Cloudflare has been down, which has affected the entire internet: https://www.youtube.com/watch?v=UgJ5qnCkIOk
We don’t use cloudflare. So that’s not the problem.
I think ti was turned off because of massive waves of scrapping bots (probably trying to train AI). Git repository web interfaces tend to have lots and lots of links (to every commit on every file) which acts like a tarpit for these bots, they just keep endlessly following all the links, until they overload the server.
There was some talk about cgit in the chat a few days ago.
cgit got overloaded by the scraping bots and the automatic restart also restarted Gerrit,which made the situation even worse.
Therefore,cgit is temporarily disabled so that at least Gerrit works until a permanent solution is found.
@kallisti5 said that GitHub - w4/rgit: 🏯 blazingly fast web frontend for git repositories could be a good replacement for cgit that can handle a lot more traffic,but it needs more memory at the server.
cgit has been down for some days already,totally not related to Cloudflare which we luckily don’t have to rely on.
I saw that a couple of times already…figured someone was working on the website. Did not realize it was another LLM scrapers scenario again. Thanks for the info!
Does anyone know how long cgit is planned to stay down? (ping @kallisti5)
I initially thought the switch to rgit will be almost instant,but it’s been almost two weeks now.
If it’s expected to take a few more weeks or even months,I can temporarily set up a rgit (or similar) installation on one of my servers.
I have the spare capacities available,so that would be a matter of a few minutes of work here.
The cgit server is lacking ressources needed for rgit, Haiku Inc will have to aprove a bigger server.
Do we need it?
I’m of course just one data point, but as long as existing links continue working, I’m OK with an external mirror for web browsing. For me, something that redirects old links would be enough.
I much prefer cgit or something like rgit to stuff like githubs or codebergs interface. cgit just shows so much more and is usually way faster (and doesn’t choke and dies on our tags)
It was supposed to be an easy switch to rgit, but it turns out rgit doesn’t scale that easily to a git repository with 25 years worth of commits (it starts by trying to load them all in ram, and there’s no space for it). Kallisti5 has been looking into patching it, I think.
Not really, we could just send each other hrev numbers and commit hashes, but people will then end up using github or some other thing not in our control. Which means:
- Use of closed source software where not necessary,
- Relying on a 3rd party to provide a service, which they may later decide to close (it already happened to us with Berlios and to other projects with google code project hosting). If they close it, no way to set up a link redirect from their urls,
- Also a risk that someone may hijack the mirror and inject commits not made by the Haiku team in there
So, yes we do need an online view of the sourcecode, and yes it should be self-hosted. Maybe we can put it behind a login wall?
Meanwhile, I have setup something good enough for me. Feel free to use it if also works for you.
A tag-decorated history. Does not update immediately, has no graph and may lose commits if a branch is merged instead of using our typical rebase + fast forward.
A redirector. Close enough for common links. If you know what you are doing and don’t mind the certificate mismatch, you can even edit your hosts file and you won’t have to copy-paste the links.