Sock Puppet(eer) Browser.

What is this?

This is a high-performance proxy for Chrome so that you can drive many simultaneous Chrome browsers easily and efficiently.

When you connect on ws://127.0.0.1:3000 as your "CDP Chrome Browser URL" URL it will always spin up a new fresh Chrome instance.

This project is the OpenSource'ed browser back-end for the amazing opensource web page change detection project.

When a request for a new Chrome CDP starts, this software will launch an individual isolated-ish Chrome process for just that request (This is a Chrome CDP "Proxy")

It is based on the excellent https://github.com/Zenika/alpine-chrome, and we add our own wrapper to launch individual chrome instances on demand.

When ever something requiring puppeteer connects via ws://.. it will spin up a new Chrome browser instance and connect you through (proxy you through) to that Chrome's DevTools connection.

It also handles throttling, scaling, and accepting extra Chrome settings on the connection query.

Under-the-hood it is a simple Python websockets wrapper using a puppeteer image, so that we can be sure that all the basic configuration required for Chrome to work will function well.

Why do I need this?

This provides a Chrome interface to applications that need it, usually for example as required when using Playwright - Playwright will launch a node instance and start issuing CDP (Chrome protocol) commands to drive the actual project. So you need this project.

(Playwright gives a high-level command set, which talks to node, that node then does the low-level CDP commands to drive Chrome directly)

It is also more efficient to not need that extra node process like with some other systems (you would end up with two node processes).

playwright -> node -> [sockpuppetserver] -> CDP protocol todo the browser business

Because this method is always built ontop of the latest puppeteer release, it's a lot more secure and reliable than relying on projects to invidually update their Chrome browsers and configurations.

You can skip the whole python -> node mess by using https://github.com/pyppeteer/pyppeteer and talk to this container directly.

How to run

wget https://raw.githubusercontent.com/jfrazelle/dotfiles/master/etc/docker/seccomp/chrome.json
docker run --rm --security-opt seccomp=$(pwd)/chrome.json -p 127.0.0.1:3000:3000 dgtlmoon/sockpuppetbrowser

seccomp security setting is highly recommended https://github.com/Zenika/alpine-chrome?tab=readme-ov-file#-the-best-with-seccomp

Statistics

Access http://127.0.0.1:8080/stats or which ever hostname you bind to, use --sport to specify something other than 8080

{
  "active_connections": 158,
  "connection_count_total": 8383,
  "mem_use_percent": 46.9,
  "special_counter_len": 0
}

You can also add this to your fetch and access 'special_counter_len' at the /stats URL, this is good for adding at the end of your scripts so you know the actual script ran all steps.

        try:
            await self.page._client.send("SOCKPUPPET.specialcounter")
        except:
            pass

Debug CDP session logs

Sometimes you need to examine the low-level Chrome CDP protocol interaction, enable ALLOW_CDP_LOG=yes environment variable and add &log-cdp=/path/somefile.txt to the connection URL.

Then the log will contain the CDP session, for example:

1712224824.5491815 - Attempting connection to ws://localhost:56745/devtools/browser/899f78ce-e7c8-4ad1-b8c9-a7aa449a93ef
1712224824.5528538 - Connected to ws://localhost:56745/devtools/browser/899f78ce-e7c8-4ad1-b8c9-a7aa449a93ef
1712224824.5529754 - Puppeteer -> Chrome: {"method": "Target.getBrowserContexts", "params": {}, "id": 1}
1712224824.553542 - Chrome -> Puppeteer: {"id":1,"result":{"browserContextIds":[]}}
...

Tuning

Some tips on high-concurrency scraping and tuning where you have a lot of chrome browsers running simultaneously

• Understand different Chrome command line options https://github.com/GoogleChrome/chrome-launcher/blob/main/docs/chrome-flags-for-tools.md and specify them on the connection URL
• Set your inotify values higher https://stackoverflow.com/questions/32281277/too-many-open-files-failed-to-initialize-inotify-the-user-limit-on-the-total
• Don't burn out your disk!! Mount the path for --user-data-dir as a RAM Disk/tmpfs disk ! This will also help to speed up Chrome

On a Intel(R) Xeon(R) E-2288G CPU @ 3.70GHz (16 core), it will sustain 150 concurrent browser sessions with a load average of about 65-70 (about 3-4 browsers per CPU core it means).

Most of the CPU load seems to occur when starting a browser, maybe in the future 1 browser could processes multiple requests.

Future ideas

• Some super cool "on the wire" hacks to add custom functionality to CDP, like issuing single commands to download files (PDF) to location dgtlmoon/changedetection.io#2019

Have fun!