athenix/hw/nix-lxc.nix

# ============================================================================
# Proxmox LXC Container Configuration
# ============================================================================
# Configuration for lightweight Linux containers running in Proxmox.
# Disables boot/disk management and enables remote development support.

{
  config,
  lib,
  pkgs,
  ...
}:

with lib;

let
  cfg = config.athenix.hw.nix-lxc;
in
{
  options.athenix.hw.nix-lxc = mkOption {
    type = types.submodule {
      options = {
        enable = mkOption {
          type = types.bool;
          default = false;
          description = "Enable Proxmox LXC container hardware configuration.";
        };
        cuda.enable = mkOption {
          type = types.bool;
          default = false;
          description = "Enable CUDA support in LXC containers (requires NVIDIA GPU and drivers on host).";
        };
        cuda.visibleDevices = mkOption {
          type = types.listOf types.str;
          default = [ "all" ];
          example = [
            "0"
            "1"
          ];
          description = ''
            NVIDIA devices exposed to containerized workloads.
            Use indexes (e.g. "0", "1"), UUIDs, or "all".
          '';
        };
        cuda.driverCapabilities = mkOption {
          type = types.listOf types.str;
          default = [
            "compute"
            "utility"
          ];
          example = [
            "compute"
            "utility"
            "video"
          ];
          description = ''
            Value used for NVIDIA_DRIVER_CAPABILITIES for container runtimes.
          '';
        };
        cuda.driver.channel = mkOption {
          type = types.str;
          default = "dc_550";
          example = "legacy_470";
          description = ''
            NVIDIA driver package channel from boot.kernelPackages.nvidiaPackages.
            Common values include: stable, latest, beta, dc_550, legacy_470, legacy_390.
            Alias values: production -> stable, datacenter -> dc_550.
          '';
        };
        cuda.driver.package = mkOption {
          type = types.nullOr types.package;
          default = null;
          description = ''
            Explicit NVIDIA driver package override. When set, this takes precedence
            over cuda.driver.channel.
          '';
        };
      };
    };
    default = { };
    description = "Proxmox LXC hardware type configuration.";
  };

  config = mkIf cfg.enable {
    # ========== Nix Configuration ==========
    nix.settings.trusted-users = [
      "root"
      "engr-ugaif"
    ];
    nix.settings.experimental-features = [
      "nix-command"
      "flakes"
    ];

    # ========== Container-Specific Configuration ==========
    boot.isContainer = true;
    boot.loader.systemd-boot.enable = lib.mkForce false; # No bootloader in container
    disko.enableConfig = lib.mkForce false; # No disk management in container
    console.enable = true;

    # Set timezone to fix /etc/localtime for Docker containers
    time.timeZone = lib.mkDefault "America/New_York";

    # NVIDIA Container Toolkit for CUDA support (optional)
    hardware.nvidia = lib.mkIf cfg.cuda.enable {
      package =
        let
          nvidiaPackages = config.boot.kernelPackages.nvidiaPackages;
          driverAliasMap = {
            production = "stable";
            datacenter = "dc_550";
          };
          driverChannel = driverAliasMap.${cfg.cuda.driver.channel} or cfg.cuda.driver.channel;
        in
        if cfg.cuda.driver.package != null then
          cfg.cuda.driver.package
        else if builtins.hasAttr driverChannel nvidiaPackages then
          builtins.getAttr driverChannel nvidiaPackages
        else
          throw "athenix.hw.nix-lxc.cuda.driver.channel '${driverChannel}' not found in boot.kernelPackages.nvidiaPackages";
    };
    hardware.nvidia-container-toolkit.enable = lib.mkIf cfg.cuda.enable true;
    hardware.nvidia-container-toolkit.suppressNvidiaDriverAssertion = lib.mkIf cfg.cuda.enable true;
    environment.systemPackages = lib.mkIf cfg.cuda.enable [
      (pkgs.writeShellScriptBin "nvidia-smi" ''
        primary="${config.hardware.nvidia.package}/bin/nvidia-smi"
        secondary="${
          if builtins.hasAttr "bin" config.hardware.nvidia.package then
            config.hardware.nvidia.package.bin
          else
            config.hardware.nvidia.package
        }/bin/nvidia-smi"

        if [ -x "$primary" ]; then
          exec "$primary" "$@"
        fi

        if [ -x "$secondary" ]; then
          exec "$secondary" "$@"
        fi

        echo "nvidia-smi binary not found in configured NVIDIA package: ${config.hardware.nvidia.package}" >&2
        exit 127
      '')
    ];
    environment.variables = lib.mkIf cfg.cuda.enable {
      NVIDIA_VISIBLE_DEVICES = lib.concatStringsSep "," cfg.cuda.visibleDevices;
      NVIDIA_DRIVER_CAPABILITIES = lib.concatStringsSep "," cfg.cuda.driverCapabilities;
    };

    # Allow getty to work in containers
    systemd.services."getty@".unitConfig.ConditionPathExists = [
      ""
      "/dev/%I"
    ];

    # Suppress unnecessary systemd units for containers
    systemd.suppressedSystemUnits = [
      "dev-mqueue.mount"
      "sys-kernel-debug.mount"
      "sys-fs-fuse-connections.mount"
    ];

    # ========== Remote Development ==========
    services.vscode-server.enable = true;

    # ========== System Configuration ==========
    system.stateVersion = "25.11";
    athenix.host.buildMethods = lib.mkDefault [
      "lxc" # LXC container tarball
      "proxmox" # Proxmox VMA archive
    ];

    athenix.sw.enable = lib.mkDefault true;
    athenix.sw.headless.enable = lib.mkDefault true;
  };
}