#lemmy/#kbin has a problem that #mastodon hasn’t even attempted to solve; groups and what happens when they get popular.
#Communities, #groups, #magazines, whatever they are called are implemented as #Actors in #ActivityPub. They are basically just *very* popular users who boost a *lot*.
You can’t just distribute them across instances the way normal actors do. Whichever server hosts @technology@lemmy.ml or @technology@beehaw.org is going to get HOSED on the regular.
So there is no way to horizontally scale?
The network can actually scale quite well thanks to the fact that other instances will act as mirrors of communities!
No. The “single source of truth” is the instance hosting the community. If it goes down the community itself goes down with the ship. The only way to prevent it is to have a IT infrastructure that can provide redundancy
having a redundant system is feasible (I’m just a dev, not an architect so don’t take my words for granted) but it have to be designed and putted together … and prices are gonna skyrocket
Lemmy / the fediverse isn’t designed this way, but it could be. There are certainly systems that share diskspace and are multimaster and keep stuff as long as someone is interested in it(i.e. accessing the data). I really start to think added to the lemmy / fediverse servers should be something like what freenet used to do in terms of hosting content.
If it’s just a temporary outage, whatever the mirror has received prior to the outage will be shown to users on that other instance but only local interactions for that instance will update it, when it comes back up, things like votes and comments will be synchronized again across all of the instances.
For permanent outages, the community will just need to be started again on a new instance.
But they could pick up where the now defunct community left off, right? Like, the cached copy from another server could be imported on a new server elsewhere?
That functionality doesn’t currently exist, but migration of communities is something that’s being actively talked about for development.
Out of curiosity, how is kbin’s magazine system designed to avoid this problem?
It’s not