Pulling the data when a user requests a post/comment (with a cooldown/cache for popular posts) isn’t any more or less scalable
That would definitely be less scalable. That would entail pulling every single time a user views a post or comment. That’s simply not feasible. There are far, far more views of content than there are posts, comments and votes.
Also what about stuff that isn’t seen? What if nobody is logged in or nobody looks at the New sort? You need the votes before you even show the user anything, otherwise you can’t sort the votes.
But if it has to be a push model, why doesn’t feddit.dk push the votes it knows about along with the rest of the data?
This has been explained elsewhere in the thread, see feddit.dk/post/7628338/10255563
catloaf@lemm.ee 1 week ago
True, but that’s why I mentioned a cache or cooldown. Once every two minutes is plenty, unless Lemmy really blows up and we have hundreds of instances trying to fetch a very popular post.
You have a point about new sort, but you could approximate it by sorting what’s known to an instance. It’s not ideal, but it’s at least something. Maybe it would make sense to push just that feed, or to fetch a subset periodically.
I read that comment tree, but it doesn’t answer my question. If someone on Mastodon likes a post on feddit.dk, I don’t see any reason feddit.dk can’t communicate that to lemm.ee when I go look at it.
SorteKanin@feddit.dk 1 week ago
That won’t work, because it would have to be once every two minute for every single comment and post forever.
I have no idea what you mean by this - you can’t fetch a feed, that’s not at all how the data is organised.
The problem is this: How does lemm.ee know that the like that feddit.dk claims is from mastodon.social actually is from mastodon.social? What prevents feddit.dk from just fabricating a like from mastodon.social? Currently there is no nice mechanism for authenticating such a forwarded like. A malicious instance could send loads of likes claiming they come from other instances.