Conversation

Notices

1. Embed this notice
    (mint@ryona.agency)'s status on Saturday, 25-May-2024 06:54:39 JST 
   Apparently effective_io_concurrency beyond 20 also produced diminishing results.
   https://portavita.github.io/2019-07-19-PostgreSQL_effective_io_concurrency_benchmarked/
   • Embed this notice
      (mint@ryona.agency)'s status on Saturday, 25-May-2024 06:58:17 JST 

     in reply to
     Decreased it from 200 to 100, also increased random_page_cost from 1.1 to 1.25. I'd take a couple extra milliseconds the queries have to run over having random 500s.
   • Embed this notice
     lauralt (lauralt@ryona.agency)'s status on Saturday, 25-May-2024 07:11:22 JST lauralt

     in reply to
     @mint
     
     Things seem to be improving, nice work mint.
     Why would such a problem appear as abruptly? Was db query time increasing over time and it just hit the threshold of not being able to serve them any more?
      likes this.
   • Embed this notice
      (mint@ryona.agency)'s status on Saturday, 25-May-2024 07:15:08 JST 

     in reply to

     • lauralt
     @lauralt Don't think so, the disk seems to be in good health even if a bit slow, and there's plenty of space as well now that I've booted salon off to an $20 SSD. I've been experimenting with postgres config for a while, sometimes not changing some corresponding values (e.g. it seems like you need to decrease work_mem when increasing max_connections). Now I did it again after yet another pg_repack session, but paying somewhat closer attention this time.
   • Embed this notice
     munir (munir@fedi.munir.tokyo)'s status on Saturday, 25-May-2024 07:17:30 JST munir

     in reply to

     • lauralt
     @mint @lauralt wait what, you run ryona, spw, cyberia, sal*n all by yourself from one server?? :akko_aaa:
   • Embed this notice
      (mint@ryona.agency)'s status on Saturday, 25-May-2024 07:26:38 JST 

     in reply to

     • lauralt
     • munir
     @munir @lauralt Agency, salon and accela run on the same server at home, with the former two proxied to a frantech VPS; camp and rawr onion run on the same frantech VPS directly. plagu.ee is sort of a staging server where I test changes before commiting it. Plus I got root access to nukie's server and a couple more I'm not going to mention.
     ✙ dcc :pedomustdie: :phear_slackware: likes this.
   • Embed this notice
     munir (munir@fedi.munir.tokyo)'s status on Saturday, 25-May-2024 07:31:05 JST munir

     in reply to

     • lauralt
     @mint @lauralt damn props to you mint 👏
     I can barely manage my own instance
     Maybe you can help me fix the blockbot thingy sometime? :002:
   • Embed this notice
      (mint@ryona.agency)'s status on Saturday, 25-May-2024 07:33:56 JST 

     in reply to

     • lauralt
     • munir
     @munir @lauralt It should work, and is working just fine on at least nine instances, so idk, recheck the config.
   • Embed this notice
      (mint@ryona.agency)'s status on Saturday, 25-May-2024 07:36:43 JST 

     in reply to

     • lauralt
     • munir
     @munir @lauralt Forgot about zetsubou and route 200::/7, but those are pretty much testing instances as well.