You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
Stefan Zabka 5a67ce87cd
Merge pull request #591 from mozilla/issue_559
4 days ago
automation Fixed log reading in python3.8.1 1 week ago
test Disable intermittent mp_logger test failure 1 week ago Corrected the docker tag to push 8 months ago
.dockerignore Don't include .git in Docker images 8 months ago
.gitignore Type annotation and refactoring in 2 weeks ago
.travis.yml Make the tests run with Python 3.8 on Travis 2 weeks ago
CHANGELOG Version bump, removing warning message 11 months ago Add Mozilla Code of Conduct file 1 year ago
Dockerfile removed python2 from scripts and dockerfile 4 months ago
LICENSE Store POST request data as JSON in the database. 3 years ago Added type hints to development pointers 2 weeks ago
VERSION Version bump, removing warning message 11 months ago Create data_directory if it doesn't exist + drop an init file in main 4 years ago Print the result of the build extension script when it has finished 9 months ago setting site_rank 2 weeks ago Callstack capturing activated in 2 months ago Add libsasl2-dev to development dependencies 1 week ago Inlined FF version upgrade description 1 month ago Split the installer into smaller scripts 9 months ago Remove --user flag from pip install 1 week ago Inlined FF version upgrade description 1 month ago Split the installer into smaller scripts 9 months ago Update requirements to work when layered 5 months ago
requirements-dev.txt Update python dependencies 2 weeks ago removing six as a project dependency 4 months ago
requirements.txt Update python dependencies 2 weeks ago Add --shm-size argument to docker osx helper script 7 months ago
setup.cfg Ignore all flake8-quotes error codes 7 months ago

OpenWPM Build Status

OpenWPM Matrix Channel

OpenWPM is a web privacy measurement framework which makes it easy to collect data for privacy studies on a scale of thousands to millions of websites. OpenWPM is built on top of Firefox, with automation provided by Selenium. It includes several hooks for data collection. Check out the instrumentation section below for more details.

Table of Contents


OpenWPM is a Python >3.6 application developed and tested for Ubuntu 18.04. Python 2 is not supported. An installation script, is included to install both the system and python dependencies automatically. A few of the python dependencies require specific versions, so you should install the dependencies in a virtual environment if you’re installing a shared machine. If you plan to develop OpenWPM’s instrumentation extension or run tests you will also need to install the development dependencies included in

It is likely that OpenWPM will work on platforms other than Ubuntu, however we do not officially support anything else. For pointers on alternative platform support see the wiki.

Quick Start

Once installed, it is very easy to run a quick test of OpenWPM. Check out for an example. This will use the default setting specified in automation/default_manager_params.json and automation/default_browser_params.json, with the exception of the changes specified in

More information on the instrumentation and configuration parameters is given below.

The wiki provides a more in-depth tutorial, including a platform demo and a description of the additional commands available. You can also take a look at two of our past studies, which use the infrastructure:

  1. The Web Never Forgets
  2. Cookies that Give You Away

Instrumentation and Data Access

OpenWPM provides several instrumentation modules which can be enabled independently of each other for each crawl. More detail on the output is available below.

  • HTTP Request and Response Headers, redirects, and POST request bodies
    • Set browser_params['http_instrument'] = True
    • Data is saved to the http_requests, http_responses, and http_redirects tables.
      • http_requests schema documentation
      • channel_id can be used to link a request saved in the http_requests table to its corresponding response in the http_responses table.
      • channel_id can also be used to link a request to the subsequent request that results after an HTTP redirect (3XX response). Use the http_redirects table, which includes a mapping between old_channel_id, the channel_id of the HTTP request that resulted in a 3XX response, and new_channel_id, the HTTP request that resulted from that redirect.
    • OCSP POST request bodies are not recorded
    • Note: request and response headers for cached content are also saved, with the exception of images. See: Bug 634073.
  • Javascript Calls
    • Records all method calls (with arguments) and property accesses for APIs of potential fingerprinting interest:
      • HTML5 Canvas
      • HTML5 WebRTC
      • HTML5 Audio
      • Plugin access (via navigator.plugins)
      • MIMEType access (via navigator.mimeTypes)
      • window.Storage, window.localStorage, window.sessionStorage, and access.
      • Navigator properties (e.g. appCodeName, oscpu, userAgent, …)
      • Window properties (via window.screen)
    • Set browser_params['js_instrument'] = True
    • Data is saved to the javascript table.
  • Response body content
    • Saves all files encountered during the crawl to a LevelDB database de-duplicated by the md5 hash of the content.
    • Set browser_params['save_content'] = True
    • The content_hash column of the http_responses table contains the md5 hash for each script, and can be used to do content lookups in the LevelDB content database.
    • NOTE: this instrumentation may lead to performance issues when a large number of browsers are in use.
    • Set browser_params['save_content'] to a comma-separated list of resource_types to save only specific types of files, for instance browser_params['save_content'] = "script" to save only Javascript files. This will lessen the performance impact of this instrumentation when a large number of browsers are used in parallel.
  • Flash Cookies
    • Recorded by scanning the respective Flash directories after each page visit.
    • To enable: call the CommandSequence::dump_flash_cookies command after a page visit. Note that calling this command will close the current tab before recording the cookie changes.
    • Data is saved to the flash_cookies table.
    • NOTE: Flash cookies are shared across browsers, so this instrumentation will not correctly attribute flash cookie changes if more than 1 browser is running on the machine.
  • Cookie Access
    • Set browser_params['cookie_instrument'] = True
    • Data is saved to the javascript_cookies table.
    • Will record cookies set both by Javascript and via HTTP Responses
  • Log Files
    • Stored in the directory specified by manager_params['data_directory'].
    • Name specified by manager_params['log_file'].
  • Browser Profile
    • Contains cookies, Flash objects, and so on that are dumped after a crawl is finished
    • Automatically saved when the platform closes or crashes by specifying browser_params['profile_archive_dir'].
    • Save on-demand with the CommandSequence::dump_profile command.
  • Rendered Page Source
    • Save the top-level frame’s rendered source with the CommandSequence::dump_page_source command.
    • Save the full rendered source (including all nested iframes) with the CommandSequence::recursive_dump_page_source command.
      • The page source is saved in the following nested json structure: { 'doc_url': "", 'source': "<html> ... </html>", 'iframes': { 'frame_1': {'doc_url': ..., 'source': ..., 'iframes: { ... }}, 'frame_2': {'doc_url': ..., 'source': ..., 'iframes: { ... }}, 'frame_3': { ... } } }
  • Screenshots
    • Selenium 3 can be used to screenshot an individual element. None of the built-in commands offer this functionality, but you can use it when writing your own. See the Selenium documentation.
    • Viewport screenshots (i.e. a screenshot of the portion of the website visible in the browser’s window) are available with the CommandSequence::save_screenshot command.
    • Full-page screenshots (i.e. a screenshot of the entire rendered DOM) are available with the CommandSequence::screenshot_full_page command.
      • This functionality is not yet supported by Selenium/geckodriver, though it is planned. We produce screenshots by using JS to scroll the page and take a viewport screenshot at each location. This method will save the parts and a stitched version in the screenshot_path.
      • Since the screenshots are stitched they have some limitations:
        • On the area of the page present when the command is called will be captured. Sites which dynamically expand when scrolled (i.e., infinite scroll) will only go as far as the original height.
        • We only scroll vertically, so pages that are wider than the viewport will be clipped.
        • In geckodriver v0.15 doing any scrolling (or having devtools open) seems to break element-only screenshots. So using this command will cause any future element-only screenshots to be misaligned.

Output Format

Local Databases

By default OpenWPM saves all data locally on disk in a variety of formats. Most of the instrumentation saves to a SQLite database specified by manager_params['database_name'] in the main output directory. Response bodies are saved in a LevelDB database named content.ldb, and are keyed by the hash of the content. In addition, the browser commands that dump page source and save screenshots save them in the sources and screenshots subdirectories of the main output directory. The SQLite schema specified by: automation/DataAggregator/schema.sql. You can specify additional tables inline by sending a create_table message to the data aggregator.

Parquet on Amazon S3 Experimental

As an option, OpenWPM can save data directly to an Amazon S3 bucket as a Parquet Dataset. This is currently experimental and hasn’t been thoroughly tested. Screenshots, and page source saving is not currently supported and will still be stored in local databases and directories. To enable S3 saving specify the following configuration parameters in manager_params:

  • Output format: manager_params['output_format'] = 's3'
  • S3 bucket name: manager_params['s3_bucket'] = 'openwpm-test-crawl'
  • Directory within S3 bucket: manager_params['s3_directory'] = '2018-09-09_test-crawl-new'

In order to save to S3 you must have valid access credentials stored in ~/.aws. We do not currently allow you to specify an alternate storage location.

NOTE: The schemas should be kept in sync with the exception of output-specific columns (e.g., instance_id in the S3 output). You can compare the two schemas by running diff -y automation/DataAggregator/schema.sql automation/DataAggregator/

Browser and Platform Configuration

The browser and platform can be configured by two separate dictionaries. The platform configuration options can be set in manager_params, while the browser configuration options can be set in browser_params. The default settings are given in automation/default_manager_params.json and automation/default_browser_params.json.

To load the default configuration parameter dictionaries we provide a helper function TaskManager::load_default_params. For example:

from automation import TaskManager
manager_params, browser_params = TaskManager.load_default_params(num_browsers=5)

where manager_params is a dictionary and browser_params is a length 5 list of configuration dictionaries.

Platform Configuration Options

  • data_directory
    • The directory in which to output the crawl database and related files. The directory given will be created if it does not exist.
  • log_directory
    • The directory in which to output platform logs. The directory given will be created if it does not exist.
  • log_file
    • The name of the log file to be written to log_directory.
  • database_name
    • The name of the database file to be written to data_directory
  • failure_limit
    • The number of successive command failures the platform will tolerate before raising a CommandExecutionError exception. Otherwise the default is set to 2 x the number of browsers plus 10.
  • testing
    • A platform wide flag that can be used to only run certain functionality while testing. For example, the Javascript instrumentation exposes its instrumentation function on the page script global to allow test scripts to instrument objects on-the-fly. Depending on where you would like to add test functionality, you may need to propagate the flag.
    • This is not something you should enable during normal crawls.

Browser Configuration Options

Note: Instrumentation configuration options are described in the Instrumentation and Data Access section and profile configuration options are described in the Browser Profile Support section. As such, these options are left out of this section.

  • bot_mitigation
    • Performs some actions to prevent the platform from being detected as a bot.
    • Note, these aren’t comprehensive and automated interaction with the site will still appear very bot-like.
  • disable_flash
    • Flash is disabled by default. Set this to False to re-enable. Note that flash cookies are shared between browsers.
  • headless
    • Launch the browser in headless mode (supported as of Firefox 56), no GUI will be visible.
    • Use this when running browsers on a remote machine or to run crawls in the background on a local machine.
  • browser
    • Used to specify which browser to launch. Currently only firefox is supported.
    • Other browsers may be added in the future.
  • tp_cookies
    • Specifies the third-party cookie policy to set in Firefox.
    • The following options are supported:
    • always: Accept all third-party cookies
    • never: Never accept any third-party cookies
    • from_visited: Only accept third-party cookies from sites that have been visited as a first party.
  • donottrack
    • Set to True to enable Do Not Track in the browser.
  • disconnect
    • Set to True to enable Disconnect with all blocking enabled
    • The filter list may be automatically updated. We recommend checking the version of the xpi located here, which may be outdated.
  • ghostery
    • Set to True to enable Ghostery with all blocking enabled
    • The filter list won’t be automatically updated. We recommend checking the version of the xpi located here, which may be outdated.
  • https-everywhere
    • Set to True to enable HTTPS Everywhere in the browser.
    • The filter list won’t be automatically updated. We recommend checking the version of the xpi located here, which may be outdated.
  • ublock-origin
    • Set to True to enable uBlock Origin in the browser.
    • The filter lists may be automatically updated. We recommend checking the version of the xpi located here, which may be outdated.
  • tracking-protection

Browser Profile Support

WARNING: Stateful crawls are currently not supported. Attempts to run stateful crawls will throw NotImplementedErrors. The work required to restore support is tracked in this project.

Stateful vs Stateless crawls

By default OpenWPM performs a “stateful” crawl, in that it keeps a consistent browser profile between page visits in the same browser. If the browser freezes or crashes during the crawl, the profile is saved to disk and restored before the next page visit.

It’s also possible to run “stateless” crawls, in which each new page visit uses a fresh browser profile. To perform a stateless crawl you can restart the browser after each command sequence by setting the reset initialization argument to True when creating the command sequence. As an example:

manager = TaskManager.TaskManager(manager_params, browser_params)

for site in sites:
    command_sequence = CommandSequence.CommandSequence(site, reset=True)
    command_sequence.get(sleep=30, timeout=60)

In this example, the browser will get the requested site, sleep for 30 seconds, dump the profile cookies to the crawl database, and then restart the browser before visiting the next site in sites.

Loading and saving a browser profile

It’s possible to load and save profiles during stateful crawls. Profile dumps currently consist of the following browser storage items:

  • cookies
  • localStorage
  • IndexedDB
  • browser history

Other browser state, such as the browser cache, is not saved. In Issue #62 we plan to expand profiles to include all browser storage.

Save a profile

A browser’s profile can be saved to disk for use in later crawls. This can be done using a browser command or by setting a browser configuration parameter. For long running crawls we recommend saving the profile using the browser configuration parameter as the platform will take steps to save the profile in the event of a platform-level crash, whereas there is no guarantee the browser command will run before a crash.

Browser configuration parameter: Set the profile_archive_dir browser parameter to a directory where the browser profile should be saved. The profile will be automatically saved when TaskManager::close is called or when a platform-level crash occurs.

Browser command: See the command definition wiki page for more information.

Load a profile

To load a profile, specify the profile_tar browser parameter in the browser configuration dictionary. This should point to the location of the profile.tar or (profile.tar.gz if compressed) file produced by OpenWPM. The profile will be automatically extracted and loaded into the browser instance for which the configuration parameter was set.

Development pointers

Much of OpenWPM’s instrumentation is included in a Firefox add-on SDK extension. Thus, in order to add or change instrumentation you will need a few additional dependencies, which can be installed with

Types Annotations in Python

We as maintainers have decided it would be helpful to have Python3 type annotations for the python part of this project to catch errors earlier, get better code completion and allow bigger changes down the line with more confidence. As such you should strive to add type annotations to all new code you add to the project as well as the one you plan to change fundamentally.

Editing instrumentation

The instrumentation extension is included in /automation/Extension/firefox/. The instrumentation itself (used by the above extension) is included in /automation/Extension/webext-instrumentation/. Any edits within these directories will require the extension to be re-built to produce a new openwpm.xpi with your updates. You can use to do this.

Debugging the platform

Manual debugging with OpenWPM can be difficult. By design the platform runs all browsers in separate processes and swallows all exceptions (with the intent of continuing the crawl). We recommend using

This utility allows manual debugging of the extension instrumentation with or without Selenium enabled, as well as makes it easy to launch a Selenium instance (without any instrumentation)

  • python -m test.manual_test builds the current extension directory and launches a Firefox instance with it.
  • python -m test.manual_test --selenium launches a Firefox Selenium instance after automatically rebuilding openwpm.xpi. The script then drops into an ipython shell where the webdriver instance is available through variable driver.
  • python -m test.manual_test --selenium --no_extension launches a Firefox Selenium instance with no instrumentation. The script then drops into an ipython shell where the webdriver instance is available through variable driver.

Managing requirements

We use pip-tools to pin requirements. This means that the requirements.txt and requirements-dev.txt files should not be edited directly. Instead, place new requirements in or Requirements necessary to run OpenWPM should be placed in the former, while those only required to run the tests (or perform other development tasks) should be placed in the latter.

To update dependencies, run the following two commands in order:

  • pip-compile --upgrade
  • pip-compile --upgrade

It’s important that these are run in order, as we layer the dev dependencies on the output of the pinned production dependencies as per the official documentation. This means you may need to manually pin some versions of a dependency in so there exists a compatible version for a dependency in Before you run an upgrade, check if the previous pins that are related to layering in are still necessary by removing them and attempting an upgrade without the pins.

Running tests

OpenWPM’s tests are build on py.test. To run the tests you will need a few additional dependencies, which can be installed by running

Once installed, execute py.test -vv in the test directory to run all tests.

Mac OSX (Limited support for developers)

We’ve added an installation file to make it easier to run tests and develop on Mac OSX. To install the dependencies on Mac OSX, run instead of and in the official getting started instructions.

This will install Python packages in a local Python 3 virtualenv, download the latest Unbranded Firefox Release into the current folder, move geckodriver next to the Firefox binary and install development dependencies. For the OpenWPM to be aware of which Firefox installation to run, set the FIREFOX_BINARY environment variable before running any commands.

Example, running a demo crawl on Mac OSX:

source venv/bin/activate

Running the OpenWPM tests on Mac OSX:

source venv/bin/activate
python -m pytest -vv

For more detailed setup instructions for Mac, see Running OpenWPM natively on macOS.

We do not run CI tests for Mac, so new issues may arise. We welcome PRs to fix these issues and add full support and CI testing for Mac.


  1. WebDriverException: Message: The browser appears to have exited before we could connect...

This error indicates that Firefox exited during startup (or was prevented from starting). There are many possible causes of this error:

  • If you are seeing this error for all browser spawn attempts check that:

    • Both selenium and Firefox are the appropriate versions. Run the following commands and check that the versions output match the required versions in and requirements.txt. If not, re-run the install script. sh cd firefox-bin/ firefox --version


        pip show selenium
    • If you are running in a headless environment (e.g. a remote server), ensure that all browsers have the headless browser parameter set to True before launching.
  • If you are seeing this error randomly during crawls it can be caused by an overtaxed system, either memory or CPU usage. Try lowering the number of concurrent browsers.

Docker Deployment for OpenWPM

OpenWPM can be run in a Docker container. This is similar to running OpenWPM in a virtual machine, only with less overhead.

Building the Docker Container

Step 1: install Docker on your system. Most Linux distributions have Docker in their repositories. It can also be installed from For Ubuntu you can use: sudo apt-get install

You can test the installation with: sudo docker run hello-world

Note, in order to run Docker without root privileges, add your user to the docker group (sudo usermod -a -G docker $USER). You will have to logout-login for the change to take effect, and possibly also restart the Docker service.

Step 2: to build the image, run the following command from a terminal within the root OpenWPM directory:

    docker build -f Dockerfile -t openwpm .

After a few minutes, the container is ready to use.

Running Measurements from inside the Container

You can run the demo measurement from inside the container, as follows:

First of all, you need to give the container permissions on your local X-server. You can do this by running: xhost +local:docker

Then you can run the demo script using:

    mkdir -p docker-volume && docker run -v $PWD/docker-volume:/root/Desktop \
    -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix --shm-size=2g \
    -it openwpm python3 /opt/OpenWPM/

Note: the --shm-size=2g parameter is required, as it increases the amount of shared memory available to Firefox. Without this parameter you can expect Firefox to crash on 20-30% of sites.

This command uses bind-mounts to share scripts and output between the container and host, as explained below (note the paths in the command assume it’s being run from the root OpenWPM directory):

  • run starts the openwpm container and executes the python /opt/OpenWPM/ command.

  • -v binds a directory on the host ($PWD/docker-volume) to a directory in the container (/root). Binding allows the script’s output to be saved on the host (./docker-volume/Desktop), and also allows you to pass inputs to the docker container (if necessary). We first create the docker-volume direction (if it doesn’t exist), as docker will otherwise create it with root permissions.

  • The -it option states the command is to be run interactively (use -d for detached mode).

  • The demo scripts runs instances of Firefox that are not headless. As such, this command requires a connection to the host display server. If you are running headless crawls you can remove the following options: -e DISPLAY=$DISPLAY -v /tmp/.X11-unix:/tmp/.X11-unix.

Alternatively, it is possible to run jobs as the user openwpm in the container too, but this might cause problems with none headless browers. It is therefore only recommended for headless crawls.

MacOS GUI applications in Docker

Requirements: Install XQuartz by following these instructions.

Given properly installed prerequisites (including a reboot), the helper script in the project root folder can be used to facilitate working with Docker in Mac OSX.

To open a bash session within the environment:

./ /bin/bash

Or, run commands directly:

./ python
./ python -m test.manual_test
./ python -m pytest
./ python -m pytest -vv -s


Note that OpenWPM is under active development, and should be considered experimental software. The repository may contain experimental features that aren’t fully tested. We recommend using a tagged release.

Although OpenWPM is actively used by our group for research studies and we regularly use of the data collected, it is still possible there are unknown bugs in the infrastructure. We are in the process of writing comprehensive tests to verify the integrity of all included instrumentation. Prior to using OpenWPM for your own research we encourage you to write tests (and submit pull requests!) for any instrumentation that isn’t currently included in our test scripts.


If you use OpenWPM in your research, please cite our CCS 2016 publication on the infrastructure. You can use the following BibTeX.

    author    = "Steven Englehardt and Arvind Narayanan",
    title     = "{Online tracking: A 1-million-site measurement and analysis}",
    booktitle = {Proceedings of ACM CCS 2016},
    year      = "2016",

OpenWPM has been used in over 30 studies.


OpenWPM is licensed under GNU GPLv3. Additional code has been included from FourthParty and Privacy Badger, both of which are licensed GPLv3+.