cross-posted from: https://poptalk.scrubbles.tech/post/3263324
Sorry for the alarming title but, Admins for real, go set up Anubis.
For context, Anubis is essentially a gatekeeper/rate limiter for small services. From them:
(Anubis) is designed to help protect the small internet from the endless storm of requests that flood in from AI companies. Anubis is as lightweight as possible to ensure that everyone can afford to protect the communities closest to them.
It puts forward a challenge that must be solved in order to gain access, and judges how trustworthy a connection is. For the vast majority of real users they will never notice, or will notice a small delay accessing your site the first time. Even smaller scrapers may get by relatively easily.
For big scrapers though, AI and trainers, they get hit with computational problems that waste their compute before being let in. (Trust me, I worked for a company that did “scrape the internet”, and compute is expensive and a constant worry for them, so win win for us!)
Anubis ended up taking maybe 10 minutes to set up. For Lemmy hosters you literally just point your UI proxy at Anubis and point Anubis to Lemmy UI. Very easy and slots right in, minimal setup.
These graphs are since I turned it on less than an hour ago. I have a small instance, only a few people, and immediately my CPU usage has gone down and my requests per minute have gone down. I have already had thousands of requests challenged, I had no idea I was being scraped this much! You can see they’re backing off in the charts.
(FYI, this only stops the web requests, so it does nothing to the API or federation. Those are proxied elsewhere, so it really does only target web scrapers).
Hmmm. I mean my instance is small, almost all of my traffic comes from federation. There will be like 50 requests in the log to forward some upvotes, and then one or two from a user or crawler. Most of them seem to behave, it was just Alibaba and Tencent who did proper DDoS attacks on me. I’ve blocked most of their ASNs and so far it seems to do 100% what I was trying to do. I suppose the minor stream of requests from South America and other places of the world isn’t humans either, but they mostly read articles and that’s kind of a cheap request on PieFed and it’s not a lot. I’ll keep an eye on it, maybe that’s an alternative solution. Though currently it’s a manual process to look up the address ranges and write the firewall rules. It should probably be automated in a way before random people adopt it.
I thought so too with mine, divide your traffic though into the API vs Web requests. If you have a small instance most of your traffic should be federation traffic hitting the api endpoints, and posts. Scraping traffic will be GETs and requesting web content, not API content. That’s what I noticed, that the vast, vast majority of my traffic wasn’t federation at all, but web scrapes. Granted my instance has been around for almost 3 years now and I’m sure most of the bot farms know I exist.
Interesting. For me on a 16 month old PieFed instance, mostly used by me, 95% of the total incoming traffic is federation. With some countermeasures activated. In the last 3 days I got 450k requests from Lemmy instances, 7k from PieFed, 10k from Mastodon, 30k from PeerTube and 30k have a browser User Agent string so those are users or crawlers. That’s roughly 5%. 5k of those have bot or crawler in the User agent, that’s 1% of my requests. And I did 1k requests from my home IP address. Fun fact: 52% of all incoming traffic is just to federate with lemmy.world.
I’m not surprised, but I have noticed a lot of the bots are fake user agents as well. For example, for my instance I know for a fact no one else was using my instance when I was testing this, and I kept getting User Agent requests from Safari. Which I would be surprised if my users were using Apple too, but knowing they weren’t on was a huge driver. I want to dive in and see how Anubis knows this, or if they were just tested and failed or didn’t bother to complete the challenge. So I’m curious when you remove POSTs and api/federation endpoint calls what your traffic looks like
I don’t know how you’re reading this, in case you didn’t know, a lot of browsers have the word “Safari” in their user agent string.
My Vanadium browser identifies as: “Mozilla/5.0 (Linux; Android 10; K) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/142.0.0.0 Mobile Safari/537.36”
So there’s Safari in it, but it’s a Chromium based browser on Android. And none of the information is correct. It’s Android 16, not 10 and Vanadium or Chromium isn’t even in it. Also doesn’t use Mozilla’s engine nor AppleWebkit. It is version 142.something however.
My Firefox (LibreWolf) identifies as: “Mozilla/5.0 (X11; Linux x86_64; rv:144.0) Gecko/20100101 Firefox/144.0” that doesn’t include anything else.
But the desktop Chromium does the same thing again: “Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0.0.0 Safari/537.36”
That’s absolutely true, looking at my logs there are definitely some weird ones:
Mozilla/5.0 (Windows NT 6.1; WOW64; rv:2.0b7) Gecko/20100101 Firefox/4.0b7Which… if that’s firefox is very out of date, and Windows 6.1 is Windows 7. Which I could believe that people are posting from Windows 7 compared to 10 or 11, but sus. A lot of them are kind of weird combinations. Anubis auto flagged that one as
bot/huawei-cloud, but I’m really curious as to how or why it did. It’s not authentic, that’s for sure, but how does it know it isn’t?Another one is
Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/101.0.4951.64 Safari/537.36 Edg/101.0.1210.47, is completely valid, but from an older Windows version and Edge, like if it was a snapshot from a few years ago. Idk it’s all very interesting. That one was also flagged asbot/huawei-cloud.Well, pretty much all browsers fake their user agent strings for a long time now. That’s for privacy reasons. Because malicious actors can single you out with all that information, the combination of IP, operating system and version and exact version of the browser and used libraries. Also they don’t want to advertise if you skimped on updates and are vulnerable for exploits. And they fake more because via Javascript a website can get your screen size, resolution and all kinds of details. That’s business as usual. These pieces of information still serve some purpose, occasionally… But they’re more a relic from the distant past when the internet was an entirely different place without an advertisement and surveillance economy.
And sure, shady bots fake them as well. It’s rare to see correct information with anything. The server to server communication in the Fediverse for example advertises the correct name and version number. Maybe some apps as well if they’re not concerned with servers being malicious.
Fair, all of that. You know as someone who really likes pristine data, I really hate that actors used such basic clean data against all of us. What should have been a simple compatibility check was completely co-opted and now essentially serves no purpose.
Anubis I know looks at much more than that, you’ve made me curious to go read through their code. It’s working for my instance, but I also see it warrants more research as to how it decides
I think the main point is to load a JavaScript and make the client perform a proof-of-work. That’s also the mechanism they advertise with.
The additional (default) heuristics seem to be here looks like they match for some known user agents and header strings and there are some IP address ranges in there. And some exception to allow-list important stuff.