Guide

There are a lot of breaking changes in the v1.0.0 release, but we're confident that updating your code will be a matter of minutes. Below, you'll find examples how to do it and also short tutorials how to use many of the new features.

Many of the new features are made with power users in mind, so don't worry if something looks complicated. You don't need to use it.

Installation

Previous versions of the SDK bundled the puppeteer package, so you did not have to install it. SDK v1 supports also playwright and we don't want to force users to install both. To install SDK v1 with Puppeteer (same as previous versions), run:

npm install apify puppeteer

To install SDK v1 with Playwright run:

npm install apify playwright

While we tried to add the most important functionality in the initial release, you may find that there are still some utilities or options that are only supported by Puppeteer and not Playwright.

Running on Apify Platform

If you want to make use of Playwright on the Apify Platform, you need to use a Docker image that supports Playwright. We've created them for you, so head over to the new Docker image guide and pick the one that best suits your needs.

Note that your package.json MUST include puppeteer and/or playwright as dependencies. If you don't list them, the libraries will be uninstalled from your node_modules folder when you build your actors.

Handler arguments are now Crawling Context

Previously, arguments of user provided handler functions were provided in separate objects. This made it difficult to track values across function invocations.

const handlePageFunction = async (args1) => {
    args1.hasOwnProperty('proxyInfo') // true
}

const handleFailedRequestFunction = async (args2) => {
    args2.hasOwnProperty('proxyInfo') // false
}

args1 === args2 // false

This happened because a new arguments object was created for each function. With SDK v1 we now have a single object called Crawling Context.

const handlePageFunction = async (crawlingContext1) => {
    crawlingContext1.hasOwnProperty('proxyInfo') // true
}

const handleFailedRequestFunction = async (crawlingContext2) => {
    crawlingContext2.hasOwnProperty('proxyInfo') // true
}

// All contexts are the same object.
crawlingContext1 === crawlingContext2 // true

Map of crawling contexts and their IDs

Now that all the objects are the same, we can keep track of all running crawling contexts. We can do that by working with the new id property of crawlingContext This is useful when you need cross-context access.

let masterContextId;
const handlePageFunction = async ({ id, page, request, crawler }) => {
    if (request.userData.masterPage) {
        masterContextId = id;
        // Prepare the master page.
    } else {
        const masterContext = crawler.crawlingContexts.get(masterContextId);
        const masterPage = masterContext.page;
        const masterRequest = masterContext.request;
        // Now we can manipulate the master data from another handlePageFunction.
    }
}

autoscaledPool was moved under crawlingContext.crawler

To prevent bloat and to make access to certain key objects easier, we exposed a crawler property on the handle page arguments.

const handePageFunction = async ({ request, page, crawler }) => {
    await crawler.requestQueue.addRequest({ url: 'https://example.com' });
    await crawler.autoscaledPool.pause();
}

This also means that some shorthands like puppeteerPool or autoscaledPool were no longer necessary.

const handePageFunction = async (crawlingContext) => {
    crawlingContext.autoscaledPool // does NOT exist anymore
    crawlingContext.crawler.autoscaledPool // <= this is correct usage
}

Replacement of PuppeteerPool with BrowserPool

BrowserPool was created to extend PuppeteerPool with the ability to manage other browser automation libraries. The API is similar, but not the same.

Access to running BrowserPool

Only PuppeteerCrawler and PlaywrightCrawler use BrowserPool. You can access it on the crawler object.

const crawler = new Apify.PlaywrightCrawler({
    handlePageFunction: async ({ page, crawler }) => {
        crawler.browserPool // <-----
    }
});

crawler.browserPool // <-----

Pages now have IDs

And they're equal to crawlingContext.id which gives you access to full crawlingContext in hooks. See Lifecycle hooks below.

const pageId = browserPool.getPageId

Configuration and lifecycle hooks

The most important addition with BrowserPool are the lifecycle hooks. You can access them via browserPoolOptions in both crawlers. A full list of browserPoolOptions can be found in browser-pool readme.

const crawler = new Apify.PuppeteerCrawler({
    browserPoolOptions: {
        retireBrowserAfterPageCount: 10,
        preLaunchHooks: [
            async (pageId, launchContext) => {
                const { request } = crawler.crawlingContexts.get(pageId);
                if (request.userData.useHeadful === true) {
                    launchContext.launchOptions.headless = false;
                }
            }
        ]
    }
})

Introduction of BrowserController

BrowserController is a class of browser-pool that's responsible for browser management. Its purpose is to provide a single API for working with both Puppeteer and Playwright browsers. It works automatically in the background, but if you ever wanted to close a browser properly, you should use a browserController to do it. You can find it in the handle page arguments.

const handlePageFunction = async ({ page, browserController }) => {
    // Wrong usage. Could backfire because it bypasses BrowserPool.
    await page.browser().close();

    // Correct usage. Allows graceful shutdown.
    await browserController.close();

    const cookies = [/* some cookie objects */];
    // Wrong usage. Will only work in Puppeteer and not Playwright.
    await page.setCookies(...cookies);

    // Correct usage. Will work in both.
    await browserController.setCookies(page, cookies);
}

The BrowserController also includes important information about the browser, such as the context it was launched with. This was difficult to do before SDK v1.

const handlePageFunction = async ({ browserController }) => {
    // Information about the proxy used by the browser
    browserController.launchContext.proxyInfo

    // Session used by the browser
    browserController.launchContext.session
}

BrowserPool methods vs PuppeteerPool

Some functions were removed (in line with earlier deprecations), and some were changed a bit:

// OLD
await puppeteerPool.recyclePage(page);

// NEW
await page.close();

// OLD
await puppeteerPool.retire(page.browser());

// NEW
browserPool.retireBrowserByPage(page);

// OLD
await puppeteerPool.serveLiveViewSnapshot();

// NEW
// There's no LiveView in BrowserPool

Updated PuppeteerCrawlerOptions

To keep PuppeteerCrawler and PlaywrightCrawler consistent, we updated the options.

Removal of gotoFunction

The concept of a configurable gotoFunction is not ideal. Especially since we use a modified gotoExtended. Users have to know this when they override gotoFunction if they want to extend default behavior. We decided to replace gotoFunction with preNavigationHooks and postNavigationHooks.

The following example illustrates how gotoFunction makes things complicated.

const gotoFunction = async ({ request, page }) => {
    // pre-processing
    await makePageStealthy(page);

    // Have to remember how to do this:
    const response = await gotoExtended(page, request, {/* have to remember the defaults */});

    // post-processing
    await page.evaluate(() => {
        window.foo = 'bar';
    });

    // Must not forget!
    return response;
}

const crawler = new Apify.PuppeteerCrawler({
    gotoFunction,
    // ...
})

With preNavigationHooks and postNavigationHooks it's much easier. preNavigationHooks are called with two arguments: crawlingContext and gotoOptions. postNavigationHooks are called only with crawlingContext.

const preNavigationHooks = [
    async ({ page }) => makePageStealthy(page)
];

const postNavigationHooks = [
    async ({ page }) => page.evaluate(() => {
        window.foo = 'bar'
    })
]

const crawler = new Apify.PuppeteerCrawler({
    preNavigationHooks,
    postNavigationHooks,
    // ...
})

launchPuppeteerOptions => launchContext

Those were always a point of confusion because they merged custom Apify options with launchOptions of Puppeteer.

const launchPuppeteerOptions = {
    useChrome: true, // Apify option
    headless: false, // Puppeteer option
}

Use the new launchContext object, which explicitly defines launchOptions. launchPuppeteerOptions were removed.

const crawler = new Apify.PuppeteerCrawler({
    launchContext: {
        useChrome: true, // Apify option
        launchOptions: {
            headless: false // Puppeteer option
        }
    }
})

LaunchContext is also a type of browser-pool and the structure is exactly the same there. SDK only adds extra options.

Removal of launchPuppeteerFunction

browser-pool introduces the idea of lifecycle hooks, which are functions that are executed when a certain event in the browser lifecycle happens.

const launchPuppeteerFunction = async (launchPuppeteerOptions) => {
    if (someVariable === 'chrome') {
        launchPuppeteerOptions.useChrome = true;
    }
    return Apify.launchPuppeteer(launchPuppeteerOptions);
}

const crawler = new Apify.PuppeteerCrawler({
    launchPuppeteerFunction,
    // ...
})

Now you can recreate the same functionality with a preLaunchHook:

const maybeLaunchChrome = (pageId, launchContext) => {
    if (someVariable === 'chrome') {
        launchContext.useChrome = true;
    }
}

const crawler = new Apify.PuppeteerCrawler({
    browserPoolOptions: {
        preLaunchHooks: [maybeLaunchChrome]
    },
    // ...
})

This is better in multiple ways. It is consistent across both Puppeteer and Playwright. It allows you to easily construct your browsers with pre-defined behavior:

const preLaunchHooks = [
    maybeLaunchChrome,
    useHeadfulIfNeeded,
    injectNewFingerprint,
]

And thanks to the addition of crawler.crawlingContexts the functions also have access to the crawlingContext of the request that triggered the launch.

const preLaunchHooks = [
    async function maybeLaunchChrome(pageId, launchContext) {
        const { request } = crawler.crawlingContexts.get(pageId);
        if (request.userData.useHeadful === true) {
            launchContext.launchOptions.headless = false;
        }
    }
]

Launch functions

In addition to Apify.launchPuppeteer() we now also have Apify.launchPlaywright().

Updated arguments

We updated the launch options object because it was a frequent source of confusion.

// OLD
await Apify.launchPuppeteer({
    useChrome: true,
    headless: true,
})

// NEW
await Apify.launchPuppeteer({
    useChrome: true,
    launchOptions: {
        headless: true,
    }
})

Custom modules

Apify.launchPuppeteer already supported the puppeteerModule option. With Playwright, we normalized the name to launcher because the playwright module itself does not launch browsers.

const puppeteer = require('puppeteer');
const playwright = require('playwright');

await Apify.launchPuppeteer();
// Is the same as:
await Apify.launchPuppeteer({
    launcher: puppeteer
})

await Apify.launchPlaywright();
// Is the same as:
await Apify.launchPlaywright({
    launcher: playwright.chromium
})