Conversation

Notices

1. Embed this notice
   Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 29-Apr-2025 17:31:25 JST Haelwenn /элвэн/ :triskell:
   Mastodon still downloading external media and keeping them for ages instead of just being a caching proxy is just so weird to me.
   I think no other fedi server does that.
   
   You go from "Just need the gigabytes for the database" to "I need an SSD for the medias"
   • Phantasm likes this.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 29-Apr-2025 17:49:19 JST Phantasm

     in reply to
     @lanodan I think there's a "cron" job for that now.
     
     I guess it was made like that so multiple application instances could share media between each other, but there are better options for that. An nginx loadbalancer in front of the cluster can serve the media directly without issues. The downside with that is no direct control over the content of the cache unless you pay for nginx (PURGE is not supported). You also avoid the stupidity of most distros now making /tmp and /var/tmp a tmpfs.
     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 29-Apr-2025 18:01:56 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm

     @phnt /tmp being a tmpfs is very old, it's just a fuss now because now enterprise distros are also adopting that, but well most of them where also annoyingly running rm -fr /tmp/* at boot (which with a non-parallelized service manager gets annoyingly slow).

     But /var/tmp should not be a tmpfs, at least to me that's like the whole point of /var/tmp, and well if some distros gets particularly annoying, /var/cache is also a possibility.

     Phantasm likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 29-Apr-2025 18:22:32 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Ⓑⓡⓔⓣⓣ
     @brettm Well to be fair only one I've used other than Pleroma is Honk and I forgot how it deals with media files.
   • Embed this notice
     Ⓑⓡⓔⓣⓣ (brettm@swarm.coiloptic.org)'s status on Tuesday, 29-Apr-2025 18:22:33 JST Ⓑⓡⓔⓣⓣ

     in reply to
     @lanodan@queer.hacktivis.me maybe gts does it as well. For my gts database was inexplicably 5gb (and still growing fast) after running for about 1 month with only 1 account (mine) and purging every 8 hours. Only snac seems to be a reasonable size (mine sits at ~2gb)
     
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 29-Apr-2025 18:27:57 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Ⓑⓡⓔⓣⓣ
     @brettm Yeah, although probably easy enough to patch in.
   • Embed this notice
     Ⓑⓡⓔⓣⓣ (brettm@swarm.coiloptic.org)'s status on Tuesday, 29-Apr-2025 18:27:58 JST Ⓑⓡⓔⓣⓣ

     in reply to
     Honk is pretty cool but doesn't impliment likes. I understand why they don't want to but that's a dealbreaker for me.
     
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 29-Apr-2025 18:31:08 JST Phantasm

     in reply to

     @lanodan Now that you mentioned rm -rf /tmp/*, systemd I think automatically deletes the /tmp/systemd-private-...-<service>-... and /var/tmp/systemd-private-...-<service>-... folders when the service exits/crashes and makes a new one, if PrivateTmp is enabled (RHEL has it enabled). So your temporary files just get deleted anyway.

     nginx crashed? Yeah, your whole cache is now gone. Have fun.

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 29-Apr-2025 18:36:28 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     @phnt Heh lovely to toss things out after a crash, not only it means worse performance after but also… good luck debugging anything.
     Phantasm likes this.
   • Embed this notice
      (mint@ryona.agency)'s status on Tuesday, 29-Apr-2025 18:37:13 JST 

     in reply to

     • Phantasm
     @lanodan @phnt >But /var/tmp should not be a tmpfs, at least to me that's like the whole point of /var/tmp
     Akshually in the Unix world the whole /var is supposed to hold variable data loss of which won't be critical, hence why BSDs keep stuff like DBs and webserver roots in /srv. Linux was the one that chanced the convention.
   • Embed this notice
     Phantasm (phnt@fluffytail.org)'s status on Tuesday, 29-Apr-2025 18:47:18 JST Phantasm

     in reply to
     @lanodan Well, at least you can disable the directive in an override and it isn't enabled by default, but it's still stupid. Especially when it's enabled for services like web servers by the distro maintainers. Someone at Red Hat must have been smoking something very strong.
   • Embed this notice
     Haelwenn /элвэн/ :triskell: (lanodan@queer.hacktivis.me)'s status on Tuesday, 29-Apr-2025 18:54:33 JST Haelwenn /элвэн/ :triskell:

     in reply to

     • Phantasm
     @phnt Well… RedHat lol
     Phantasm likes this.