75
flamingos-cant (hopepunk arc)
Feddit UKSorry about the long downtime everyone. What happened is sometime around 20:00 yesterday the instance went down. I wouldn't be able to ssh into the server until the morning, so I tried restarting the server through our hosting's web interface, but this didn't work.
When I got home I found the reason Lemmy wasn't working was because the database had suffered some corruption. Thankfully, the only lost data was 3 posts, but some data got duplicated in random tables that shouldn't be and this wasn't the easiest thing to track down.
Why did the server go down in the first place?
I don't actually know the real reason, what I suspect happened is that the frontend got taken down by the rampant scraping we've been subject to. To put it into prospective, just look at the jump in the size of our web logs:
These are compressed plain text, like it's unreal.
But back to the server, I think when I restarted the server to get everything back up, docker killed the database before it had shutdown properly hence the corruption.
Anyway, it's sorted now. It's going to be a bit quiet until we've caught up with the activities we've missed, so enjoy the quiet while you can.
Glad to see we are back. Those access logs are crazy.
I’m sure you will fine a lot of Chrome/12X, Chrome/13X and Chrome/14X user agents there. Seems to be the favourites of attackers/scrapers recently.
On another note, do the server has a profile somewhere like mastodon or something we could check in case it goes down in future.
Thank you so much for the work you do!
Thanks Flamingos!
Rotate your logs!!!
Looks like they're already rotated daily.
Anyway by my understanding it wasn't the logs that caused the problem, they just demonstrate the extent of the sudden increase in requests by bots/scrapers.
All hail flamingo-cant!
More like flamingo-can!
Hails!
Yeah, those gigabyte sized access logs is something we have seen for weeks.
Anubis has also gotten largely ineffective, and yesterday I had some time to sample some of the access log IPs and it is clear that this is a coordinated scraping attack from the Huawei ASN that additionally employs residential proxies mostly in Pakistan, Vietnam and Brazil.
I don't really have a good idea how to deal with it, but it is a complete PITA.
Well done! We knew you would come through, honest!
On the scraping.... That's crazy. Is there anything that can be done to protect things?
Using Anubis. But i'm not technical, anyzay that's a good new :)
Thanks for all your hard work!
Thank you for your efforts to fix it!
Just wondering, is there a backup place where things like this can be communicated? Blue sky or mastodon for example?
There's a matrix channel here: https://matrix.to/#/#feddit.uk:matrix.org
Do you have anything in place for detecting and mitigating DOS attacks? I wonder if you were to grep the logs by IP address you'd see the offending IP addresses. Though if you're using a load balancer or some other kind of proxy and don't have IP address forwarding configured, all requests might come through as the same IP address.
some data got duplicated in random tables that shouldn’t be
I'd also check that there's plenty of disk space spare, given that the log files are so large, as I've seen environments start to behave unusually once disk space has ran out, including SSH locking up.
There's a setting in Lemmy that only let's logged in users make API requests. I'm going to activate that over night for now.
Awesome work, flamingos-cant. Appreciate all you do for us.
Phew! I was worried for a bit.
I notice that feddit.org use Anubis for bot mitigation. Is this something we do too?
Thank you for fixing it and keeping things going :)
Great job. Although I think my eyes have been widened as to how big the fediverse actually is... I found all sorts of stuff that I still don't understand, communities I didn't know existed, tools that I don't know how to use and a lot of communities that have warnings i can't get past. Along with a whole heap of other stuff that I don't know how to get into.
Thanks for fixing it
Those logs are insane! 1.6GB compressed?! Any idea what the load/throughput was like during that time?
