SeleniumBase: Python APIs for web automation and bypassing bot-detection

https://github.com/seleniumbase/SeleniumBase

by seleniumbase • 2 years ago

177 67 2 years ago

I've been working with scrapers quite a lot. I started with python requests, then to scrapy, then selenium, then selenium via undetected_chromedriver, and once that started being detected during a chrome update about a year ago, I've switched over to seleniumbase. It got by undetected, but to get it working with pre-downloaded drivers, I had to look into the code. I have never, and I mean never, in all my python years, seen such a horrible mess of code. We are talking 1000lines long methods, with 20-30 different flags and branches Just horrible. I have since switched to Playwright, which seems to be also undetected, and offers a much saner interface.

seleniumbase16• 2 years ago

SeleniumBase modifies the webdriver so that it doesn't get detected when used alongside the CDP stealth mode and methods. It'll download chromedriver for you. Not sure what you mean by the multiple branches, as there's just the primary one. What 1000-line methods are you referring to? By "flags", do you mean the different command-line options available? As for Playwright, they aren't undetected: See https://github.com/microsoft/playwright/issues/23884#issueco... - "Playwright is an end-to-end testing framework, where we expect you test on your own environments. Bypassing any form of bot protection is not something we can act on. Thanks for your understanding." On the contrary, SeleniumBase is OK with bypassing bot detection: https://github.com/seleniumbase/SeleniumBase/blob/master/exa...

cyanmagenta• 2 years ago

Not the commenter, but “multiple branches” in this context is referring to if/else statements in the code, not source-control branches. Similarly, “flags” is referring to function arguments like a boolean “is_original.” More generally, they are just saying that the code has long, complicated, bug-prone functions.

That said, I just spent a few minutes browsing the SeleniumBase repro, and honestly it didn’t seem that unusual to me. Would be interested in seeing a specific example of what the commenter had in mind.

mdaniel• 2 years ago

rather than point-by-point rebuttal as the sibling requests, I think this sums up the coding style pretty well: https://github.com/seleniumbase/SeleniumBase/blob/v4.33.11/s...

seleniumbase16• 2 years ago

That method came from code that I accepted in a PR from December 31, 2019: https://github.com/seleniumbase/SeleniumBase/pull/459 Not a true representation of most of the code today.

harrall• 2 years ago

That's not amazing code but that's not that bad. In the grand scheme of things, that's not code debt that would ever seriously make my life any harder.

edm0nd5• 2 years ago

Not sure if you have explored rolling captcha solving services into your code. Its easy as fuck and you can do it in a few lines of code. Check out DeathByCaptcha or AntiCaptcha. It's like $2.99 per 1,000 successfully solved captchas.

I guess my point is, you dont have to be undetected nor write 1000 lines of code to scrape or do whatever you are needing to do always. Saved me a ton of headaches and time when captchas are involved.

mintzworld7• 2 years ago

SeleniumBase is free, open-source, can bypass CAPTCHAs with a few lines of code, and it works from the free tier of GitHub Actions.

pryelluw3• 2 years ago

Enterprise Python code. Somehow ends up being worse than Java enterprise code. I’m too used to it at this point.

seleniumbase16• 2 years ago

The "Python vs Java" debate is probably one for a different Hacker News post. :)

bryanrasmussen3• 2 years ago

Maybe I am just a cynic but I would expect Playwright to be detected when using Chrome, I mean I would expect it was to the benefit of Google to make that happen for the sake of making reCaptcha detect bots better.

That's actually why I've been scrapping my Playwright automation (because I expect I will encounter problems even if hasn't happened yet, cynical and paranoid) and moving towards writing a browser extension to automate Firefox.

Basically my use case is automating tedious things for myself not running bots at scale, so that's why it is imperative not to get caught being "not human", because then risk account problems.

robertlagrant• 2 years ago

How can Google make that happen? Playwright's made by Microsoft. It can use Firefox as a browser as well as Chrome.

lyu072822• 2 years ago

https://github.com/seleniumbase/SeleniumBase/blob/master/sel...

That's.. that works?? :D

seleniumbase16• 2 years ago

That patches chromedriver, (which gets renamed to uc_driver), but patching by itself isn't enough to bypass bot-detection. SeleniumBase also sets specific Chrome options and modifies methods to use the Chrome Devtools Protocol.

lyu072822• 2 years ago

I was more astonished that you could just search and replace a string in a PE/ELF binary without breaking everything, but I take your solution over recompiling chrome anytime. Awesome job, very well done!

mintzworld7• 2 years ago

The "CDP Mode" used for bypassing CAPTCHAs and bot-detection has it's own ReadMe within SeleniumBase: https://github.com/seleniumbase/SeleniumBase/blob/master/exa... And there's a recent YouTube video with live demos: https://www.youtube.com/watch?v=Mr90iQmNsKM

cruffle_duffle• 2 years ago

As somebody who is now on the “need to scrape a website to get my customers data for them” side of the fence… I get the reason bot detection exists. If you want people to not scrape, offer API’s that allow customers or their software to log in using oauth and let their software / LLM agent grab their data for them.

chii• 2 years ago

If you want people to not scrape, offer API’s

many sites want to prevent scrapers because they don't want their information aggregated - things like price lists and product availability etc.

I know groceries sites do this, to prevent customers from knowing price histories of products. They want to raise prices, then offer a discount to make it seem like the discount is legitimate.

mintzworld7• 2 years ago

On the topic of scraping grocery sites, here's an example of bypassing bot-detection on Albertsons: https://github.com/seleniumbase/SeleniumBase/blob/master/exa... (A demo of that is in https://www.youtube.com/watch?v=Mr90iQmNsKM)

theanonymousone• 2 years ago

Is it demonstrably better than Playwright in bypassing Cloudflare measures? I have some scraping projects and the "cat and mouse game" (what's the right expression here?) got so much energy that I finally went with an external dedicated scraping service. It doesn't feel right that some scrapers are considered friendly (e.g. Google), while smaller ones are vilified..

seleniumbase16• 2 years ago

There are live demos on YouTube of SeleniumBase bypassing CAPTCHAs (if you want see first): https://www.youtube.com/watch?v=Mr90iQmNsKM

diamondfist252• 2 years ago

Playwright failed to bypass cloud flare, while seleniumbase did for me.

I was scraping my own oai data

coppsilgold4• 2 years ago

Is there a reason why the crawling and browser automation people don't just patch the browser to be controlled with no possibility of detection?

The web page is heavily restricted in what is can access through various interfaces and you can feed it anything you want by patching the browser. Once you do that the problem becomes just simulating a legitimate user to a sufficient degree.

I wonder if that's what's already happening with CDP and ReCAPTCHA and hCaptcha - the two services mentioned that are strong and a problem. Are they detecting the "Stealth" or is it just the lack of user activity and reputation? Is CDP by itself detectable by some means?

seleniumbase16• 2 years ago

Patching chromedriver is a lot easier than patching the browser. Plus, if you're just using a regular Chrome browser for the automation, then there's nothing to patch. Automated CDP calls aren't detectable if they don't leave any trace of automation activity. However, since Google created CDP, they might have ways of detecting automated CDP in ways that other services cannot.

coppsilgold4• 2 years ago

What about faking mouse movement from inside the browser? PyAutoGUI is not the right way to be doing this for interacting with JavaScript that has no hope of interrogating user operating system GUI interactions.

And it seems like it would be important to try and adopt user-like mouse movement since JavaScript has access to this information.

ghxst5• 2 years ago

The reason in my experience is that there's a high barrier of entry for most devs when it comes to setting up an environment for Chromium and a workflow for patches that still allows you to quickly and easily pull in and apply upstream changes whenever a new Chromium version releases.

In reality, if you know how to use CDP correctly and you have control over the environment that you run the browser in, you have to make very few browser patches.

What I mean with using CDP correctly is that, yes it is detectable to a certain extent but it comes down to things like enabling Runtime domain for example which you can easily mitigate in your own solution but is something that libraries like puppeteer / playwright often do out of the box (this is where the "stealth" versions of these libraries come in, they will either mitigate by disabling features or use some hacky approaches to instrument the JS that runs on the pages).

Then when you move into an environment that is a lot more stripped down (let's say from your home machine to docker) now you run into A LOT of issues that you definitely are better off fixing with browser patches, however figuring out what those issues are and how to fix them is a huge feat in itself and often will require you to have the ability to reverse engineer things like Cloudflare, Akamai and other anti bot vendors just to know what leaks you still have to patch.

It doesn't help that there is no end to misinformed articles on things like "browser fingerprinting" that you encounter when you try to solve your issues the first time you encounter them, a lot of articles based on nothing but superstition, articles that basically say "proxies are never good enough", "captchas are getting out of hand" that get things wrong and will just eat away at your sanity while trying to debug issues.

This is long enough of a rant already but maybe offers you some insight, if you have any specific questions feel free to ask.

seleniumbase16• 2 years ago

The biggest issue with going from a home machine to a server is that you may lose having a "residential IP address", which is something that you'll want to have in order to prevent automation from being blocked outright. Hence the popularity of residential proxies. However, some servers live in a residential IP space, which makes them optimal for running web automation in. As was partially covered in https://www.youtube.com/watch?v=Mr90iQmNsKM, GitHub Actions appears to live in a "Residential IP space", which makes it a good server choice for web automation.

coppsilgold4• 2 years ago

Why not create a library that you inject into the Chrome process though?

It seems to me that playing a cat and mouse game with these anti-bot systems is unnecessary. Design a system which mimics a legitimate user to such a degree that it's either indistinguishable from an actual user or would produce an unacceptable level of false positives for the detection system. This is not an even playing field, the bot has all the advantages.

For example:

- Enumerate all the possible ways in which the webpage can glean insight into user input/activity.

- Hook all these functions by injecting code into the browser. At a level above and completely inaccessible to anything the web page can do to detect/interfere.

- Create functions that mimic user activities (mouse pathing, aimless mouse wondering, random scrolls, clicks, text selections, etc)

- Feed the outputs of these functions into the functions that you hooked.

- Rip out whatever information you want from the Chrome data structures in memory. Can probably reuse CDP code here.

After all this, the only challenge that would remain is to perfect the input functions that are supposed to mimic a legitimate user. Depending on how sophisticated these anti-bot systems can/will get, you may also need to cultivate user browsing habit profiles to enter advertising/spying databases as real humans.

nullday• 2 years ago

Nice, but I still use puppeteer these days with just rebrowser-patches - https://github.com/rebrowser/rebrowser-patches - works just fine with vanilla chrome

bpiroman• 2 years ago

can it bypass Kasada??

a good example is realestate.com.au

seleniumbase16• 2 years ago

Here's an example that bypasses Kasada on the Hyatt website: https://github.com/seleniumbase/SeleniumBase/blob/master/exa...

diamondfist252• 2 years ago

I was just looking at seleniumbase.

I was scraping my own oai chatgpt.com with playwright, and cloudflare blocked any attempts and the same with selenium and puppeteer. Only seleniumbase got pass it

hildenae• 2 years ago

Because reasons i don't remember, I ended up doing all my scraping through Firefox (i think it was easier to make PDFs?). Is there something like this that works for running (non-headless) Firefox?

seleniumbase16• 2 years ago

You can set `browser="firefox"`, but there's no stealth mode for it.

SeleniumBase: Python APIs for web automation and bypassing bot-detection

Related Stories