k8scd/nix-infra-machine/cli

#!/usr/bin/env bash
WORK_DIR=${WORK_DIR:-"$(dirname "$0")"}
NIX_INFRA=${NIX_INFRA:-"nix-infra"}
NIXOS_VERSION=${NIXOS_VERSION:-"25.11"}
SSH_KEY=${SSH_KEY:-"nixinfra-machine"}
SSH_EMAIL=${SSH_EMAIL:-"your-email@example.com"}
SECRETS_PWD=${SECRETS_PWD:-"my_secrets_password"}
LOCATION=${LOCATION:-"nbg1"}
MACHINE_TYPE=${MACHINE_TYPE:-"cx23"}
ENV=${ENV:-"$WORK_DIR/.env"}

read -r -d '' __help_text__ <<EOF || true
nix-infra-machine CLI
=====================

Usage: $0 <command> [options]

Commands:
  create <nodes>      Provision and initialize machines
  update <nodes>      Update node configuration and deploy apps
  upgrade <nodes>     Upgrade NixOS version on nodes
  rollback <nodes>    Rollback to previous NixOS configuration
  destroy             Destroy machines
      --target=<nodes>    Nodes to destroy

  ssh <node>          SSH into a node
  cmd                 Run command on node(s)
      --target=<nodes>    Target node(s)
      <command>           Command to execute
  action              Run app module action
      --target=<node>     Target node
      <module>            App module name
      <cmd>               Action command to run
  port-forward        Forward port from remote node to local
      --target=<node>         Target node to forward from
      --port-mapping=<l:r>    Port mapping as local:remote
  claude              Launch Claude with nix-infra-machine-mcp

Options:
  --env=<file>            Environment file (default: .env)
  --target=<nodes>        Target node(s) for commands
  --node-module=<file>    Node module file (default: node_types/standalone_machine.nix)
  --port-mapping=<l:r>    Port mapping as local:remote (for port-forward)

Examples:
  # Create a new machine
  $0 create --env=.env node001

  # Create multiple machines
  $0 create --env=.env node001 node002 node003

  # SSH into a machine
  $0 ssh --env=.env node001

  # Run a command on a machine
  $0 cmd --env=.env --target=node001 "systemctl status nginx"

  # Update configuration
  $0 update --env=.env node001

  # Destroy machines
  $0 destroy --env=.env --target="node001 node002"

  # Launch Claude with dev MCP
  $0 claude-dev --env=.env
EOF

if [[ "create upgrade rollback destroy update ssh cmd action port-forward claude claude-dev" == *"$1"* ]]; then
  CMD="$1"
  shift
else
  echo "$__help_text__"
  exit 1
fi

for i in "$@"; do
  case $i in
    --help)
    echo "$__help_text__"
    exit 0
    ;;
    --env=*)
    ENV="${i#*=}"
    shift
    ;;
    --target=*)
    TARGET="${i#*=}"
    shift
    ;;
    --port-mapping=*)
    PORT_MAPPING="${i#*=}"
    shift
    ;;
    --node-module=*)
    NODE_MODULE="${i#*=}"
    shift
    ;;
    *)
    REST="$@"
    ;;
  esac
done

# Read the environment file if provided
if [ "$ENV" != "" ] && [ -f "$ENV" ]; then
  source "$ENV"
fi

# Check for nix-infra CLI
if [ "$NIX_INFRA" = "nix-infra" ] && ! command -v nix-infra >/dev/null 2>&1; then
  echo "The 'nix-infra' cli is required for this script to work."
  echo "Visit https://github.com/jhsware/nix-infra for instructions on installation"
  exit 1
fi

# A Hetzner Cloud token is required
if [ -z "$HCLOUD_TOKEN" ]; then
  echo "Missing env-var HCLOUD_TOKEN. Load through .env-file that is specified through --env."
  exit 1
fi

# ============================================================================
# Validation helpers
# ============================================================================

check_required_vars() {
  local missing=""
  for var in "$@"; do
    if [ -z "${!var}" ]; then
      missing="$missing $var"
    fi
  done
  if [ -n "$missing" ]; then
    die "Missing required environment variables:$missing
Load through .env-file specified with --env option."
  fi
}

check_required_args() {
  local name="$1"
  local value="$2"
  if [ -z "$value" ]; then
    die "Missing required argument: $name"
  fi
}

# ============================================================================
# Helper Functions
# ============================================================================

printTime() {
  local _start=$1; local _end=$2; local _secs=$((_end-_start))
  printf '%02dh:%02dm:%02ds' $((_secs/3600)) $((_secs%3600/60)) $((_secs%60))
}

destroyNodes() {
  $NIX_INFRA fleet destroy -d "$WORK_DIR" --env="$ENV" --batch \
      --target="$TARGET"
}

cleanupOnFail() {
  if [ $1 -ne 0 ]; then
    echo "$2"
    destroyNodes
    exit 1
  fi
}

# ============================================================================
# Interactive Commands
# ============================================================================

if [ "$CMD" = "ssh" ]; then
  if [ -z "$REST" ]; then
    echo "Usage: $0 ssh --env=.env <node>"
    exit 1
  fi
  $NIX_INFRA fleet ssh -d "$WORK_DIR" --env="$ENV" --target="$REST"
  exit 0
fi

if [ "$CMD" = "cmd" ]; then
  if [ -z "$TARGET" ] || [ -z "$REST" ]; then
    echo "Usage: $0 cmd --env=.env --target=<node> <command>"
    exit 1
  fi
  $NIX_INFRA fleet cmd -d "$WORK_DIR" --env="$ENV" --target="$TARGET" "$REST"
  exit 0
fi

if [ "$CMD" = "action" ]; then
  if [ -z "$TARGET" ] || [ -z "$REST" ]; then
    echo "Usage: $0 action --env=.env --target=<node> <module> <cmd>"
    exit 1
  fi
  
  read -r module action_cmd <<< "$REST"
  $NIX_INFRA fleet action -d "$WORK_DIR" --env="$ENV" --target="$TARGET" --app-module="$module" \
    --cmd="$action_cmd"
  exit 0
fi

if [ "$CMD" = "port-forward" ]; then
  if [ -z "$TARGET" ] || [ -z "$PORT_MAPPING" ]; then
    echo "Usage: $0 port-forward --env=.env --target=<node> --port-mapping=<local:remote>"
    exit 1
  fi

  OLD_IFS=$IFS
  IFS=: read LOCAL_PORT REMOTE_PORT <<< "$PORT_MAPPING"
  IFS=$OLD_IFS

  $NIX_INFRA fleet port-forward -d "$WORK_DIR" --env="$ENV" \
    --target="$TARGET" \
    --local-port="$LOCAL_PORT" \
    --remote-port="$REMOTE_PORT"
  exit 0
fi

# ============================================================================
# Fleet Management Commands
# ============================================================================

if [ "$CMD" = "destroy" ]; then
  if [ -z "$TARGET" ]; then
    echo "Usage: $0 destroy --env=.env --target=\"<node1> <node2> ...\""
    exit 1
  fi

  destroyNodes
  exit 0
fi

if [ "$CMD" = "update" ]; then
  if [ -z "$REST" ]; then
    echo "Usage: $0 update --env=.env <node1> <node2> ..."
    exit 1
  fi

  NODE_MODULE=${NODE_MODULE:-"node_types/standalone_machine.nix"}
  
  $NIX_INFRA fleet update -d "$WORK_DIR" --env="$ENV" --batch \
    --nixos-version="$NIXOS_VERSION" \
    --target="$REST" \
    --node-module="$NODE_MODULE" \
    --no-rebuild
  
  $NIX_INFRA fleet deploy-apps -d "$WORK_DIR" --env="$ENV" --batch \
    --target="$REST"

  $NIX_INFRA fleet cmd -d "$WORK_DIR" --env="$ENV" --target="$REST" "nixos-rebuild switch"
  exit 0
fi

if [ "$CMD" = "upgrade" ]; then
  if [ -z "$REST" ]; then
    echo "Usage: $0 upgrade --env=$ENV [--nixos-version='$NIXOS_VERSION'] [node1 node2 ...]"
    exit 1
  fi
  # (cd "$WORK_DIR" && git fetch origin && git reset --hard origin/$(git branch --show-current))
  $NIX_INFRA cluster upgrade-nixos -d "$WORK_DIR" --env="$ENV" --batch \
    --nixos-version="$NIXOS_VERSION" \
    --target="$REST"
  exit 0
fi

if [ "$CMD" = "rollback" ]; then
  if [ -z "$REST" ]; then
    echo "Usage: $0 rollback --env=.env <node1> <node2> ..."
    exit 1
  fi

  $NIX_INFRA fleet cmd -d "$WORK_DIR" --env="$ENV" --target="$REST" "nixos-rebuild switch --rollback"
  exit 0
fi

# ============================================================================
# Create Command - Provision and Initialize Machines
# ============================================================================

if [ "$CMD" = "create" ]; then
  if [ -z "$REST" ]; then
    echo "Usage: $0 create --env=.env <node1> <node2> ..."
    exit 1
  fi

  TARGET="$REST"

  _start=$(date +%s)

  # Initialize if not already done (check for ssh directory)
  if [ ! -d "$WORK_DIR/ssh" ]; then
    echo "Initializing nix-infra..."
    $NIX_INFRA init -d "$WORK_DIR" --env="$ENV" --no-cert-auth --batch
  fi

  # Add SSH key to agent
  ssh-add "$WORK_DIR/ssh/$SSH_KEY" 2>/dev/null || true

  echo "*** Provisioning NixOS $NIXOS_VERSION ***"

  $NIX_INFRA fleet provision -d "$WORK_DIR" --env="$ENV" --batch \
      --nixos-version="$NIXOS_VERSION" \
      --ssh-key="$SSH_KEY" \
      --location="$LOCATION" \
      --machine-type="$MACHINE_TYPE" \
      --node-names="$TARGET"

  cleanupOnFail $? "WARNING: Provisioning failed! Cleaning up..."

  _provision=$(date +%s)

  echo "*** Initializing machines ***"

  NODE_MODULE=${NODE_MODULE:-"node_types/standalone_machine.nix"}

  $NIX_INFRA fleet init-machine -d "$WORK_DIR" --env="$ENV" --batch \
      --nixos-version="$NIXOS_VERSION" \
      --target="$TARGET" \
      --node-module="$NODE_MODULE"

  $NIX_INFRA fleet cmd -d "$WORK_DIR" --env="$ENV" --target="$TARGET" "nixos-rebuild switch --fast"

  _init_nodes=$(date +%s)

  echo "*** Deploying apps ***"

  $NIX_INFRA fleet deploy-apps -d "$WORK_DIR" --env="$ENV" --batch \
    --target="$TARGET"
  $NIX_INFRA fleet cmd -d "$WORK_DIR" --env="$ENV" --target="$TARGET" "nixos-rebuild switch --fast"

  _end=$(date +%s)

  echo ""
  echo "=========================================="
  printf '+ provision    %s\n' "$(printTime $_start $_provision)"
  printf '+ init nodes   %s\n' "$(printTime $_provision $_init_nodes)"
  printf '+ deploy apps  %s\n' "$(printTime $_init_nodes $_end)"
  printf '= TOTAL        %s\n' "$(printTime $_start $_end)"
  echo "=========================================="
  echo "Done!"
  exit 0
fi

if [ "$CMD" = "claude" ]; then
  if [ -z "$SSH_KEY" ] || [ -z "$HCLOUD_TOKEN" ]; then
    echo "You need a .env file in the working directory with SSH_KEY and HCLOUD_TOKEN defined" >&2
    exit 1
  fi

  if ! type nix-infra-machine-mcp &> /dev/null; then
    echo "nix-infra-machine-mcp is not installed" >&2
    exit 1
  fi

  # Detect OS and set paths accordingly
  case "$(uname -s)" in
    Darwin)
      path_to_claude="$HOME/Library/Application Support/Claude"
      claude_bin="/Applications/Claude.app/Contents/MacOS/Claude"
      ;;
    Linux)
      path_to_claude="${XDG_CONFIG_HOME:-$HOME/.config}/Claude"
      claude_bin="claude"
      ;;
    *)
      echo "Unsupported operating system: $(uname -s)" >&2
      exit 1
      ;;
  esac

  if [ ! -d "$path_to_claude" ]; then
    echo "You do not have Claude Application Support files in your user. Start the app once manually and retry." >&2
    exit 1
  fi

  if [ -f "$path_to_claude/claude_desktop_config.json" ]; then
    cp -f "$path_to_claude/claude_desktop_config.json" "$path_to_claude/claude_desktop_config.json.nix-infra.bak"
  fi

  # Create mcp configuration for Claude
  cat > "$path_to_claude/claude_desktop_config.json" << EOF
{
  "mcpServers": {
      "nix-infra-machine-mcp": {
          "command": "nix-infra-machine-mcp"
      }
  }
}
EOF

  # Run claude
  "$claude_bin" 2>&1

  # Reset mcp configuration for Claude
  if [ -f "$path_to_claude/claude_desktop_config.json.nix-infra.bak" ]; then
    cp -f "$path_to_claude/claude_desktop_config.json.nix-infra.bak" "$path_to_claude/claude_desktop_config.json"
  fi
fi

# ============================================================================
# MCP Support
# ============================================================================

if [ "$CMD" = "claude" ] || [ "$CMD" = "claude-dev" ]; then
  check_required_vars SSH_KEY HCLOUD_TOKEN

  # Detect OS and set paths accordingly
  case "$(uname -s)" in
    Darwin)
      path_to_claude="$HOME/Library/Application Support/Claude"
      claude_bin="/Applications/Claude.app/Contents/MacOS/Claude"
      ;;
    Linux)
      path_to_claude="${XDG_CONFIG_HOME:-$HOME/.config}/Claude"
      claude_bin="claude"
      ;;
    *)
      die "Unsupported operating system: $(uname -s)"
      ;;
  esac

  if [ ! -d "$path_to_claude" ]; then
    die "You do not have Claude Application Support files in your user. Start the app once manually and retry."
  fi


  # ============================================================================
  # Command: claude
  # ============================================================================

  if [ "$CMD" = "claude" ]; then
    if ! command -v nix-infra-cluster-mcp &>/dev/null; then
      die "nix-infra-cluster-mcp is not installed"
    fi
        
    # Backup existing config
    if [ -f "$path_to_claude/claude_desktop_config.json" ]; then
      cp -f "$path_to_claude/claude_desktop_config.json" "$path_to_claude/claude_desktop_config.json.nix-infra.bak"
    fi
    
    # Create mcp configuration for Claude
    cat >"$path_to_claude/claude_desktop_config.json" <<EOF
{
  "mcpServers": {
    "nix-infra-cluster-mcp": {
      "command": "nix-infra-cluster-mcp"
    }
  }
}
EOF
    
    # Run claude
    "$claude_bin" 2>&1
    
    # Reset mcp configuration for Claude
    if [ -f "$path_to_claude/claude_desktop_config.json.nix-infra.bak" ]; then
      cp -f "$path_to_claude/claude_desktop_config.json.nix-infra.bak" "$path_to_claude/claude_desktop_config.json"
    fi
    
    exit 0
  fi

  # ============================================================================
  # Command: claude-dev
  # ============================================================================

  if [ "$CMD" = "claude-dev" ]; then
    # if ! command -v nix-infra-dev-mcp &>/dev/null; then
    #   die "nix-infra-dev-mcp is not installed"
    # fi
        
    # Backup existing config
    if [ -f "$path_to_claude/claude_desktop_config.json" ]; then
      cp -f "$path_to_claude/claude_desktop_config.json" "$path_to_claude/claude_desktop_config.json.nix-infra.bak"
    fi
    
    # Create mcp configuration for Claude
    cat >"$path_to_claude/claude_desktop_config.json" <<EOF
{
  "mcpServers": {
    "nix-infra-dev-mcp": {
      "command": "nix-infra-dev-mcp"
    }
  }
}
EOF
    
    # Run claude
    "$claude_bin" 2>&1
    
    # Reset mcp configuration for Claude
    if [ -f "$path_to_claude/claude_desktop_config.json.nix-infra.bak" ]; then
      cp -f "$path_to_claude/claude_desktop_config.json.nix-infra.bak" "$path_to_claude/claude_desktop_config.json"
    fi
    
    exit 0
  fi
fi

# If we get here, command wasn't handled
die "Unknown command: $CMD"