I wanted to have plex handled as part of swarm, but:
a) it doesn't fucking matter becuase I don't intend to have fault
tolerance for plex; i'm not even sure it can run well in a clustered
way
b) its just much much easier to get compose working rather than
swarm. One day maybe i'd like to do that so that its a single stack to
deliver, but meh.
Pantheon is actually really nice, but unusable for me on nixos because
of how I use ssh agent. KDE provides ksshaskpass as a total
replacement for pinentry, it seems, and Pantheon just uses pinentry.
On nixos, pinentry appears to be Fucked Up, and changing pinentry
binaries to other nixos packages doesn't result in anything besides
more of regular pinentry, which doesn't allow copy paste, which is a
NON STARTER.
FUcking linux.
This is a nice step forward, for me. But there are some challenges
here I need to fix.
The first, and biggest one, is that DO rate limits API requests per
minute to ~250. If I'm doing even a small amount of changes and
iterating a lot I will run into issues with that just due to recrods;
each record pull appears to be its OWN CALL. God.
With that in mind, an easy next step seems like splitting the domain
management into their own submodules to prevent unnecessary calls.
For now, i'm just happy to have this managed in code :).
Create 2 different subfolders:
- jowj dot net owns / will own DNS for jowj.net and its subdomains
- Bikeshed owns my provisioning VMs against my proxmox cluster
env_creds.sh is just a simple way to export secrets from `pass` to my
local environment.
The last of my fucking around with Nix and Flakes prior to killing the
nixos portion of my k8s project. This was just too hard to figure out,
honestly. Docs around nixflakes suck. They appear to be the future! I
believe the hype! But I can't figure out DICK ALL.
Committing, will probably rip out once I'm fully over on PM.
1. Fuck around with Goof configuration. I think ultimately this never
went anywhere, and I'll probably rip it out, but I forgot to commit it
at the time so i'm including it just to be safe.
2. Home services and home net refactor. Smaller clean ups
3. Removal of wg stuff.
Deploy-RS will follow module imports, but seems to do some fucking
DUMB SHIT when it comes to actually building the whole package. I
found that unless I defined the user in the core configuration.nix
file, regardless of other imports, I would lose access to the users
halfway through the deploy.
I'm not totally sure why. I never tried setting the import inside the
configuration.nix, but I /was/ setting it inside the fucking
flake.nix, so who knows.
This way does make some sense, it feels safer to have my deploy user /
root defs inside the core config file so there's never a possibility
of losing access remotely, but still. What a fucking pain. Morph
doesn't have this problem! Morph is good! What are you doing deploy-rs!!
This appears to be working out pretty well.
Alice is now a required user for all machines to manage remote
deployment requirements.
Probably should figure out a reasonable refactor in order to move
Alice into a "secrets" sort of thing, so that I'm not giving away user
info for free on git, but i'm not that worried about it.
