Upgrading to v3
This page summarizes most of the breaking changes between Crawlee (v3) and Apify SDK (v2). Crawlee is the spiritual successor to Apify SDK, so we decided to keep the versioning and release Crawlee as v3.
Up until version 3 of apify
, the package contained both scraping related tools and Apify platform related helper methods. With v3 we are splitting the whole project into two main parts:
Crawlee monorepo
The crawlee
package consists of several smaller packages, released separately under @crawlee
namespace:
@crawlee/core
: the base for all the crawler implementations, also contains things likeRequest
,RequestQueue
,RequestList
orDataset
classes@crawlee/cheerio
: exportsCheerioCrawler
@crawlee/playwright
: exportsPlaywrightCrawler
@crawlee/puppeteer
: exportsPuppeteerCrawler
@crawlee/jsdom
: exportsJSDOMCrawler
@crawlee/basic
: exportsBasicCrawler
@crawlee/http
: exportsHttpCrawler
(which is used for creating@crawlee/jsdom
and@crawlee/cheerio
)@crawlee/browser
: exportsBrowserCrawler
(which is used for creating@crawlee/playwright
and@crawlee/puppeteer
)@crawlee/memory-storage
:@apify/storage-local
alternative@crawlee/browser-pool
: previouslybrowser-pool
package@crawlee/utils
: utility methods@crawlee/types
: holds TS interfaces mainly about theStorageClient
Installing Crawlee
Most of the Crawlee packages are extending and reexporting each other, so it's enough to install just the one you plan on using, e.g. @crawlee/playwright
if you plan on using playwright
- it already contains everything from the @crawlee/browser
package, which includes everything from @crawlee/basic
, which includes everything from @crawlee/core
.
If we don't care much about additional code being pulled in, we can just use the crawlee
meta-package, which contains (re-exports) most of the @crawlee/*
packages, and therefore contains all the crawler classes.
npm install crawlee
Or if all we need is cheerio support, we can install only @crawlee/cheerio
.
npm install @crawlee/cheerio
When using playwright
or puppeteer
, we still need to install those dependencies explicitly - this allows the users to be in control of which version will be used.
npm install crawlee playwright
# or npm install @crawlee/playwright playwright
Alternatively we can also use the crawlee
meta-package which contains (re-exports) most of the @crawlee/*
packages, and therefore contains all the crawler classes.
Sometimes you might want to use some utility methods from
@crawlee/utils
, so you might want to install that as well. This package contains some utilities that were previously available underApify.utils
. Browser related utilities can be also found in the crawler packages (e.g.@crawlee/playwright
).
Full TypeScript support
Both Crawlee and Apify SDK are full TypeScript rewrite, so they include up-to-date types in the package. For your TypeScript crawlers we recommend using our predefined TypeScript configuration from @apify/tsconfig
package. Don't forget to set the module
and target
to ES2022
or above to be able to use top level await.
The
@apify/tsconfig
config hasnoImplicitAny
enabled, you might want to disable it during the initial development as it will cause build failures if you left some unused local variables in your code.
{
"extends": "@apify/tsconfig",
"compilerOptions": {
"module": "ES2022",
"target": "ES2022",
"outDir": "dist",
"lib": ["DOM"]
},
"include": [
"./src/**/*"
]
}
Docker build
For Dockerfile
we recommend using multi-stage build, so you don't install the dev dependencies like TypeScript in your final image:
# using multistage build, as we need dev deps to build the TS source code
FROM apify/actor-node:20 AS builder
# copy all files, install all dependencies (including dev deps) and build the project
COPY . ./
RUN npm install --include=dev \
&& npm run build
# create final image
FROM apify/actor-node:20
# copy only necessary files
COPY /usr/src/app/package*.json ./
COPY /usr/src/app/README.md ./
COPY /usr/src/app/dist ./dist
COPY /usr/src/app/apify.json ./apify.json
COPY /usr/src/app/INPUT_SCHEMA.json ./INPUT_SCHEMA.json
# install only prod deps
RUN npm --quiet set progress=false \
&& npm install --only=prod --no-optional \
&& echo "Installed NPM packages:" \
&& (npm list --only=prod --no-optional --all || true) \
&& echo "Node.js version:" \
&& node --version \
&& echo "NPM version:" \
&& npm --version
# run compiled code
CMD npm run start:prod
Browser fingerprints
Previously we had a magical stealth
option in the puppeteer crawler that enabled several tricks aiming to mimic the real users as much as possible. While this worked to a certain degree, we decided to replace it with generated browser fingerprints.
In case we don't want to have dynamic fingerprints, we can disable this behaviour via useFingerprints
in browserPoolOptions
:
const crawler = new PlaywrightCrawler({
browserPoolOptions: {
useFingerprints: false,
},
});
Session cookie method renames
Previously, if we wanted to get or add cookies for the session that would be used for the request, we had to call session.getPuppeteerCookies()
or session.setPuppeteerCookies()
. Since this method could be used for any of our crawlers, not just PuppeteerCrawler
, the methods have been renamed to session.getCookies()
and session.setCookies()
respectively. Otherwise, their usage is exactly the same!
Memory storage
When we store some data or intermediate state (like the one RequestQueue
holds), we now use @crawlee/memory-storage
by default. It is an alternative to the @apify/storage-local
, that stores the state inside memory (as opposed to SQLite database used by @apify/storage-local
). While the state is stored in memory, it also dumps it to the file system, so we can observe it, as well as respects the existing data stored in KeyValueStore (e.g. the INPUT.json
file).
When we want to run the crawler on Apify platform, we need to use Actor.init
or Actor.main
, which will automatically switch the storage client to ApifyClient
when on the Apify platform.
We can still use the @apify/storage-local
, to do it, first install it pass it to the Actor.init
or Actor.main
options:
@apify/storage-local
v2.1.0+ is required for Crawlee
import { Actor } from 'apify';
import { ApifyStorageLocal } from '@apify/storage-local';
const storage = new ApifyStorageLocal(/* options like `enableWalMode` belong here */);
await Actor.init({ storage });
Purging of the default storage
Previously the state was preserved between local runs, and we had to use --purge
argument of the apify-cli
. With Crawlee, this is now the default behaviour, we purge the storage automatically on Actor.init/main
call. We can opt out of it via purge: false
in the Actor.init
options.
Renamed crawler options and interfaces
Some options were renamed to better reflect what they do. We still support all the old parameter names too, but not at the TS level.
handleRequestFunction
->requestHandler
handlePageFunction
->requestHandler
handleRequestTimeoutSecs
->requestHandlerTimeoutSecs
handlePageTimeoutSecs
->requestHandlerTimeoutSecs
requestTimeoutSecs
->navigationTimeoutSecs
handleFailedRequestFunction
->failedRequestHandler
We also renamed the crawling context interfaces, so they follow the same convention and are more meaningful:
CheerioHandlePageInputs
->CheerioCrawlingContext
PlaywrightHandlePageFunction
->PlaywrightCrawlingContext
PuppeteerHandlePageFunction
->PuppeteerCrawlingContext
Context aware helpers
Some utilities previously available under Apify.utils
namespace are now moved to the crawling context and are context aware. This means they have some parameters automatically filled in from the context, like the current Request
instance or current Page
object, or the RequestQueue
bound to the crawler.
Enqueuing links
One common helper that received more attention is the enqueueLinks
. As mentioned above, it is context aware - we no longer need pass in the requestQueue
or page
arguments (or the cheerio handle $
). In addition to that, it now offers 3 enqueuing strategies:
EnqueueStrategy.All
('all'
): Matches any URLs foundEnqueueStrategy.SameHostname
('same-hostname'
) Matches any URLs that have the same subdomain as the base URL (default)EnqueueStrategy.SameDomain
('same-domain'
) Matches any URLs that have the same domain name. For example,https://wow.an.example.com
andhttps://example.com
will both be matched for a base url ofhttps://example.com
.
This means we can even call enqueueLinks()
without any parameters. By default, it will go through all the links found on current page and filter only those targeting the same subdomain.
Moreover, we can specify patterns the URL should match via globs:
const crawler = new PlaywrightCrawler({
async requestHandler({ enqueueLinks }) {
await enqueueLinks({
globs: ['https://crawlee.dev/*/*'],
// we can also use `regexps` and `pseudoUrls` keys here
});
},
});
Implicit RequestQueue
instance
All crawlers now have the RequestQueue
instance automatically available via crawler.getRequestQueue()
method. It will create the instance for you if it does not exist yet. This mean we no longer need to create the RequestQueue
instance manually, and we can just use crawler.addRequests()
method described underneath.
We can still create the
RequestQueue
explicitly, thecrawler.getRequestQueue()
method will respect that and return the instance provided via crawler options.
crawler.addRequests()
We can now add multiple requests in batches. The newly added addRequests
method will handle everything for us. It enqueues the first 1000 requests and resolves, while continuing with the rest in the background, again in a smaller 1000 items batches, so we don't fall into any API rate limits. This means the crawling will start almost immediately (within few seconds at most), something previously possible only with a combination of RequestQueue
and RequestList
.
// will resolve right after the initial batch of 1000 requests is added
const result = await crawler.addRequests([/* many requests, can be even millions */]);
// if we want to wait for all the requests to be added, we can await the `waitForAllRequestsToBeAdded` promise
await result.waitForAllRequestsToBeAdded;
Less verbose error logging
Previously an error thrown from inside request handler resulted in full error object being logged. With Crawlee, we log only the error message as a warning as long as we know the request will be retried. If you want to enable verbose logging like in v2, use the CRAWLEE_VERBOSE_LOG
env var.
Request.label
shortcut
Labeling requests used to work via the Request.userData
object. With Crawlee, we can also use the Request.label
shortcut. It is implemented as a get/set
pair, using the value from Request.userData
. The support for this shortcut is also added to the enqueueLinks
options interface.
async requestHandler({ request, enqueueLinks }) {
if (request.label !== 'DETAIL') {
await enqueueLinks({
globs: ['...'],
label: 'DETAIL',
});
}
}
Removal of requestAsBrowser
In v1 we replaced the underlying implementation of requestAsBrowser
to be just a proxy over calling got-scraping
- our custom extension to got
that tries to mimic the real browsers as much as possible. With v3, we are removing the requestAsBrowser
, encouraging the use of got-scraping
directly.
For easier migration, we also added context.sendRequest()
helper that allows processing the context bound Request
object through got-scraping
:
const crawler = new BasicCrawler({
async requestHandler({ sendRequest, log }) {
// we can use the options parameter to override gotScraping options
const res = await sendRequest({ responseType: 'json' });
log.info('received body', res.body);
},
});
How to use sendRequest()
?
Removed options
The useInsecureHttpParser
option has been removed. It's permanently set to true
in order to better mimic browsers' behavior.
Got Scraping automatically performs protocol negotiation, hence we removed the useHttp2
option. It's set to true
- 100% of browsers nowadays are capable of HTTP/2 requests. Oh, more and more of the web is using it too!
Renamed options
In the requestAsBrowser
approach, some of the options were named differently. Here's a list of renamed options:
payload
This options represents the body to send. It could be a string
or a Buffer
. However, there is no payload
option anymore. You need to use body
instead. Or, if you wish to send JSON, json
. Here's an example:
// Before:
await Apify.utils.requestAsBrowser({ …, payload: 'Hello, world!' });
await Apify.utils.requestAsBrowser({ …, payload: Buffer.from('c0ffe', 'hex') });
await Apify.utils.requestAsBrowser({ …, json: { hello: 'world' } });
// After:
await gotScraping({ …, body: 'Hello, world!' });
await gotScraping({ …, body: Buffer.from('c0ffe', 'hex') });
await gotScraping({ …, json: { hello: 'world' } });
ignoreSslErrors
It has been renamed to https.rejectUnauthorized
. By default, it's set to false
for convenience. However, if you want to make sure the connection is secure, you can do the following:
// Before:
await Apify.utils.requestAsBrowser({ …, ignoreSslErrors: false });
// After:
await gotScraping({ …, https: { rejectUnauthorized: true } });
Please note: the meanings are opposite! So we needed to invert the values as well.
header-generator
options
useMobileVersion
, languageCode
and countryCode
no longer exist. Instead, you need to use headerGeneratorOptions
directly:
// Before:
await Apify.utils.requestAsBrowser({
…,
useMobileVersion: true,
languageCode: 'en',
countryCode: 'US',
});
// After:
await gotScraping({
…,
headerGeneratorOptions: {
devices: ['mobile'], // or ['desktop']
locales: ['en-US'],
},
});
timeoutSecs
In order to set a timeout, use timeout.request
(which is milliseconds now).
// Before:
await Apify.utils.requestAsBrowser({
…,
timeoutSecs: 30,
});
// After:
await gotScraping({
…,
timeout: {
request: 30 * 1000,
},
});
throwOnHttpErrors
throwOnHttpErrors
→ throwHttpErrors
. This options throws on unsuccessful HTTP status codes, for example 404
. By default, it's set to false
.
decodeBody
decodeBody
→ decompress
. This options decompresses the body. Defaults to true
- please do not change this or websites will break (unless you know what you're doing!).
abortFunction
This function used to make the promise throw on specific responses, if it returned true
. However, it wasn't that useful.
You probably want to cancel the request instead, which you can do in the following way:
const promise = gotScraping(…);
promise.on('request', request => {
// Please note this is not a Got Request instance, but a ClientRequest one.
// https://nodejs.org/api/http.html#class-httpclientrequest
if (request.protocol !== 'https:') {
// Unsecure request, abort.
promise.cancel();
// If you set `isStream` to `true`, please use `stream.destroy()` instead.
}
});
const response = await promise;
Removal of browser pool plugin mixing
Previously, you were able to have a browser pool that would mix Puppeteer and Playwright plugins (or even your own custom plugins if you've built any). As of this version, that is no longer allowed, and creating such a browser pool will cause an error to be thrown (it's expected that all plugins that will be used are of the same type).
As an example, this change disallows a pool to mix Puppeteer with Playwright. You can still create pools that use multiple Playwright plugins, each with a different launcher if you want!
Handling requests outside of browser
One small feature worth mentioning is the ability to handle requests with browser crawlers outside the browser. To do that, we can use a combination of Request.skipNavigation
and context.sendRequest()
.
Take a look at how to achieve this by checking out the Skipping navigation for certain requests example!
Logging
Crawlee exports the default log
instance directly as a named export. We also have a scoped log
instance provided in the crawling context - this one will log messages prefixed with the crawler name and should be preferred for logging inside the request handler.
const crawler = new CheerioCrawler({
async requestHandler({ log, request }) {
log.info(`Opened ${request.loadedUrl}`);
},
});
Auto-saved crawler state
Every crawler instance now has useState()
method that will return a state object we can use. It will be automatically saved when persistState
event occurs. The value is cached, so we can freely call this method multiple times and get the exact same reference. No need to worry about saving the value either, as it will happen automatically.
const crawler = new CheerioCrawler({
async requestHandler({ crawler }) {
const state = await crawler.useState({ foo: [] as number[] });
// just change the value, no need to care about saving it
state.foo.push(123);
},
});
Apify SDK
The Apify platform helpers can be now found in the Apify SDK (apify
NPM package). It exports the Actor
class that offers following static helpers:
ApifyClient
shortcuts:addWebhook()
,call()
,callTask()
,metamorph()
- helpers for running on Apify platform:
init()
,exit()
,fail()
,main()
,isAtHome()
,createProxyConfiguration()
- storage support:
getInput()
,getValue()
,openDataset()
,openKeyValueStore()
,openRequestQueue()
,pushData()
,setValue()
- events support:
on()
,off()
- other utilities:
getEnv()
,newClient()
,reboot()
Actor.main
is now just a syntax sugar around calling Actor.init()
at the beginning and Actor.exit()
at the end (plus wrapping the user function in try/catch block). All those methods are async and should be awaited - with node 16 we can use the top level await for that. In other words, following is equivalent:
import { Actor } from 'apify';
await Actor.init();
// your code
await Actor.exit('Crawling finished!');
import { Actor } from 'apify';
await Actor.main(async () => {
// your code
}, { statusMessage: 'Crawling finished!' });
Actor.init()
will conditionally set the storage implementation of Crawlee to the ApifyClient
when running on the Apify platform, or keep the default (memory storage) implementation otherwise. It will also subscribe to the websocket events (or mimic them locally). Actor.exit()
will handle the tear down and calls process.exit()
to ensure our process won't hang indefinitely for some reason.
Events
Apify SDK (v2) exports Apify.events
, which is an EventEmitter
instance. With Crawlee, the events are managed by EventManager
class instead. We can either access it via Actor.eventManager
getter, or use Actor.on
and Actor.off
shortcuts instead.
-Apify.events.on(...);
+Actor.on(...);
We can also get the
EventManager
instance viaConfiguration.getEventManager()
.
In addition to the existing events, we now have an exit
event fired when calling Actor.exit()
(which is called at the end of Actor.main()
). This event allows you to gracefully shut down any resources when Actor.exit
is called.
Smaller/internal breaking changes
Apify.call()
is now just a shortcut for runningApifyClient.actor(actorId).call(input, options)
, while also taking the token inside env vars into accountApify.callTask()
is now just a shortcut for runningApifyClient.task(taskId).call(input, options)
, while also taking the token inside env vars into accountApify.metamorph()
is now just a shortcut for runningApifyClient.task(taskId).metamorph(input, options)
, while also taking the ACTOR_RUN_ID inside env vars into accountApify.waitForRunToFinish()
has been removed, useApifyClient.waitForFinish()
insteadActor.main/init
purges the storage by default- remove
purgeLocalStorage
helper, move purging to the storage class directlyStorageClient
interface now has optionalpurge
method- purging happens automatically via
Actor.init()
(you can opt out viapurge: false
in the options ofinit/main
methods)
QueueOperationInfo.request
is no longer availableRequest.handledAt
is now string date in ISO formatRequest.inProgress
andRequest.reclaimed
are nowSet
s instead of POJOsinjectUnderscore
from puppeteer utils has been removedAPIFY_MEMORY_MBYTES
is no longer taken into account, useCRAWLEE_AVAILABLE_MEMORY_RATIO
instead- some
AutoscaledPool
options are no longer available:cpuSnapshotIntervalSecs
andmemorySnapshotIntervalSecs
has been replaced with top levelsystemInfoIntervalMillis
configurationmaxUsedCpuRatio
has been moved to the top level configuration
ProxyConfiguration.newUrlFunction
can be async..newUrl()
and.newProxyInfo()
now return promises.prepareRequestFunction
andpostResponseFunction
options are removed, use navigation hooks insteadgotoFunction
andgotoTimeoutSecs
are removed- removed compatibility fix for old/broken request queues with null
Request
props fingerprintsOptions
renamed tofingerprintOptions
(fingerprints
->fingerprint
).fingerprintOptions
now acceptuseFingerprintCache
andfingerprintCacheSize
(instead ofuseFingerprintPerProxyCache
andfingerprintPerProxyCacheSize
, which are now no longer available). This is because the cached fingerprints are no longer connected to proxy URLs but to sessions.