| @fungicide:matrix.org | lists.o.o is taking forever to respond to web requests. i see uwsgi processes eating up all available processors, and trying to grab /server-status locally on the server with curl has been waiting over a minute for a response even | 17:21 |
|---|---|---|
| @fungicide:matrix.org | ah yep, all worker slots are full | 17:22 |
| @fungicide:matrix.org | WWWWWWWWWWWWWWWWWW_WWWWWWWWRRWWWWWRWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWRWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWRWWWWWWWWWWWWWWWWWWWWWWWWWW | 17:22 |
| @fungicide:matrix.org | i can't wait until the ai craze crashes and burns, at this point i don't even care if it takes the world economy with it, this insanity can't go on forever | 17:23 |
| @fungicide:matrix.org | seeing as how we've been maxing out swap on that server courtesy of the webui, i don't think it's safe to add more slots in apache unless we resize the vm | 17:25 |
| @fungicide:matrix.org | where resize probably means replace, and then we get new ip addresses and a delivery reputation reset | 17:26 |
| @fungicide:matrix.org | though maybe we could run hyperkitty/postorius on a separate vm from mailman-core? | 17:26 |
| @fungicide:matrix.org | that would probably be less ridiculous than turning the current server into a smarthost | 17:28 |
| @fungicide:matrix.org | getting simultaneously crawled by chatgpt, amazonbot, baidu, semrush, claude, and a bunch more | 17:34 |
| @fungicide:matrix.org | oh, bing's scraping right now too | 17:34 |
| @fungicide:matrix.org | and applebot | 17:35 |
| @fungicide:matrix.org | mj12bot | 17:35 |
| @fungicide:matrix.org | ahrefsbot | 17:36 |
| @fungicide:matrix.org | it's like everybody decided saturday was a good day to redownload all the content from that server | 17:36 |
| @fungicide:matrix.org | dotbot, googlebot, bytespider, petalbot... just watching these scroll by and it's clear that between the search engine and llm training crawlers we're not serving this content to humans with web browsers any more, they're apparently a fraction of a fraction of a percent of the requests | 17:42 |
| @fungicide:matrix.org | and now it's almost completely idle | 18:09 |
| @fungicide:matrix.org | .........................______________________________W_____________W_____........................._____W___________________......................... | 18:09 |
| @clarkb:matrix.org | fungi: note that often times those those named bots aren't the problem as they tend to respect things liek our crawl delay value | 20:08 |
| @clarkb:matrix.org | fungi: what I find useful is to sort all the apache requests by user agent to get a sense of relative request counts. On gitea we've also identified particularly expensive requests and filter on those. Maybe we need to identify similar requests for mailman? | 20:08 |
| @clarkb:matrix.org | we do set a crawl delay of 2 here. If we decide these properly named bots are the issue and they are respecting crawl delay we might increase that value | 20:09 |
| @clarkb:matrix.org | I agree we shouldn't add more slots without changing anything else. Another option may be to try and have apache cache things? I think the main risk there is you'll go to an archive link for a thread and new thread responses may not appear until the cache is cleared? that may be ok considering that list traffic tends to be low and slow like a smoker? | 20:11 |
| @clarkb:matrix.org | anyway we can think about this more on Monday | 20:11 |
| @fungicide:matrix.org | oh, yeah disk cache might help, even a very, very large one. we have plenty of storage | 20:22 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!