NixOS System Operations | 620 Members | |
| About system administration for running NixOS systems in production. Declaratively manage your operations. | Room recommendations: #networking:nixos.org | 170 Servers |
| Sender | Message | Time |
|---|---|---|
| 4 Apr 2026 | ||
| it looks cool | 20:29:11 | |
| it looks like an Ansible with nix instead of yaml, am I getting that correct? | 20:31:34 | |
| I don't think that's correct. I think it would be more fair to compare nix to ansible, but even that would be unfair to both in my opinion. What are you looking to achieve exactly? Perhaps that tells me whether clan is a good fit or not | 20:34:19 | |
| Fair enough currently I'm running a ton of my infrastructure on Kubernetes with gitops using fluxcd, it's amazing because the state in git is the state of my cluster the problem I have is that some applications are not fun to run in Kubernetes, like TURN servers, so I have those in separate VMs which are not tracked by git so I have a notebook with the "state" but nothing guarantees that there will be no state drift I'm looking for a way to move part of my infrastructure that's currently in and outside of Kubernetes to VMs that I manage declaratively. I already run NixOS as my daily driver so it was a no brainer to use it for this kinda thing on the VMs, but I still need a good way to make sure that whatever is in git also represents the state of my machines ideally all machines also use the same build server so that I don't store the same nix build cache on $n VMs :P | 20:39:05 | |
Download as an example, renovate opens pull requests on my Kubernetes repository with commits like this to update certain containers, and I can merge them from my phone and the container updates in real time | 20:43:05 | |
| Ok, yep, I think that you would like clan in this case. I myself am(/was, long hiatus..) building a self hosted cloud with clan with k3s and crossplane. https://git.amarth.cloud/amarth/infra Also in the middle of migrating my personal machines to clan. | 20:45:30 | |
| cool | 20:46:29 | |
| how high/low level is it? I've built a small NixOS POC with colemena where I can deploy a k3s cluster to multiple machines at once, but you're basically writing a normal NixOS system configuration yourself and you also manage the bootstrapping and secret management of nodes yourself | 20:47:29 | |
| The deploying to multiple machines is what a clan service is for, this is what I was builing with my "cloud machines" I have 4 machines, 1 control node and 3 compute nodes. I made a clan serive which wraps k3s config with two roles in that service. Then I have a single instance of that service in which I assign the machines to that instance's roles. You can see all this in https://git.amarth.cloud/amarth/infra/clan.nix If this sounds nice than I suggest you read through the getting started of the clan docs and join clan's matrix room. Plenty of smart helpful people there 😁 | 20:53:54 | |
| I'll check out your repo, thanks :) | 20:56:42 | |
| it sounds amazing | 20:56:45 | |
| * The deploying to multiple machines is what a clan service is for, this is what I was builing with my "cloud machines" I have 4 machines, 1 control node and 3 compute nodes. I made a clan serive which wraps k3s config with two roles in that service. Then I have a single instance of that service in which I assign the machines to that instance's roles. You can see all this in https://git.amarth.cloud/amarth/infra/src/branch/main/clan.nix If this sounds nice than I suggest you read through the getting started of the clan docs and join clan's matrix room. Plenty of smart helpful people there 😁 | 20:58:05 | |
| 5 Apr 2026 | ||
| 06:55:47 | ||
| 21:03:55 | ||
| just found out Clan has the same flaw when onboarding new machines as my own setup (changing IP adressess), feeling kinda defeated right now 😭 | 22:14:31 | |
| What flaw? | 22:34:33 | |
| when you onboard a new server, in the installer it'll have a different IP than after the installation in most dhcp setups | 22:41:04 | |
| Really? Id expect that to not be the case since the mac should be stable, so id expect the router to hand back the same ip | 22:45:11 | |
| Though i havent tested this | 22:45:18 | |
| But ig the router can just decide to give back another one cause why the fuck not. Not sure if anything can really done about that | 22:45:54 | |
| Depends on how the dhcp server and client handle the lease. | 22:46:50 | |
| Well id expect the lease to last :P but ig not | 22:47:09 | |
| I do wonder why the servers switching ips is a problem and if it is a problem, why are they using dhcp with non-static leases at all | 22:47:53 | |
| But, why not disable dhcp if you want static adresses? | 22:47:59 | |
| That seems like relying on finicky behavior | 22:48:06 | |
| Yeah exactly my point, or just static leases. But somehow pin the damn thing | 22:48:23 | |
| There are 3 options imo:
Anything else will be finicky | 22:50:29 | |
| 6 Apr 2026 | ||
| 09:42:52 | ||
| then machine onboarding is still clunky since you'd have to manually assign an ip adress either:
| 09:57:54 | |
| having an overlay network is nice but you'd still need to bootstrap it (and have ways around it when you're messing with its options and it restarts during a rebuild or whateverK | 09:58:36 | |