Can’t post images because they’re too big so here’s imgur: https://imgur.com/a/Fm52ZTB

Edit: lemmy.ml and lemmy.world seem to have come back, I’m just a bit worried that it’s another one of those hacks.

Edit 2: Most of those I’ve tried came back. reddthat.com and sh.itjust.works seems to still be down

  • hitagiEnglish
    10•1 year ago

    There is a GitHub issue on it and I experienced the exact same thing with my instance. A timeout occurs and the only way to fix it is to restart it seems. Like everyone else, it’s strange that it all happened at the same time.

    • zalack
      5•1 year ago

      It’s not that strange. A timeout occurs on several servers overnight, and maybe a bunch of Lemmy instances are all run in the same timezone, so all their admins wake up around the same time and fix it.

      Well it’s a timeout, so by fixing it at the same time the admins have “synchronized” when timeouts across their servers are likely to occur again since it’s tangentially related to time. They’re likely to all fail again around the same moment.

      It’s kind of similar to the thundering herd where a bunch of things getting errors will synchronize their retries in a giant herd and strain the server. It’s why good clients will add exponential backoff AND jitter (a little bit of randomness to when the retry is done, not just every x^2 seconds). That way if you have a million clients, it’s less likely that all 1,000,000 of them will attempt a retry at the extract same time, because they all got an error from your server at the same time when it failed.

      Edit: looked at the ticket and it’s not exactly the kind of timeout I was thinking of.

      This timeout might be caused by something that’s loosely a function of time or resources usage. If it’s resource usage, because the servers are federated, those spikes might happen across servers as everything is pushing events to subscribers. So, failure gets synchronized.

      Or it could just be a coincidence. We as humans like to look for patterns in random events.

      • hitagiEnglish
        1•1 year ago

        This probably makes more sense although the issue I was experiencing earlier had similar logs as the issue I linked and others have commented on it too around the same time. I’m guessing they’re related.

        • ShadowEnglish
          2•1 year ago

          The original issue is just a symptom of all database threads being tied up. People just don’t know how to follow an error message to the root cause.

          The real source of the issue is db locking from triggers and cascading deletes on a major user change.

          My report in https://github.com/LemmyNet/lemmy/issues/3649 has the offending query.