My new video about setting up akkoma fediverse activitypub server has been published https://www.youtube.com/watch?v=GDCvCkSWKak

akkoma is hardfork of pleroma. After pleroma caught itself in web of politics, feature creep and bloat, alternatives emerged. Akkoma is one of them.

Some highlights, comparing to pleroma:

• removed ssh/gopher frontend support
• removed chats
• removed scrobling support
• other stuff slowing down server were chopped off
• added elasticsearch and meilisearch support
• EDIT posts 🔦

In a nutshell, what I do in video:

• order cheapest cloud server
• setup basic security measures
• “boost” it with enormous swapfile
• build & install akkoma
• cloudflare Argo tunnel for security & ingress
• some tweeks instance to lower threat & spam

#selfhosting #akkoma #mastodon #fediverse

@netcat hey... i have yet to watch the video.... this looks nice. have you documented these steps in text? how about an install script or a docker or something? that would make it much easier

@randy yes, text post is coming. Video is better for not much technical people.

@netcat agreed 100%. i understand the appeal. being someone who is contemplating setting up an instance (i am using pleroma right now, cawfee.club) i dont want to go mastodon as it is too expensive to host and maintain. have talked to akkoma guys and they seem nice.
would you be willing to keep the script on github so that people can help improve it?

Text version of video tuttorial

https://gist.github.com/maratsh/0d7a7f715b64c8bd5a5f139837c02926

@randy

@netcat why use cloudflare Argo?

@randy you get cdn, security and it is easy hto configure. With tunneling and external firewall it is hard to break something

@netcat OK. I have not checked this fully out but https://support.cloudflare.com/hc/en-us/articles/115000224192-Billing-for-Argo
This says that there is a per gb bandwidth cost.

1tb costs $105/month. Fine, I usually use cheap vps provider like hostnamaste.com or racknerd.com and they give monthly bandwidth like 3-6TB.

We should do one thing, in the text version, mention the cost aspect.

Why I'm saying this because you can get a racknerd 4gb ram vps for like $30/year with monthly 6tb bandwidth. That is quite a lot for small projects and they shouldn't bother with CDN or security. Maybe If one grows big, then they would have to think but just starting out,

@randy
tunnels are free: https://blog.cloudflare.com/tunnel-for-everyone/

Argo smart routing is different feature. Cloudflare product naming is very questionable.

The point of tuttorial is to give good starting point for newcomers with possibilty to grow. One of the serious problems in fedireverse is security and performance, that is why starting with CDN and tunnel is good idea.

As soon as instance will get more traction, traffic load and attention of bad and good actors, scaling and optimizing question will come.
Cloudflare traffic is quite cheap - if you get TBs of traffic, you may as well start monetizing or open donations.

That is also why akkoma was chosen by me. Akkoma could be scaled in and out .

Sure, you can find better options for upfront charges. I've used hetzner because i know them. I see VPS you linked are more US related and I'm EU based.

I would not commit for a year now, because idk my usage for next year.

@netcat cool. I read that link and I "know" about free cloudflare CDN but Argo is paid. Anyway, do you have any specific idea about the scalability of pleroma/akkoma?
Like if we have a server with 500 daily users for example, what kind of vcpu/ram are we looking at?

My understanding is you can't do horizontal scaling by adding more nodes and load balancing so if you have a 4vcpu/4gb ram server, your only chance to grow is by getting a bigger server? Is that correct or am I with out of date info?

@randy both tunnel and cdn have free tier. “Argo” is strange product umbrella covering 2 lousy products. I just double checked and i have zero billed $ in my cf account for last month.

Scalability is good question. For real data it is better to ask author and operator. From architecture standpoint is good choice of bringing optional elasticsearch, stripping down unneeded features.

Sure you can scale out horizontally. You can bring more elixir web apps, postgresql replicas (for scaling reads), use elasticsearch(very scalable) for search and s3 compatible backend for media.

My concern is write scalability of postgresql. Hovewer, before your instance will grew out of capabilities of big postgresql server, you will have many other non-technical “scaling” issues and you may want to just create more instances and bubble them (which will ultimately solve write scalability issue. Another issue is federation peer-to-peer http connection pool / queue scalability.

My expectation is that 500 DAU will be handled on single instance, just by changing VPS size to (?) and tuning postgres for new instance.