Lemmy

12576 readers

2 users here now

Everything about Lemmy; bugs, gripes, praises, and advocacy.

For discussion about the lemmy.ml instance, go to !meta@lemmy.ml.

founded 4 years ago

MODERATORS

nutomic@lemmy.ml

How Subscribing Helps Federate (lemmy.directory)

submitted 1 year ago* (last edited 1 year ago) by kiwi@lemmy.directory to c/lemmy@lemmy.ml

5 comments fedilink hide all child comments

I recently setup a new Lemmy instance and was surprised when my feed was mostly empty. I've since learned that a key part of Lemmy's federation is based on a user from your instance subscribing to communities on other instances. Only then, will your instance pull in posts from the subscribed community to your "All" feed.

This means that subscribing to new communities is especially important if you're on a young Lemmy instance since it helps to build out everyone's feed on that instance.

I've found discovering new communities to subscribe to on other instances can be difficult. To help me search for new communities I may be interested in, I tried aggregating as much of the Lemmy fediverse together into a single feed by subscribing to the widest range of Lemmy communities possible. This offers a Lemmy feed that's kind of like reddit.com/r/all. If it's interesting to anyone else, you can find the instance here: https://lemmy.directory.

Hopefully this offers another way to find new communities to subscribe to on other instances.

Here's a better description of my understanding on how Lemmy federates communities and why you might be interested in checking out lemmy.directory: https://lemmy.directory/post/34207.

Hope this helps ease the orientation to how Lemmy federation and communities work.

top 5 comments

sorted by: hot top controversial new old

[–] PriorProject@lemmy.world 0 points 1 year ago* (last edited 1 year ago) (1 children)

This offers a Lemmy feed that's kind of like reddit.com/r/all. If it's interesting to anyone else, you can find the instance here: https://lemmy.directory.

You might want to update your instance link to be https://lemmy.directory/home/data_type/Post/listing_type/All/sort/Active/page/1. Ironically, it defaults to the local feed which is... empty. You could probably make nginx rewrite the homepage to be the all feed as well so the simple/nice url does what you want.

You might also want to add a section to your post writeup about federation load and why Lemmy doesn't do this by default. In a world where Lemmy is very successful and there are lots of instances (many thousands) that subscribe to all communities like you're doing...

An instance hosting lots of niche/unpopular communities is going to be sending thousands of copies of every post/comment off to instances where no one is going to read them. The federated network is small right now, and federation replication load is not a big issue... but the way it remains a non-issue as the network grows is for communities to replicate only when users on an instance are interested in them.
Each instance would have to handle the replication and storage of the entire lemmyverse. I don't know if that's a meaningful workload yet... but again... part of what keeps single-user instances viable in the future as the network grows is ensuring that they only have to fetch and store content that their user is interested in.

Very cool project though. Having an "all" instance sounds like a great service for discoverability. I'd also be interested to see a writeup from you about the hosting requirements. Does it take a lot of CPU/ram/disk to receive the full lemmyverse right now, or is it trivial? Your performance profile could be an interesting leading indicator of replication load as the network grows.

[–] calculuschild@vlemmy.net 0 points 1 year ago (1 children)

Each instance would have to handle the replication and storage of the entire lemmyverse.

Do instances fully replicate and locally store remote subscribed communities? My understanding is they are still solely hosted on the original instance; subscribing just opens a window to the community by making your instance aware it exists.

[–] PriorProject@lemmy.world 0 points 1 year ago (1 children)

Do instances fully replicate and locally store remote subscribed communities?

To a first approximation yes, they replicate the posts and comments made after the time of subscription. Images aren't replicated, but posts, comments, votes, and mod actions do replicate.

My understanding is they are still solely hosted on the original instance; subscribing just opens a window to the community by making your instance aware it exists.

I don't know what you consider "a window" to be, and maybe this is a fuzzy description of some steps of community discovery... but it's definitely not a complete or coherent view of how instances interact through federation.

[–] calculuschild@vlemmy.net 1 points 1 year ago* (last edited 1 year ago) (1 children)

Thank you for taking the time to answer. I hope you might be willing to clarify a bit more for me. By "window", I meant just... having access to a remote community via an API gateway, I guess.

I was under the impression that if I try to subscribe to a remote community hosted on lemmy.world from vlemmy.net, that is simply registering the URL of that community into some local directory in my instance, not duplicating the entire community contents into vlemmy.net. And then when I view a thread in that remote community, I am just retrieving the thread data from the host server at lemmy.world straight to my browser, not loading some local duplicate of the thread from vlemmy.net. Seems like it would get out of sync quickly if we are all reading separate local copies of the original.

So based on your answer, I am still misunderstanding something. What is the purpose of all the duplication then? Is it just for local caching purposes? Does this not needlessly drive up the amount of traffic because each instance is frantically trying to keep up to date with every other instance, rather than just letting each instance handle the requests for its own communities?

[–] PriorProject@lemmy.world 1 points 1 year ago* (last edited 1 year ago)

So based on your answer, I am still misunderstanding something. What is the purpose of all the duplication then? Is it just for local caching purposes?

Pretty much everything in your summary was wrong. I can't reasonably type out a complete primer on federation here, but in short... When the first user on an instance subscribes to a remote community, the subscribing server tells the community-hosting server "send me future updates about posts, comments, and votes and such for this community". The subscribing server then stores them locally.

Why do this actual replication rather than just an API gateway?

That's what federation is. I don't mean to be glib, but there's more than one way to potentially do this stuff, and some systems designers think federation is a useful approach for a variety of complex reasons. Federation involves federated replication, API proxies aren't federation.
If the community-hosting server goes down temporarily, content replicated via federation remains available on the subscribing nodes. This enabled me to read communities homed on lemmy.ml recently when it was struggling and frequently down due to overload.
It spreads the browse workload, which is generally bigger than the replication workload. For example, lemmy.world has commonly has like 4k active users this week. But it only takes one batch of federated replication messages for the community's instance to serve the browse traffic of all those users... then the reads come out of lemmy.world's db. This spreads the browse workload around the federated network.
There are many other reasons to work this way as well, but as noted... at some point you'll just want to Google primers on federated apps/networks. They definitely don't operate according to your current intuition though, and the differences aren't accidental.

Can federated networks have problems where the replication traffic becomes "too much"? Yes, they must be carefully designed to avoid that problem. And for some apps, it makes single-user instances sometimes anti-efficient as the federation workload for the instance can exceed the browse workload from the user, but for multi-user instances the federation messages are a rounding error compared to the browse work. But it's a problem that federated networks generally weather, and the replication offers benefits that on-balance outweigh the costs.