Puppeteer Cluster

Create a cluster of puppeteer workers. This library spawns a pool of Chromium instances via [Puppeteer] and helps to keep track of jobs and errors. This is helpful if you want to crawl multiple pages or run tests in parallel. Puppeteer Cluster takes care of reusing Chromium and restarting the browser in case of errors.

• Installation
• Usage
• Examples
• Concurrency implementations
• Typings for input/output (via TypeScript Generics)
• Debugging
• API
• License

What does this library do?

• Handling of crawling errors
• Auto restarts the browser in case of a crash
• Can automatically retry if a job fails
• Different concurrency models to choose from (pages, contexts, browsers)
• Simple to use, small boilerplate
• Progress view and monitoring statistics (see below)

Installation

Install using your favorite package manager:

npm install --save puppeteer # in case you don't already have it installed 
npm install --save puppeteer-cluster

Alternatively, use yarn:

yarn add puppeteer puppeteer-cluster

Usage

The following is a typical example of using puppeteer-cluster. A cluster is created with 2 concurrent workers. Then a task is defined which includes going to the URL and taking a screenshot. We then queue two jobs and wait for the cluster to finish.

const { Cluster } = require('puppeteer-cluster');

(async () => {
  const cluster = await Cluster.launch({
    concurrency: Cluster.CONCURRENCY_CONTEXT,
    maxConcurrency: 2,
  });

  await cluster.task(async ({ page, data: url }) => {
    await page.goto(url);
    const screen = await page.screenshot();
    // Store screenshot, do something else
  });

  cluster.queue('http://www.google.com/');
  cluster.queue('http://www.wikipedia.org/');
  // many more pages

  await cluster.idle();
  await cluster.close();
})();

Examples

• Simple example
• Wait for a task to be executed
• Minimal screenshot server with express
• Deep crawling the Google search results
• Crawling the Alexa Top 1 Million
• Queuing functions (simple)
• Queuing functions (complex)
• Error handling
• Using a different puppeteer library (like puppeteer-core or puppeteer-firefox)
• Provide types for input/output with TypeScript generics

Concurrency implementations

There are different concurrency models, which define how isolated each job is run. You can set it in the options when calling Cluster.launch. The default option is Cluster.CONCURRENCY_CONTEXT, but it is recommended to always specify which one you want to use.

Concurrency	Description	Shared data
CONCURRENCY_PAGE	One [Page] for each URL	Shares everything (cookies, localStorage, etc.) between jobs.
CONCURRENCY_CONTEXT	Incognito page (see BrowserContext) for each URL	No shared data.
CONCURRENCY_BROWSER	One browser (using an incognito page) per URL. If one browser instance crashes for any reason, this will not affect other jobs.	No shared data.
Custom concurrency (experimental)	You can create your own concurrency implementation. Copy one of the files of the concurrency/built-in directory and implement ConcurrencyImplementation. Then provide the class to the option concurrency. This part of the library is currently experimental and might break in the future, even in a minor version upgrade while the version has not reached 1.0.	Depends on your implementation

Typings for input/output (via TypeScript Generics)

To allow proper type checks with TypeScript you can provide generics. In case no types are provided, any is assumed for input and output. See the following minimal example or check out the more complex typings example for more information.

  const cluster: Cluster<string, number> = await Cluster.launch(/* ... */);

  await cluster.task(async ({ page, data }) => {
    // TypeScript knows that data is a string and expects this function to return a number
    return 123;
  });

  // Typescript expects a string as argument ...
  cluster.queue('http://...');

  // ... and will return a number when execute is called.
  const result = await cluster.execute('https://www.google.com');

Debugging

Try to checkout the puppeteer debugging tips first. Your problem might not be related to puppeteer-cluster, but puppeteer itself. Additionally, you can enable verbose logging to see which data is consumed by which worker and some other cluster information. Set the DEBUG environment variable to puppeteer-cluster:*. See an example below or checkout the debug docs for more information.

# Linux
DEBUG='puppeteer-cluster:*' node examples/minimal
# Windows Powershell
$env:DEBUG='puppeteer-cluster:*';node examples/minimal

API

• class: Cluster
• Cluster.launch(options)
• cluster.task(taskFunction)
• cluster.queue([data] [, taskFunction])
• cluster.execute([data] [, taskFunction])
• cluster.idle()
• cluster.close()

class: Cluster

Cluster module provides a method to launch a cluster of Chromium instances.

event: 'taskerror'

• <[Error]>
• <[string]|[Object]>
• <[boolean]>

Emitted when a queued task ends in an error for some reason. Reasons might be a network error, your code throwing an error, timeout hit, etc. The first argument will the error itself. The second argument is the URL or data of the job (as given to [Cluster.queue]). If retryLimit is set to a value greater than 0, the cluster will automatically requeue the job and retry it again later. The third argument is a boolean which indicates whether this task will be retried. In case the task was queued via [Cluster.execute] there will be no event fired.

  cluster.on('taskerror', (err, data, willRetry) => {
      if (willRetry) {
        console.warn(`Encountered an error while crawling ${data}. ${err.message}\nThis job will be retried`);
      } else {
        console.error(`Failed to crawl ${data}: ${err.message}`);
      }
  });

event: 'queue'

• <\?[Object]>
• <\?[function]>

Emitted when a task is queued via [Cluster.queue] or [Cluster.execute]. The first argument is the object containing the data (if any data is provided). The second argument is the queued function (if any). In case only a function is provided via [Cluster.queue] or [Cluster.execute], the first argument will be undefined. If only data is provided, the second argument will be undefined.

Cluster.launch(options)

• options <[Object]> Set of configurable options for the cluster. Can have the following fields:
• concurrency <Cluster.CONCURRENCY_PAGE|Cluster.CONCURRENCY_CONTEXT|Cluster.CONCURRENCY_BROWSER|ConcurrencyImplementation> The chosen concurrency model. See Concurreny implementations for more information. Defaults to Cluster.CONCURRENCY_CONTEXT. Alternatively you can provide a class implementing ConcurrencyImplementation.
• maxConcurrency <[number]> Maximal number of parallel workers. Defaults to 1.
• puppeteerOptions <[Object]> Object passed to [puppeteer.launch]. See puppeteer documentation for more information. Defaults to {}.
• perBrowserOptions <[Array]<[Object]>> Object passed to [puppeteer.launch] for each individual browser. If set, puppeteerOptions will be ignored. Defaults to undefined (meaning that puppeteerOptions will be used).
• retryLimit <[number]> How often do you want to retry a job before marking it as failed. Ignored by tasks queued via [Cluster.execute]. Defaults to 0.
• retryDelay <[number]> How much time should pass at minimum between the job execution and its retry. Ignored by tasks queued via [Cluster.execute]. Defaults to 0.
• sameDomainDelay <[number]> How much time should pass at minimum between two requests to the same domain. If you use this field, the queued data must be your URL or data must be an object containing a field called url.
• skipDuplicateUrls <[boolean]> If set to true, will skip URLs which were already crawled by the cluster. Defaults to false. If you use this field, the queued data must be your URL or data must be an object containing a field called url.
• timeout <[number]> Specify a timeout for all tasks. Defaults to 30000 (30 seconds).
• monitor <[boolean]> If set to true, will provide a small command line output to provide information about the crawling process. Defaults to false.
• workerCreationDelay <[number]> Time between creation of two workers. Set this to a value like 100 (0.1 seconds) in case you want some time to pass before another worker is created. You can use this to prevent a network peak right at the start. Defaults to 0 (no delay).
• puppeteer <[Object]> In case you want to use a different puppeteer library (like puppeteer-core or puppeteer-extra), pass the object here. If not set, will default to using puppeteer. When using puppeteer-core, make sure to also provide puppeteerOptions.executablePath.
• returns: <[Promise]<[Cluster]>>

The method launches a cluster instance.

cluster.task(taskFunction)

• taskFunction <function> Sets the function, which will be called for each job. The function will be called with an object having the following fields:
• page <[Page]> The page given by puppeteer, which provides methods to interact with a single tab in Chromium.
• data The data of the job you provided to [Cluster.queue].
• worker <[Object]> An object containing information about the worker executing the current job.
  • id <[number]> ID of the worker. Worker IDs start at 0.
• returns: <[Promise]>

Specifies a task for the cluster. A task is called for each job you queue via [Cluster.queue]. Alternatively you can directly queue the function that you want to be executed. See [Cluster.queue] for an example.

cluster.queue([data] [, taskFunction])

• data Data to be queued. This might be your URL (a string) or a more complex object containing data. The data given will be provided to your task function(s). See [examples] for a more complex usage of this argument.
• taskFunction <[function]> Function like the one given to [Cluster.task]. If a function is provided, this function will be called (only for this job) instead of the function provided to [Cluster.task]. The function will be called with an object having the following fields:
• page <[Page]> The page given by puppeteer, which provides methods to interact with a single tab in Chromium.
• data The data of the job you provided as first argument to [Cluster.queue]. This might be undefined in case you only specified a function.
• worker <[Object]> An object containing information about the worker executing the current job.
  • id <[number]> ID of the worker. Worker IDs start at 0.
• returns: <[Promise]>

Puts a URL or data into the queue. Alternatively (or even additionally) you can queue functions. See the examples about function queuing for more information: (Simple function queuing, complex function queuing).

Be aware that this function only returns a Promise for backward compatibility reasons. This function does not run asynchronously and will immediately return.

cluster.execute([data] [, taskFunction])

• data Data to be queued. This might be your URL (a string) or a more complex object containing data. The data given will be provided to your task function(s). See [examples] for a more complex usage of this argument.
• taskFunction <[function]> Function like the one given to [Cluster.task]. If a function is provided, this function will be called (only for this job) instead of the function provided to [Cluster.task]. The function will be called with an object having the following fields:
• page <[Page]> The page given by puppeteer, which provides methods to interac