I imagine there’s excitement for the increase of activity but worries about the potential toxic side of Reddit coming along too.

I’d especially be interested in the Lemmy devs’ opinions.

  • darkfoe@lemmy.serverfail.party
    link
    fedilink
    arrow-up
    13
    ·
    1 year ago

    I fired up my own personal test instance so I can experiment with figuring out ways to reduce bottlenecks on the sysadmin/devops side - used to run the various PHP forums back in the day, so hoping to pass on some knowledge eventually.

    I figure the toxic side(s) will gravitate towards instances that will tolerate their behaviour which is easier to deal with. Mods will be busy for a little bit though, and I wouldn’t be surprised if registrations closed for a bit on some of the bigger instances so they can catch up if they don’t just fall flat over on the heavy days. But, lots of smart folks trying to prep for this.

    • Valmond@lemmy.ml
      link
      fedilink
      arrow-up
      4
      ·
      1 year ago

      Any idea what hardware specs you need to run an instance? Like for 100 users, 1.000, 10k etc?

      Or the hardware lemmy.ml runs on and the userbase?

      • darkfoe@lemmy.serverfail.party
        link
        fedilink
        arrow-up
        11
        ·
        1 year ago

        It’s still a little unknown at this time what you need to handle X number of users, beyond a few hundred. Beehaw.org is pretty open about what they’re using though in their financial statements if you’re curious, but there’s of operational optimization being tried out to see what’ll help.

        The stack is: postgres, pictrs, lemmy (Rust), lemmy-ui (nodejs), and nginx. RAM usage isn’t too bad, but so far I see CPU and disk I/O (pictrs) as the limitation. Websockets are being removed which was another hurdle - would cause nginx worker threads to max out and drop instances off.

        I’m on a 6$/month droplet as a reference for my single user instance and I’m subbed to a boatload of communities. So far I’m not having problems, but I made a 2GB swapfile for safety if RAM somehow spiked. CPU usage for me tends to spike when a community is being loaded for the first time due to image processing, but otherwise things are pretty idle.

        • TheDude@sh.itjust.works
          link
          fedilink
          arrow-up
          5
          ·
          1 year ago

          I’m looking forward to the increase in traffic tbh. I have setup a pretty beefy instance with a ton of monitoring on it so that hopefully after the wave I can create a nice write up on what it would take to scale lemmy in the future. I’ll keep everyone updated with the results!

          • darkfoe@lemmy.serverfail.party
            link
            fedilink
            arrow-up
            3
            ·
            1 year ago

            Yeah, this is a golden moment for those of us who like to learn from sudden heavy load on server software! There are not very many teachable moments like this out there, so I’m trying to soak everything up for work experience

        • Valmond@lemmy.ml
          link
          fedilink
          arrow-up
          2
          ·
          1 year ago

          I have an i-5 6core dell sitting so why shouldn’t I spin up a node?

          I’m mostly worried about maintenance and it breaking down one day, how do you deal with that in a good way?

          • darkfoe@lemmy.serverfail.party
            link
            fedilink
            arrow-up
            2
            ·
            1 year ago

            Regular backups should do the job. It’s all run in docker instances with mapped volumes, so you can just backup those contents regularly and roll-back worst case if things completely pooped out. Otherwise maintenance isn’t really much worse than a normal webserver - great for learning Linux CLI if you’re not already familiar.

            No reason you shouldn’t spin up a node though! The more the better - lets load spread out.