Everything about Lemmy; bugs, gripes, praises, and advocacy.
For discussion about the lemmy.ml instance, go to [email protected].
I run a single user instance and I subscribe to 47 communities from 12 servers.
Most of the time I see 3 requests per second from the other servers.
I'm running on the cheapest Hetzner VPS with aditional storage, because the 20 GB are not enough for lemmy and the OS.
The load is most of the time at 0.2
Cheapest Hetzner ARM64¹, 72 communities from 14 servers:
¹ CAX11: 2 vCPU, RAM 4 GB, Disk space 40 GB
edit: Weekly CPU usage is also pretty low:
Oh you went for the ARM CPU? I was looking at it but then I thought that there is no ARM docker container for lemmy, was I wrong?
The Lemmy Easy Deploy script available at GitHub supports Arm64 in theory. In practice, some required binary refused to run on my Pi 4 so I think it's still a WIP.
The Lemmy-Easy-Deploy maintainer gave up and created their own docker build for arm64 ;)
Thanks for sharing this!
So how are you making it work?
I am running it in a VM now, using Linux and Docker.
I meant if you're building the ARM version yourself or where are you getting it from?
Docker volumes folder
Started server on July 4th
July 18th: 3.1 GB
July 28th: 5.2 GB
August 1st: 2.8 GB
Today’s volumes are divided between pictrs (1.5 GB) and the database (1.2 GB)
This is for a single user instance, but it seems pretty okay.
How many communities do you subscribe to? I thin that is what will have the biggest impact on the database size.
73 currently. Notably, no meme or "great pic" communities, though. That would probably be relevant for the pictrs volume.
I have ~160 users and have been up about two months. We have subscribers to pretty much every large community and hundreds of smaller ones, plus media heavy ones like the NSFW communities.
Currently sitting at 6.6GB for Postgres (there was an update recently that drastically reduced the size, dunno exactly what it was at before), 54GB on pictrs/media. (I host the media on object storage, specifically Cloudflare R2)
For memory I'm pushing 3GB now, with Postgres taking about 1.5GB. That's higher than it used to be, might be related to the update that reduced the DB size.
CPU: I have a cheap 2 core x86 virtual server/vps - and haven't seen it go over 10% during normal operations...
To answer your question directly though, the most work is OUTGOING federation. So it really depends on how active your communities are/how many unique instances subscribe.
Lemmy can certainly be used like an old school forum with federation disabled in the settings. The developers even made a phpBB-like front-end for it: https://github.com/LemmyNet/lemmyBB
Yea ... but I like the reddit-like interface. It's richer with more options than an old school forum.
Also ... they're running an instance of this somewhere too.
On my instance, the owner said that just the cached text content is something like 25GB.
So it's very storage intensive as it seems Lemmy doesn't delete the cached content.
Some storage optimisation came in the latest version ... it may have been to do with what you describe.
Indeed, especially looking at the other coment with 2.8 GB
The main table that all the growth was in is now split and the cached federated content can be dropped with out much work, and there was talk of tuning it to expire at 3 months.
The content on no.lastname.nz (very small instance) is only about 10-15GB including cached images
one of the achievements of lemmy might be as a nice platform for simply running a forum for whatever community you want all without needing to worry about federation.
I'm interested in this question for the inverse reason: being able to run a federated community on a Lemmy server which is not open-invite
People in the Lemmyverse would be able to use the community as normal, but running the community on its own server would not involve opening the door to registration by randos on that server.
I think that's interesting too!
More broadly, it might become a story of the fediverse as it grows ... whether it makes sense to move on from an architecture where every node is more or less of the same kind (full instance) to one where nodes take on more specialised and perhaps more optimised roles within the network.
Not sure about resource usage but a single instance for a given community mean a single point of failure.
Even with federation the community is bound to an instance, so once the instance goes down the community is down too, at least the federation of the community as far as I understand. So it is already a single point of failure.
Why is that? That shouldn't be necessary, should it?
Ah, interesting. I did not know that, but it makes sense.
Sure. I'm talking about any group of people that want a forum on the internet without bothering about being on the fediverse. Would just spinning up an unfederated lemmy server be a good option for this from a resources/efficiency standpoint?
Yes it would. Another one written in elixir and no federation stuff in is https://github.com/ckruse/cforum_ex