PSA: Something to keep in mind when inviting people to a mastodon server you're on is that if your server is quite small, taking on even one multi-thousand+ follower account might result in a significant increase in the cost and difficulty of running that server. Even just the migration itself might be a costly event.

People with a lot of followers should be really be careful themselves about moving to a server, and check in with the admins about if they're willing to take on the increased load, but it's also good to keep that in mind when suggesting your server that you have no part in sysadmining to your favorite fedi celeb.

@megmac I don’t have time but I’d love to dive into the scaling factors of Mastodon.

It feels like the there are some real scary things that could keep admins awake at night.

@linux_mclinuxface mastodon specifically has some really unfortunate design decisions that make it pretty bad at scaling in a non-"throw hardware at the problem" way. Gts and some other fedi software can do better, so it's not inherent to the network, afaik.

@megmac seems like it.

TBH, throwing hardware at problems is sometimes more straight forward than making folks go down paths only experts can follow, but when it’s merely_joining_ an instance causes pain and cost, that feels like compounded anti-patterns.

Like, I understand if your presence on a particular instance causes pain because of traffic or what not… but joining? Yowza

@linux_mclinuxface well, the joining part is mostly if you migrate. That causes a huge queue of new network events to inform other servers of the change and refollows. Most of the time problems with mastodon scaling are of this kind of big deep job queue sort afaik. But then also every post needs to be sent out to other servers so once you're there, if you're active you can trigger it again when you post.

@megmac @linux_mclinuxface Is joining actually any worse than posting? Every time I post I get a big spike in server usage, it's just life with the fediverse. I guess maybe migrating is O(followers) and posting is just O(follower-instances)?

Still, given Mastodon is pretty heavily queued, this seems like it would self resolve reasonably if configured properly.

The biggest issue I had was posting a video on the edge of the "automatically fetch+cache" size threshold. That had my networking maxed out for half an hour. A caching CDN fixed that though, and it's like $7/month, so not exactly breaking the bank.

But yeah obviously if you have 10k followers, don't join an instance that has accounts with 500 followers max without talking to them first.

@lina @megmac @linux_mclinuxface From what I remember of one of the "server maids" here talking about keeping the thing running smoothly, it still takes some tuning to keep the thing responsive to "real-time" requests while working through a large queue, though that was quite a while ago so I have no idea if the defaults for that have improved at all.