Conversation

Notices

1. Embed this notice
   Poppy (Back from tha dead) :neofox_flag_trans: (radmin@limepeeps.perchinup.top)'s status on Tuesday, 19-Mar-2024 17:36:44 JST Poppy (Back from tha dead) :neofox_flag_trans:

   • Sexy Moon
   • Moffin'tosh

   @moffintosh@den.raccoon.quest @Moon@shitposter.club From what I've heard from Moon and others, it's most likely database problems. The issue of databases in Pleroma and Akkoma rotting is widespread, even to the point of becoming a meme.
   
   It's honestly awful to see, though. I know that instance death is just what happens in internet culture (as someone who killed an instance herself). Yet, there's something intriguing about an instance that started near the beginning, especially when its community is still thriving today. They've seen when GNU Social was still prevalent or when Pleroma was still starting out. Many people haven't even experienced Fedi when Twitter was only shitty and not dying. Other instances at the time, such as https://mastodon.cloud, seem like shells of their former shelves (plus, it did you dirty), and instances like https://mastodon.social have let themselves become a Twitteroid hellscape. So, seeing an instance like that crumble apart feels like a major part of Fedi might also be gone.

   • Sexy Moon likes this.
   • Embed this notice
     Sexy Moon (moon@shitposter.club)'s status on Tuesday, 19-Mar-2024 17:40:01 JST Sexy Moon

     in reply to

     • Moffin'tosh
     @radmin @moffintosh This is pretty accurate. Basically the problem is that as the database grows, performance gets worse. There's not a good explanation for it because it shouldn't be an issue from everything I've read but the fact remains that the database is ultra fast when it's nearly empty but slogs when it gets big.
     
     I'm running high end nvme drives and it's still uncomfortably slow.
     
     Another aspect is when you have all this data scrapers are merciless and it the performance problem multiplies until the the site literally goes down from load.
     
     I wanted to keep this database as a living archive of the early Fediverse. It still will be to some extent but I can't keep all the data around forever anymore.
   • Embed this notice
     Sexy Moon (moon@shitposter.club)'s status on Tuesday, 19-Mar-2024 18:17:41 JST Sexy Moon

     in reply to

     • Neko McCatface v2023 :verified::makemeneko:
     • Moffin'tosh
     @roboneko @moffintosh @radmin yeah I am gonna do that possibly.
   • Embed this notice
     Neko McCatface v2023 :verified::makemeneko: (roboneko@bae.st)'s status on Tuesday, 19-Mar-2024 18:17:42 JST Neko McCatface v2023 :verified::makemeneko:

     in reply to

     • Sexy Moon
     • Moffin'tosh
     @Moon @moffintosh @radmin
     
     > I wanted to keep this database as a living archive of the early Fediverse. It still will be to some extent but I can't keep all the data around forever anymore.
     
     what about splitting it off into an archive database so that it doesn't tank the main site?
   • Embed this notice
     Sexy Moon (moon@shitposter.club)'s status on Tuesday, 19-Mar-2024 20:25:39 JST Sexy Moon

     in reply to

     • cjd
     • Moffin'tosh
     @cjd @moffintosh @radmin there is not sufficient tracing in Pleroma for me to easily detect what's going on. Well, not at my skill level anyway.
   • Embed this notice
     cjd (cjd@pkteerium.xyz)'s status on Tuesday, 19-Mar-2024 20:25:40 JST cjd

     in reply to

     • Sexy Moon
     • Moffin'tosh
     Did you flip on slow query logs to find out what's causing the problem?
     
     Also re scrapers, how are they manifesting?
     Is it via the AP endpoints or client endpoints?
     
     In either case, I suspect they can be rate limited to the point of not being a problem..
   • Embed this notice
     Sexy Moon (moon@shitposter.club)'s status on Tuesday, 19-Mar-2024 20:26:22 JST Sexy Moon

     in reply to

     • cjd
     • Moffin'tosh
     @cjd @moffintosh @radmin I recently turned on rate limiting so we'll see.
   • Embed this notice
     cjd (cjd@pkteerium.xyz)'s status on Tuesday, 19-Mar-2024 20:27:03 JST cjd

     in reply to

     • Sexy Moon
     • Moffin'tosh
     Check log_min_duration_statement on postgres, it will give you the slow queries.
     Sexy Moon likes this.
   • Embed this notice
     cjd (cjd@pkteerium.xyz)'s status on Tuesday, 19-Mar-2024 20:30:50 JST cjd

     in reply to

     • Sexy Moon
     • cjd
     • Moffin'tosh
     What I'd do is:
     1. Log the slow queries
     2. Run a slow query manually to confirm it's always slow - not just a server load issue
     3. EXPLAIN ANALYZE the slow query and paste the result on the fedi and ask other people what the esoteric garbledigook means and what CREATE INDEX commands I ought to run
     Sexy Moon likes this.
   • Embed this notice
     iced depresso (icedquinn@blob.cat)'s status on Sunday, 24-Mar-2024 10:48:01 JST iced depresso

     in reply to

     • Sexy Moon
     • Moffin'tosh
     @Moon @moffintosh @radmin i guess the design i wanted to use where events were stored in something like haystack and loosely tracked by db records was the right call :neofox_up_paws: