TerraHome/nixos/kubeadm/modules/k8s-common.nix

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

let
  pinnedK8s = lib.attrByPath [ "kubernetes_1_31" ] pkgs.kubernetes pkgs;
  kubeVipImage = "ghcr.io/kube-vip/kube-vip:v0.8.9";
in
{
  options.terrahome.kubeadm = {
    k8sMinor = lib.mkOption {
      type = lib.types.str;
      default = "1.31";
    };

    controlPlaneInterface = lib.mkOption {
      type = lib.types.str;
      default = "eth0";
    };

    controlPlaneVipSuffix = lib.mkOption {
      type = lib.types.int;
      default = 250;
    };

    podSubnet = lib.mkOption {
      type = lib.types.str;
      default = "10.244.0.0/16";
    };

    serviceSubnet = lib.mkOption {
      type = lib.types.str;
      default = "10.96.0.0/12";
    };

    clusterDomain = lib.mkOption {
      type = lib.types.str;
      default = "cluster.local";
    };
  };

  config = {
  boot.kernelModules = [ "overlay" "br_netfilter" ];

  boot.kernel.sysctl = {
    "net.ipv4.ip_forward" = 1;
    "net.bridge.bridge-nf-call-iptables" = 1;
    "net.bridge.bridge-nf-call-ip6tables" = 1;
  };

  virtualisation.containerd.enable = true;
  virtualisation.containerd.settings = {
    plugins."io.containerd.grpc.v1.cri".containerd.runtimes.runc.options.SystemdCgroup = true;
  };

  swapDevices = lib.mkForce [ ];

  services.openssh.enable = true;
  services.openssh.settings = {
    PasswordAuthentication = false;
    KbdInteractiveAuthentication = false;
  };

  users.users.micqdf = {
    isNormalUser = true;
    extraGroups = [ "wheel" ];
  };

  security.sudo.wheelNeedsPassword = false;

  nix.settings.trusted-users = [ "root" "micqdf" ];
  nix.gc = {
    automatic = true;
    dates = "daily";
    options = "--delete-older-than 3d";
  };
  nix.settings.auto-optimise-store = true;

  environment.variables = {
    KUBECONFIG = "/etc/kubernetes/admin.conf";
    KUBE_VIP_IMAGE = kubeVipImage;
  };

  environment.systemPackages = (with pkgs; [
    containerd
    cri-tools
    cni-plugins
    pinnedK8s
    kubernetes-helm
    conntrack-tools
    socat
    ethtool
    ipvsadm
    iproute2
    iptables
    ebtables
    jq
    curl
    vim
    gawk
  ]) ++ [
    (pkgs.writeShellScriptBin "th-kubeadm-init" ''
      set -euo pipefail

      unset http_proxy https_proxy HTTP_PROXY HTTPS_PROXY no_proxy NO_PROXY

      iface="${config.terrahome.kubeadm.controlPlaneInterface}"
      if ! ip link show "$iface" >/dev/null 2>&1; then
        iface="$(ip -o -4 route show to default | awk 'NR==1 {print $5}')"
      fi

      if [ -z "''${iface:-}" ]; then
        echo "Could not determine network interface for kube-vip"
        exit 1
      fi

      suffix="${toString config.terrahome.kubeadm.controlPlaneVipSuffix}"
      pod_subnet="${config.terrahome.kubeadm.podSubnet}"
      service_subnet="${config.terrahome.kubeadm.serviceSubnet}"
      domain="${config.terrahome.kubeadm.clusterDomain}"

      local_ip_cidr=$(ip -4 -o addr show dev "$iface" | awk 'NR==1 {print $4}')
      if [ -z "''${local_ip_cidr:-}" ]; then
        echo "Could not determine IPv4 CIDR on interface $iface"
        exit 1
      fi

      subnet_prefix=$(echo "$local_ip_cidr" | cut -d/ -f1 | awk -F. '{print $1"."$2"."$3}')
      vip="$subnet_prefix.$suffix"

      echo "Using control-plane endpoint: $vip:6443"
      echo "Using kube-vip interface: $iface"

      mkdir -p /etc/kubernetes/manifests
      ctr image pull "${kubeVipImage}"

      ctr run --rm --net-host "${kubeVipImage}" kube-vip /kube-vip manifest pod \
        --interface "$iface" \
        --address "$vip" \
        --controlplane \
        --services \
        --arp \
        --leaderElection \
        > /etc/kubernetes/manifests/kube-vip.yaml

      systemctl unmask kubelet || true
      systemctl stop kubelet || true
      systemctl reset-failed kubelet || true
      env -i PATH=/run/current-system/sw/bin:/usr/bin:/bin kubeadm reset -f || true

      systemctl daemon-reload
      systemctl unmask kubelet || true

      env -i PATH=/run/current-system/sw/bin:/usr/bin:/bin kubeadm init \
        --control-plane-endpoint "$vip:6443" \
        --upload-certs \
        --ignore-preflight-errors=NumCPU,HTTPProxyCIDR,Port-10250 \
        --pod-network-cidr "$pod_subnet" \
        --service-cidr "$service_subnet" \
        --service-dns-domain "$domain"

      mkdir -p /root/.kube
      cp /etc/kubernetes/admin.conf /root/.kube/config
      chmod 600 /root/.kube/config

      echo
      echo "Next: install Cilium, then generate join commands:"
      echo "  kubeadm token create --print-join-command"
      echo "  kubeadm token create --print-join-command --certificate-key <key>"
    '')

    (pkgs.writeShellScriptBin "th-kubeadm-join-control-plane" ''
      set -euo pipefail
      unset http_proxy https_proxy HTTP_PROXY HTTPS_PROXY no_proxy NO_PROXY
      if [ "$#" -lt 1 ]; then
        echo "Usage: th-kubeadm-join-control-plane '<kubeadm join ... --control-plane --certificate-key ...>'"
        exit 1
      fi

      iface="${config.terrahome.kubeadm.controlPlaneInterface}"
      if ! ip link show "$iface" >/dev/null 2>&1; then
        iface="$(ip -o -4 route show to default | awk 'NR==1 {print $5}')"
      fi

      if [ -z "''${iface:-}" ]; then
        echo "Could not determine network interface for kube-vip"
        exit 1
      fi

      suffix="${toString config.terrahome.kubeadm.controlPlaneVipSuffix}"
      local_ip_cidr=$(ip -4 -o addr show dev "$iface" | awk 'NR==1 {print $4}')
      if [ -z "''${local_ip_cidr:-}" ]; then
        echo "Could not determine IPv4 CIDR on interface $iface"
        exit 1
      fi

      subnet_prefix=$(echo "$local_ip_cidr" | cut -d/ -f1 | awk -F. '{print $1"."$2"."$3}')
      vip="$subnet_prefix.$suffix"

      mkdir -p /etc/kubernetes/manifests
      ctr image pull "${kubeVipImage}"
      ctr run --rm --net-host "${kubeVipImage}" kube-vip /kube-vip manifest pod \
        --interface "$iface" \
        --address "$vip" \
        --controlplane \
        --services \
        --arp \
        --leaderElection \
        > /etc/kubernetes/manifests/kube-vip.yaml

      systemctl unmask kubelet || true
      systemctl stop kubelet || true
      systemctl reset-failed kubelet || true
      systemctl daemon-reload
      eval "$1"
    '')

    (pkgs.writeShellScriptBin "th-kubeadm-join-worker" ''
      set -euo pipefail
      unset http_proxy https_proxy HTTP_PROXY HTTPS_PROXY no_proxy NO_PROXY
      if [ "$#" -lt 1 ]; then
        echo "Usage: th-kubeadm-join-worker '<kubeadm join ...>'"
        exit 1
      fi

      systemctl unmask kubelet || true
      systemctl stop kubelet || true
      systemctl reset-failed kubelet || true
      systemctl daemon-reload
      eval "$1"
    '')

    (pkgs.writeShellScriptBin "th-kubeadm-status" ''
      set -euo pipefail
      systemctl is-active containerd || true
      systemctl is-active kubelet || true
      crictl info >/dev/null && echo "crictl: ok" || echo "crictl: not-ready"
    '')
  ];

  systemd.services.kubelet = {
    description = "Kubernetes Kubelet";
    wantedBy = lib.mkForce [];
    wants = [ "network-online.target" ];
    after = [ "containerd.service" "network-online.target" ];
    serviceConfig = {
      Environment = "KUBELET_CONFIG_ARGS=--config=/var/lib/kubelet/config.yaml";
      EnvironmentFile = [
        "-/var/lib/kubelet/kubeadm-flags.env"
        "-/etc/default/kubelet"
      ];
      ExecStart = "${pinnedK8s}/bin/kubelet $KUBELET_CONFIG_ARGS $KUBELET_KUBEADM_ARGS $KUBELET_EXTRA_ARGS";
      Restart = "always";
      RestartSec = "10";
    };
  };

  systemd.tmpfiles.rules = [
    "d /etc/kubernetes 0755 root root -"
    "d /etc/kubernetes/manifests 0755 root root -"
  ];
  };
}