Works fine for me. Havent had any issues at all. I'm only running a few services for me and my gf.

There is one public facing service hosted on the same vps as the pangolin stack and that gets a fair bit of traffic, but everything runs fine

How well does it run on a shared CPU VPS with the recommended specs?

Pangolin can do exactly all of this, but it does require a vps

If it's just for your own usage tailscale is probably the easiest solution. It create a tunnel between you and your server.

If you want to share your services with friends and family and not everyone want to use tailscale then I'm using pangolin on a VPS and I'm quite happy with it.

I do have a static public IP from my ISP but I prefer to not have DNS records pointing directly home, I'm more comfortable with it passing through a public VPS

I'm behind CGNAT with my ISP. I have a VPS which I use to tunnel traffic back to my home network with Wireguard. That gets terminated on the outside of my router and specific ports get forward in from there. Works well enough.

Tailscale also may be an option.

Yes, I would recommend to setup a wireguard tunnel between your vps and homeserver.

Then you can just expose whatever port your want on the VPS and forward the traffic to your server. I use nginx to forward the traffic to my server, but there is different reverse-proxy you could use. Bonus is that your don't expose your home IP

A Cloudflare tunnel would bypass NAT. I don't know enough to recommend for or against it.

These look great, thanks!

Maybe something like https://owncast.online/ or https://github.com/bluenviron/mediamtx

You could set it up to record and maybe even use a hook and some cli signal client to send a message with the stream url to one of your friends?

i dont know exact software, but you could maybe set up an rtmp server (nginx) at home, either with ports open or vpn home (tailscale) and stream video to the server from your phone (does that exist?). then have ffmpeg constantly looking for content on the rtmp server and saving it to a shared place like nextcloud or syncthing the folder to the trusted friend.

nah I think it's just to appease the Orange Ape. Other then China AI companies wouldn't care if people ran models in a different country, after all they would still make money just via different part of the bubble

What stops someone from reselling within the US. They didn't sign the agreement with Micro Center.

Something tells me this is the AI companies not wanting people to run local AI models in other countries.

Also imagine some clown saying what you can and can't do with your £4k+ GPU! At that far above what they're meant to retail for it should be delivered by helicopter and come with some complimentary lines.

What is the point of that, does the US need to keep consumer cards in the hands of random citizens for some reason? UK here and I could just buy one and not have that restriction surely.

Papers, please!

We are a fucking joke of a country at this point.

Not if you don't live in the US. Mind you, it'll still cost as much as a small car so its not exactly straight forward.

thank you!!!!!

I'm 100% biased towards this. It's really cool to see a system get frankenstiened or frankenstein a system from code. It's like getting to leap frog from all of their work without needing to ever explicitly knowing or working with each other.

The real deciding factor isn't whether to automate or not but what is the expected technical skills of who expect to contribute or maintain it the future.

So I personally layer levels to it prefering the common automation where applicable. So tofu for infra and spin up. Ansible for OS config and network equipment. Kubernetes for everything else with a fallback to tofu for some apps that have no k8s API but a good tofu provider.

With that people can actually piecemeal what they want and where they want it if some underlying assumption changed (it's no longer bare metal or aws, it's Ubuntu or SuSE now, it's RKE2 or it's GKE now etc)

That said for bare metal I actually skip tofu and do a cloud native bare metal management service like TinkerBell or Metal3, and that's really only if it's worth it.

The other, when you see it work it's beautiful, thing about the approach. GitOps. Watching what was a thousand shadow IT ops become PRs through automated testing and review is really magical. Coming from a IT back ground for life saving infrastructure the stress difference between the old hat way and this is really hard to understate.

This is truly an amazing guide on building resilient systems. It's so common for people to ignore human based failure domains too!

That said some of the early recommendations suggest fully mapping dependencies such as system accounts, payment accounts, etc. I've looked at that from the "asCode" side but always felt quite lacking. With digital twin modeling of an orginization being a highly academic work (how do you truly capture the reality right?) and smart contracts such DAOs being a heavy tool for enforcement. I was just wondering if you have recommendation for anything or templates?

If not I might try to templatize something based on those lessons myself just to see if there is an quick start way for people to do this.

By Autistici/Inventati ? Maybe some people from them were involved, yes.

Pretty sure that whole thing was generated by AI

That also means resisting the temptation to make ourselves indispensable. Being the person everyone has to call when something breaks can feel like a sign that we are contributing something important, but [...] the goal should be to make ourselves increasingly unnecessary by making the knowledge we carry increasingly available to other people.

What a beautiful sentiment. Very well said.

To my chagrin, I've seen the "private sector" learning this lesson very well, with the practice of codifying configurations of machines and networks in "infrastructure as code" documents. Though these still require special training to decipher and apply, they make people (or workers) much easier to replace.

Although, I'm not sure I want reproduce the patterns I've learned from the corporate world, I'm curious if there's merit to using some of this technology to make infrastructure more easily "spin-up-able. " Or maybe SH hardware will vary too much to make something like OpenTofu useful in a non-cloud setting.

I thought this was an article, but I'm glad I checked it out. It looks like a really very well put together course (too comprehensive to call it tutorial). With guides for people who just want to learn part of the stuff. Had no time to dig deeper, but if you (want to) selfhost and are active in building resilience go have a look.

Thank you!