Hacker News Reader

Comment by huimang | Hacker News Reader

I'll use this opportunity to mention Pleroma [1], my preferred alternative to Mastodon & others. It's much more lightweight than Mastodon.

1: https://pleroma.social/

View on HN

My years of scar tissue with RabbitMQ have given me a healthy suspicion of OTP ( https://git.pleroma.social/pleroma/pleroma#otp-releases-reco... ), and I would never want to be responsible for keeping it alive and healthy. For comparison, while I'm sure people also have strong opinions about rails and sidekiq, at least there is a non-trivial body of existing blog and SO posts about it

That's fair. I selfhost Pleroma as a single-user instance, so I couldn't tell you about running it at scale. I went with it because it could run on a raspberry pi a few years back whereas mastodon was too resource-hungry.

I've been working on Elixir sites for years now and have never had an issue with OTP underneath, although none of them (yet) have been high-demand. The reliability and performance characteristics have been top-notch. Is it possible that with the mass influx of Elixir users into the Erlang space since the RabbitMQ days, OTP has made leaps and bounds of improvement?

I thought OTP was supposed to have a very strong operations/reliability story ("nine nines of reliability in telecoms" and all that)?

For what it's worth, I remember someone shared Pleroma code in the Elixir forum and some veterans there said it was a very non-standard way of organizing an Elixir codebase.

It could be possible that Pleroma is not very idiomatic Elixir / OTP.

That's odd given OTP's focus on stability and reliability.

RabbitMQ has been my message bus of choice for over a decade. Configuration is a total PITA, but it's a one-time thing. Otherwise, it's been rock solid and I'll happily continue to use.

Note also that recently a fork called Akkoma[1] sprung up. If one is looking to self-host, it might be worth looking into this vs Pleroma (I don't have a horse in this race, was just doing some research on different ActivityPub servers).

[1]: https://coffee-and-dreams.uk/development/2022/06/24/akkoma.h...

More alternatives:

https://codeberg.org/fediverse/delightful-fediverse-apps

I'm partial to WordPress with the ActivityPub plugin and would highly recommend pushing for inclusion of core AP functionality into your content management system of choice (so everyone can just stand up and participate on their own domain name, like media/gov/your local school etc).

Mastodon itself is fun but the real value in the fedi is going to be diversity of interoperable (or even incompatible) implementations. There (on a long enough timeline) will be a successor to Masto's fedi domination.

I also prefer pleroma, plus they just have larger instances (like poa.st, which had more active users than mastodon.social before Elon bought twitter, although I'm unsure if that's the case now). Sites like Gab started off on Mastodon but found it was too hard to scale and moved off it.

That's neat. Do you happen to know if it's possible to migrate from Mastodon to Pleroma without having to recreate the account and setting up redirects? I already host my own Mastodon instance since a few years back, and don't want to do another redirect if I can help it.

To my knowledge you would have to create a new account and set up redirects, since it's a new server. Migrating is definitely a pain point in the fediverse. That's why I ended up going with a single-user instance that I selfhost.

Yeah, I mean migrating the software of a single-user instance, not migrating between instances, sorry if that was unclear.

So I currently run a single-user instance of Mastodon. Wondering if it's possible to start running a single-user instance of Pleroma instead, without having to change the username+domain?

My understanding is in theory yes. But in practice things might be more complicated.

Like merely keeping the same name and domain is easy enough. Just shut down the old system, start the new one, create an account with the same name. Of course you would need to refollow everyone. I'm not sure what happens to your followers. Your new server will not know they are followers, so unless this triggers mastodon to try to automatically refollow, I think they are left in a half-broken state. You definitely lose all your old content. All in all, this is not really a great option.

For a full migration where in theory everything works, you could try to follow the linked process to migrate to Pleroma 2.0.5, and then obviously just upgrade to latest following plemora documentation for upgrading. description of migration process: https://icyphox.sh/blog/mastodon-to-pleroma/ (This is based on the following migrator: https://gitlab.com/soapbox-pub/migrator). I have no idea if it works with newer mastodon minor releases.

> TL;DR: Mastodon’s Sidekiq deferred execution jobs are the limiting factor for scaling federated traffic in a single-server or small cluster deployment. Sidekiq performance scales poorly under a single process model, and can be limited by database performance in a deployment of the default Dockerized configuration.

Well, that in short explains why Pleroma/Elixir is a much better stack to build this in. You don't need Sidekiq (https://sipsandbits.com/2020/08/07/do-we-need-background-job...) and you aren't stuck with a GIL (Global Interpreter Lock).

Just to note, Pleroma also uses a job queue: Oban. https://hexdocs.pm/oban/Oban.html

Oban is great. (I subscribe to Oban Pro.) It's also seemingly very superior to Sidekiq-like solutions for job management.

But you don't need that. It's certainly a nice-to-have.

> Well, that in short explains why Pleroma/Elixir is a much better stack to build this in

As someone who has extensive experience in both elixir and ruby, and mastodon itself, no this is doesn't.

I read that a Pleroma instance can handle at least 10-50x the traffic of a Mastodon instance running on the same hardware. Based on the reports from other sites that have switched from Ruby to Elixir, this claim rings true. Are you claiming otherwise? I mean, the lack of a GIL alone (thank you, functional language with immutable data!) should give a pretty massive performance improvement

I like Elixir but as of the latest Ruby version, you can use ractors. They aren't bound by the gil

Did not hear of them. Interesting. Also requires a code rewrite to take advantage of them, and still considered very experimental.

How much memory/cpu does Pleroma need for a single user instance?

I spun up a test server of Pleroma recently. Following just a few people, 0 people following me AFAIK, I did join a few relays so my federated timeline is populated. It is currently showing 360MB RSS and the CPU load maxes out at ~50% of a single vCPU.

I've run a single-user pleroma server since January 2019. I ran it on a digitalocean droplet with 2 GB RAM, and then an AWS instance (t3.small, also 2 GB) for the better part of two years; I recently switched to a t3.medium. Beam's current process size: 250MB resident, uses 5-7% CPU.

I discard remote posts after 120 days, and my database (using PostgreSQL 15, no RUM indexes) takes up just over 5.4 GB (as reported by `pg_database_size`). It's grown a lot faster in the past couple of weeks. I follow 245 accounts.

Reply on news.ycombinator.com