| 18 Oct 2022 |
 Linux Hackerman | the advantage of this approach is that the secrets can theoretically not get copied to the nix store (in practice they do, because I use Flakes) | 09:05:53 |
| Linux Hackerman | Shyim: only on the machine from which you're deploying | 09:06:01 |
 Shyim | AHHH. cool | 09:06:43 |
| Linux Hackerman | but I think I'll be switching to agenix or sops-nix at some point in the future, partly because this approach is colmena-specific and I'm not sure I want to stick with colmena | 09:06:57 |
| Linux Hackerman | and partly because the secrets are persisted unencrypted on the target machines' disks, which I'd also prefer not to do | 09:07:26 |
| Shyim | I currently join ssh and just do nixos-rebuild so. tbh: I don't care if it's in nix store. I just don't want to have them on Github publicly accessible as my nix files are public :D | 09:08:14 |
| Linux Hackerman | both agenix and sops-nix copy the encrypted secrets to the store as part of the system closure, and they're encrypted to a key that lies on the machine (often the SSH host key) to be decrypted only to a ramfs at runtime | 09:08:37 |
| Linux Hackerman | then again, having both the encrypted secrets and the unencrypted SSH host key on disk isn't much different from having the unencrypted secrets on disk I guess | 09:09:27 |
| Linux Hackerman | the next-level thing would be to add a TPM and measured boot into the setup, but I'm nowhere near getting there lol | 09:10:05 |
 @v:meowy.tech | In reply to @linus:schreibt.jetzt
then again, having both the encrypted secrets and the unencrypted SSH host key on disk isn't much different from having the unencrypted secrets on disk I guess the various permissions on the file also matter, the nix store is world readable but the ssh host key probably isnt | 09:10:16 |
| Linux Hackerman | right, but none of these approaches make the unencrypted secrets world-readable | 09:10:36 |
| Linux Hackerman | that's definitely off the table for me ^^ | 09:10:56 |
| @v:meowy.tech | My setup with a hashicorp vault is probably overkill, but when set up it does work quite nicely I have to say | 09:11:40 |
| Linux Hackerman | I kind of want to have something like that but I also don't want to set it up and maintain it x) | 09:12:05 |
| Shyim | I feel that too xD | 09:12:33 |
| Shyim | more services more pain | 09:12:42 |
| Linux Hackerman | Some services are pretty painless to maintain. Vault is not "some services" as far as I can tell ^^ | 09:13:21 |
| Shyim | :D | 09:13:41 |
| Shyim | so then I go to test this out. I have at work a benchmark cluster of 10 vps with ansible and it hurts setting up that always. So hope with colmena it will be better :D | 09:14:20 |
| Linux Hackerman | oof | 09:14:52 |
| Linux Hackerman | Do they need to be long-lived? It may be more convenient to build disk images and use those rather than maintaining long-lived stateful servers | 09:15:26 |
| Linux Hackerman | at that scale | 09:15:36 |
| Shyim | and randomly apt-get servers are down. or some weird ansible modules fail to install | 09:15:39 |
| @v:meowy.tech | In reply to @linus:schreibt.jetzt
Some services are pretty painless to maintain. Vault is not "some services" as far as I can tell ^^ Yeah it can take a bit of fiddling to get right, also the need to unlock the vault manually after a restart can be a bit of a bother | 09:15:48 |
| Shyim | It's just for benchmarking for 2 hours. automated in ci. Testing that the software scales | 09:16:23 |
| @v:meowy.tech | I maintain quite a few NixOS configurations with colmena without too much issue, it is definitely one of the better deployment tools for bigger sets of machines in my experience | 09:20:39 |
| @v:meowy.tech | 
Download image.png | 09:20:41 |
| Linux Hackerman | "null" is that the actual hostname? | 09:21:01 |
| @v:meowy.tech | In reply to @linus:schreibt.jetzt
"null" is that the actual hostname? yeah it is haha, it's my laptop | 09:21:10 |
| Linux Hackerman | I'm going to write the deployment tool of my dreams one day | 09:23:14 |
