Pim Kunis
ea84627e59
fix: Improve flake outputs yet again refactor: Delete dead code related to hamnet |
||
---|---|---|
docs | ||
flake-parts | ||
kubenix-modules | ||
machines | ||
my-lib | ||
nixos-modules | ||
secrets | ||
.gitignore | ||
.sops.yaml | ||
configuration.nix | ||
container-images.nix | ||
flake.lock | ||
flake.nix | ||
README.md |
nixos-servers
Nix definitions to configure our servers at home.
Acknowledgements
- deploy-rs: NixOS deploy tool with rollback functionality
- disko: declarative disk partitioning
- dns.nix: A Nix DSL for defining DNS zones
- flake-utils: Handy utilities to develop Nix flakes
- nixos-hardware: Hardware-specific NixOS modules. Doing the heavy lifting for our Raspberry Pi
- kubenix: declare and deploy Kubernetes resources using Nix
- nixhelm: Nix-digestible Helm charts
- sops-nix: Sops secret management for Nix
Installation
Prerequisites
Bootstrapping
We bootstrap our servers using nixos-anywhere. This reformats the hard disk of the server and installs a fresh NixOS. Additionally, it deploys an age identity, which is later used for decrypting secrets.
⚠️ This will wipe your server completely ⚠️
- Make sure your have a Secret service running (such as Keepassxc) that provides the age identity.
- Ensure you have root SSH access to the server.
- Run nixos-anywhere:
nix run '.#bootstrap' <servername> <hostname>
Deployment
To deploy all servers at once: nix run 'nixpkgs#deploy-rs' -- '.#' -k
To deploy only one server: nix run 'nixpkgs#deploy-rs' -- -k --targets '.#<host>'
Deploying to Kubernetes
To deploy to the Kubernetes cluster, first make sure you have an admin account on the cluster.
You can generate this using nix run '.#gen-k3s-cert' <username> <servername> ~/.kube
, assuming you have SSH access to the master node.
This puts a private key, signed certificate and a kubeconfig in the kubeconfig directory
We are now ready to deploy to the Kubernetes cluster. Deployments are done through an experimental Kubernetes feature called ApplySets. Each applyset is responsible for a set number of resources within a namespace.
If the cluster has not been initialized yet, we must bootstrap it first. Run these deployments:
nix run '.#bootstrap-default'
nix run '.#bootstrap-kube-system'
Now the cluster has been initialized and we can deploy applications.
To explore which applications we can deploy, run nix flake show
.
Then, for each application, run nix run '.#<application>'
.
Known bugs
Rsync not available during bootstrap
The rsync
command was removed from recent NixOS ISO which causes nixos-anywhere to fail when copying extra files.
See this issue.
Solution is to execute nix-env -iA nixos.rsync
on the host.