#!/bin/bash # DartNode Node Deployment Script # Supports Proxmox VE 8.3+ (Linux) and macOS Apple Silicon hosts # One-liner: curl -sSL https://pkg.dev.snaju.com/dn/dn-deploy.sh | bash # With token: curl -sSL https://pkg.dev.snaju.com/dn/dn-deploy.sh | bash -s -- --token YOUR_TOKEN # ============================================ # macOS Detection — delegate to dartnode.sh # ============================================ if [ "$(uname -s)" = "Darwin" ]; then echo "[DartNode] macOS detected — delegating to macOS installer..." # Filter out "--" separator and pass remaining args to dartnode.sh MACOS_ARGS=() for arg in "$@"; do [ "$arg" != "--" ] && MACOS_ARGS+=("$arg") done exec bash <(curl -sSL https://pkg.dev.snaju.com/dn/dartnode.sh) install "${MACOS_ARGS[@]}" exit $? fi set -e # Ensure sbin paths are in PATH (needed for mdadm, pvcreate, etc.) export PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:$PATH" # Colors RED='\033[0;31m' GREEN='\033[0;32m' YELLOW='\033[1;33m' BLUE='\033[0;34m' CYAN='\033[0;36m' MAGENTA='\033[0;35m' NC='\033[0m' # No Color BOLD='\033[1m' DIM='\033[2m' # Configuration INTERFACES_FILE="/etc/network/interfaces" MDADM_CONF="/etc/mdadm/mdadm.conf" API_BASE="https://api.dartnode.net" # Registration token (can be set via --token flag or DARTNODE_TOKEN env var) REGISTRATION_TOKEN="${DARTNODE_TOKEN:-}" # State tracking NETWORK_CONFIGURED=false RAID_CONFIGURED=false LVM_CONFIGURED=false STORAGE_CONFIGURED=false DAEMON_INSTALLED=false REGISTERED_WITH_API=false RAID_DEVICE="" # Restore mode: node lost its boot drive but customer data volumes survived. # All storage steps become strictly non-destructive (assemble/activate/re-add # only — never wipefs/mdadm --create/pvcreate/vgcreate). RESTORE_MODE=false DISCOVERED_VG="" function header() { clear echo -e "${CYAN}" echo " ___ _ __ _ " echo " / \\__ _ _ __| |_ /\\ \\ \\___ __| | ___ " echo " / /\\ / _\` | '__| __|/ \\/ / _ \\ / _\` |/ _ \\" echo " / /_// (_| | | | |_/ /\\ / (_) | (_| | __/" echo "/___,' \\__,_|_| \\__\\_\\ \\/ \\___/ \\__,_|\\___|" echo -e "${NC}" echo -e "${BOLD}DartNode Proxmox Node Deployment Script${NC}" echo "Copyright (c) 2024-2026, Snaju Inc. All Rights Reserved." echo "" echo "==============================================" echo "" } function log_info() { echo -e "${BLUE}[INFO]${NC} $1" } function log_success() { echo -e "${GREEN}[SUCCESS]${NC} $1" } function log_warn() { echo -e "${YELLOW}[WARNING]${NC} $1" } function log_error() { echo -e "${RED}[ERROR]${NC} $1" } function log_step() { echo -e "${MAGENTA}[STEP]${NC} $1" } function check_root() { if [ "$EUID" -ne 0 ]; then log_error "Please run as root" exit 1 fi } function check_proxmox() { if [ ! -f /etc/pve/datacenter.cfg ]; then log_error "This doesn't appear to be a Proxmox node." exit 1 fi PVE_VERSION=$(pveversion | grep -oP 'pve-manager/\K[0-9]+\.[0-9]+') log_info "Detected Proxmox VE version: $PVE_VERSION" } function fix_proxmox_repos() { log_step "Configuring Proxmox repositories..." local REPOS_CHANGED=false # Disable enterprise PVE repo if it exists if [ -f /etc/apt/sources.list.d/pve-enterprise.list ]; then if ! grep -q "^#" /etc/apt/sources.list.d/pve-enterprise.list 2>/dev/null || grep -q "^deb" /etc/apt/sources.list.d/pve-enterprise.list 2>/dev/null; then log_info "Disabling Proxmox VE enterprise repository..." sed -i 's/^deb/#deb/' /etc/apt/sources.list.d/pve-enterprise.list REPOS_CHANGED=true fi fi # Disable enterprise Ceph repo if it exists if [ -f /etc/apt/sources.list.d/ceph.list ]; then if grep -q "enterprise.proxmox.com" /etc/apt/sources.list.d/ceph.list 2>/dev/null; then log_info "Disabling Proxmox Ceph enterprise repository..." sed -i 's/^deb/#deb/' /etc/apt/sources.list.d/ceph.list REPOS_CHANGED=true fi fi # Add no-subscription PVE repo if not present local PVE_NO_SUB="/etc/apt/sources.list.d/pve-no-subscription.list" if [ ! -f "$PVE_NO_SUB" ] || ! grep -q "pve-no-subscription" "$PVE_NO_SUB" 2>/dev/null; then log_info "Adding Proxmox VE no-subscription repository..." echo "deb http://download.proxmox.com/debian/pve bookworm pve-no-subscription" > "$PVE_NO_SUB" REPOS_CHANGED=true fi # Add no-subscription Ceph repo if not present (for Proxmox 8+) local CEPH_NO_SUB="/etc/apt/sources.list.d/ceph-no-subscription.list" if [ ! -f "$CEPH_NO_SUB" ] || ! grep -q "ceph-quincy" "$CEPH_NO_SUB" 2>/dev/null; then log_info "Adding Ceph no-subscription repository..." echo "deb http://download.proxmox.com/debian/ceph-quincy bookworm no-subscription" > "$CEPH_NO_SUB" REPOS_CHANGED=true fi if [ "$REPOS_CHANGED" = true ]; then log_info "Updating package lists..." apt-get update -qq log_success "Repositories configured!" else log_success "Repositories already configured." fi } function confirm() { local prompt="$1" local default="${2:-y}" if [[ "$default" == "y" ]]; then prompt="${CYAN}▶${NC} $prompt ${DIM}[Y/n]${NC}: " else prompt="${CYAN}▶${NC} $prompt ${DIM}[y/N]${NC}: " fi echo "" echo -ne "$prompt" read response < /dev/tty response=${response:-$default} [[ "$response" =~ ^[Yy]$ ]] } # ============================================ # JSON Parsing Helpers # ============================================ function json_get() { local json="$1" local key="$2" if command -v python3 &> /dev/null; then echo "$json" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('data',d).get('$key',''))" 2>/dev/null else echo "$json" | grep -o "\"$key\":\"[^\"]*\"" | cut -d'"' -f4 fi } function json_get_nested() { local json="$1" local parent="$2" local key="$3" if command -v python3 &> /dev/null; then echo "$json" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d.get('data',d).get('$parent',{}).get('$key',''))" 2>/dev/null else echo "$json" | grep -o "\"$key\":[^,}]*" | head -1 | sed 's/.*://' | tr -d '"' fi } function json_get_bool() { local json="$1" local key="$2" if command -v python3 &> /dev/null; then echo "$json" | python3 -c "import sys,json; d=json.load(sys.stdin); print('true' if d.get('data',d).get('$key') else 'false')" 2>/dev/null else echo "$json" | grep -q "\"$key\":true" && echo "true" || echo "false" fi } # ============================================ # Token Validation Functions # ============================================ function validate_token() { local token="$1" log_step "Validating registration token..." local response response=$(curl -sS "${API_BASE}/node/status?token=${token}" 2>/dev/null) if [ -z "$response" ]; then log_error "Unable to connect to DartNode API" return 1 fi if echo "$response" | grep -q '"success":true'; then local is_valid is_used node_name is_valid=$(json_get_bool "$response" "isValid") is_used=$(json_get_bool "$response" "isUsed") node_name=$(json_get "$response" "nodeName") if [ "$is_valid" = "true" ]; then log_success "Token is valid for node: ${node_name}" TOKEN_NODE_NAME="$node_name" return 0 elif [ "$is_used" = "true" ]; then log_warn "Token has already been used" return 2 else log_error "Token is expired" return 1 fi else local error_msg error_msg=$(json_get "$response" "error") log_error "Invalid token: ${error_msg:-Unknown error}" return 1 fi } function fetch_token_config() { local token="$1" log_step "Fetching configuration from DartNode..." local response response=$(curl -sS "${API_BASE}/node/config?token=${token}" 2>/dev/null) if [ -z "$response" ]; then log_error "Unable to fetch configuration" return 1 fi if echo "$response" | grep -q '"success":true'; then # Parse and store config values TOKEN_NODE_NAME=$(json_get "$response" "nodeName") TOKEN_NODE_DB_ID=$(json_get "$response" "nodeDbId") TOKEN_NODE_TYPE=$(json_get "$response" "nodeType") TOKEN_STORAGE_NAME=$(json_get "$response" "storageName") TOKEN_STORAGE_TYPE=$(json_get "$response" "storageType") TOKEN_MODE=$(json_get "$response" "mode") # NFS shares the fleet actually mounts, as "name|server|export|path|content|options" # lines. Empty when the API couldn't reach a reference node (or no python3). TOKEN_NFS_STORAGES="" if command -v python3 &> /dev/null; then TOKEN_NFS_STORAGES=$(echo "$response" | python3 -c " import sys, json d = json.load(sys.stdin) d = d.get('data', d) for s in (d.get('nfsStorages') or []): print('|'.join([str(s.get(k) or '') for k in ('storage', 'server', 'export', 'path', 'content', 'options')])) " 2>/dev/null) fi TOKEN_REDIS_HOST=$(json_get_nested "$response" "redis" "host") TOKEN_REDIS_PORT=$(json_get_nested "$response" "redis" "port") TOKEN_REDIS_PASSWORD=$(json_get_nested "$response" "redis" "password") TOKEN_REDIS_DB=$(json_get_nested "$response" "redis" "db") TOKEN_AUTH_SECRET=$(json_get "$response" "authSecret") # JSON null comes back as "None" (python3) or "null" — treat as unset case "$TOKEN_STORAGE_NAME" in null|None) TOKEN_STORAGE_NAME="" ;; esac case "$TOKEN_STORAGE_TYPE" in null|None) TOKEN_STORAGE_TYPE="" ;; esac case "$TOKEN_NODE_DB_ID" in null|None) TOKEN_NODE_DB_ID="" ;; esac # Set defaults for required fields TOKEN_REDIS_HOST=${TOKEN_REDIS_HOST:-23.165.104.28} TOKEN_REDIS_PORT=${TOKEN_REDIS_PORT:-6379} TOKEN_REDIS_DB=${TOKEN_REDIS_DB:-0} TOKEN_REDIS_PASSWORD=${TOKEN_REDIS_PASSWORD:-} TOKEN_MODE=${TOKEN_MODE:-pve} TOKEN_AUTH_SECRET=${TOKEN_AUTH_SECRET:-dartnode-vnc-secret-2024} # Debug: Show what was parsed if [ -n "$DEBUG" ]; then echo "[DEBUG] Parsed config:" echo " NODE_NAME: $TOKEN_NODE_NAME" echo " REDIS_HOST: $TOKEN_REDIS_HOST" echo " REDIS_PORT: $TOKEN_REDIS_PORT" echo " REDIS_PASSWORD: ${TOKEN_REDIS_PASSWORD:0:8}..." fi log_success "Configuration fetched for node: ${TOKEN_NODE_NAME}" return 0 else local error_msg error_msg=$(json_get "$response" "error") log_error "Failed to fetch configuration: ${error_msg:-Unknown error}" return 1 fi } # ============================================ # Network Configuration # ============================================ function list_network_interfaces() { echo "" log_info "Available network interfaces:" echo "" ip -o link show | awk -F': ' '{print $2}' | grep -vE '^(lo|veth|tap|fwbr|fwln|fwpr)' | while read iface; do STATE=$(cat /sys/class/net/$iface/operstate 2>/dev/null || echo "unknown") SPEED=$(cat /sys/class/net/$iface/speed 2>/dev/null || echo "?") MAC=$(cat /sys/class/net/$iface/address 2>/dev/null || echo "unknown") if [ "$STATE" == "up" ]; then STATE_COLOR="${GREEN}UP${NC}" else STATE_COLOR="${RED}DOWN${NC}" fi printf " %-20s State: %-12b Speed: %s Mb/s MAC: %s\n" "$iface" "$STATE_COLOR" "$SPEED" "$MAC" done echo "" } function prompt_choice() { local prompt="$1" local default="$2" echo "" > /dev/tty echo -ne "${CYAN}▶${NC} $prompt ${DIM}[$default]${NC}: " > /dev/tty read response < /dev/tty echo "${response:-$default}" } function setup_network_bridges() { echo "" echo -e "${BOLD}Network Configuration Mode:${NC}" echo "" echo -e " ${CYAN}1)${NC} Standard - Separate physical ports for WAN and LAN" echo -e " ${CYAN}2)${NC} VLAN - Single trunk port with VLAN tagging" echo -e " ${CYAN}3)${NC} Skip network configuration" NET_MODE=$(prompt_choice "Select mode" "1") case $NET_MODE in 1) setup_network_standard ;; 2) setup_network_vlan ;; 3) log_info "Network configuration skipped." return 0 ;; *) log_error "Invalid selection." return 1 ;; esac } function setup_network_standard() { echo "" log_info "Standard network configuration (separate ports)" log_warn "Common SuperMicro interfaces: enp1s0f0 (WAN), enp1s0f1 (LAN)" list_network_interfaces # Check existing configuration EXISTING_VMBR0=$(grep -A5 "^iface vmbr0" $INTERFACES_FILE 2>/dev/null | grep bridge-ports | awk '{print $2}' || echo "") EXISTING_VMBR0_IP=$(grep -A5 "^iface vmbr0" $INTERFACES_FILE 2>/dev/null | grep "address" | awk '{print $2}' || echo "") EXISTING_VMBR1=$(grep -A5 "^iface vmbr1" $INTERFACES_FILE 2>/dev/null | grep bridge-ports | awk '{print $2}' || echo "") EXISTING_VMBR1_IP=$(grep -A5 "^iface vmbr1" $INTERFACES_FILE 2>/dev/null | grep "address" | awk '{print $2}' || echo "") SKIP_WAN=false SKIP_LAN=false # ===== WAN (vmbr0) Configuration ===== echo "" echo -e "${BOLD}━━━ WAN Bridge (vmbr0) ━━━${NC}" if [ -n "$EXISTING_VMBR0" ]; then echo "" log_success "vmbr0 already configured:" echo -e " ${DIM}Interface:${NC} $EXISTING_VMBR0" echo -e " ${DIM}IP:${NC} ${EXISTING_VMBR0_IP:-DHCP}" if ! confirm "Reconfigure vmbr0 (WAN)?"; then SKIP_WAN=true WAN_IFACE="$EXISTING_VMBR0" log_info "Keeping existing vmbr0 configuration." fi fi if [ "$SKIP_WAN" = false ]; then WAN_IFACE=$(prompt_choice "Enter WAN interface for vmbr0" "${EXISTING_VMBR0:-enp1s0f0}") if [ ! -d "/sys/class/net/$WAN_IFACE" ]; then log_error "Interface $WAN_IFACE does not exist!" if ! confirm "Continue anyway?" "n"; then return 1 fi fi echo "" echo -e "${BOLD}WAN IP Configuration:${NC}" echo -e " ${CYAN}1)${NC} Static IP" echo -e " ${CYAN}2)${NC} DHCP" WAN_IP_MODE=$(prompt_choice "Select" "1") if [ "$WAN_IP_MODE" == "1" ]; then WAN_IP=$(prompt_choice "Enter WAN IP address (e.g., 38.134.40.96/24)" "$EXISTING_VMBR0_IP") WAN_GW=$(prompt_choice "Enter WAN gateway (e.g., 38.134.40.1)" "") fi fi # ===== LAN (vmbr1) Configuration ===== echo "" echo -e "${BOLD}━━━ LAN Bridge (vmbr1) ━━━${NC}" if [ -n "$EXISTING_VMBR1" ]; then echo "" log_success "vmbr1 already configured:" echo -e " ${DIM}Interface:${NC} $EXISTING_VMBR1" echo -e " ${DIM}IP:${NC} ${EXISTING_VMBR1_IP:-DHCP}" if ! confirm "Reconfigure vmbr1 (LAN)?"; then SKIP_LAN=true LAN_IFACE="$EXISTING_VMBR1" log_info "Keeping existing vmbr1 configuration." fi fi if [ "$SKIP_LAN" = false ]; then LAN_IFACE=$(prompt_choice "Enter LAN interface for vmbr1" "${EXISTING_VMBR1:-enp1s0f1}") if [ ! -d "/sys/class/net/$LAN_IFACE" ]; then log_error "Interface $LAN_IFACE does not exist!" if ! confirm "Continue anyway?" "n"; then return 1 fi fi echo "" echo -e "${BOLD}LAN IP Configuration:${NC}" echo -e " ${CYAN}1)${NC} Static IP" echo -e " ${CYAN}2)${NC} DHCP" LAN_IP_MODE=$(prompt_choice "Select" "1") if [ "$LAN_IP_MODE" == "1" ]; then LAN_IP=$(prompt_choice "Enter LAN IP address (e.g., 10.1.1.7/16)" "$EXISTING_VMBR1_IP") fi # Ask about static route for internal network echo "" if confirm "Add static route for 10.0.0.0/8 network (internal/LAN traffic)?"; then ADD_LAN_ROUTE="y" # Try to guess the gateway from the LAN IP if [ -n "$LAN_IP" ]; then DEFAULT_LAN_GW=$(echo "$LAN_IP" | sed 's/\.[0-9]*\/.*/.1/') else DEFAULT_LAN_GW="10.1.0.1" fi LAN_ROUTE_GW=$(prompt_choice "Enter LAN gateway for 10.0.0.0/8 route" "$DEFAULT_LAN_GW") else ADD_LAN_ROUTE="n" fi else ADD_LAN_ROUTE="n" fi # ===== Summary and Apply ===== if [ "$SKIP_WAN" = true ] && [ "$SKIP_LAN" = true ]; then log_info "Both bridges already configured. No changes needed." NETWORK_CONFIGURED=true return 0 fi # If skipping WAN, load existing WAN config if [ "$SKIP_WAN" = true ]; then WAN_IFACE="$EXISTING_VMBR0" if [ -n "$EXISTING_VMBR0_IP" ]; then WAN_IP_MODE="1" WAN_IP="$EXISTING_VMBR0_IP" WAN_GW=$(grep -A10 "^iface vmbr0" $INTERFACES_FILE 2>/dev/null | grep "gateway" | awk '{print $2}' || echo "") else WAN_IP_MODE="2" fi fi # If skipping LAN, load existing LAN config if [ "$SKIP_LAN" = true ]; then LAN_IFACE="$EXISTING_VMBR1" if [ -n "$EXISTING_VMBR1_IP" ]; then LAN_IP_MODE="1" LAN_IP="$EXISTING_VMBR1_IP" else LAN_IP_MODE="2" fi # Check for existing route if grep -q "10.0.0.0/8" $INTERFACES_FILE 2>/dev/null; then ADD_LAN_ROUTE="y" LAN_ROUTE_GW=$(grep -A20 "^iface vmbr1" $INTERFACES_FILE 2>/dev/null | grep "10.0.0.0/8" | grep -oE '[0-9]+\.[0-9]+\.[0-9]+\.[0-9]+' | head -1 || echo "") fi fi echo "" log_info "Configuration summary:" echo " vmbr0 (WAN): $WAN_IFACE" if [ "$WAN_IP_MODE" == "1" ]; then echo " IP: $WAN_IP, Gateway: $WAN_GW" else echo " IP: DHCP" fi if [ "$SKIP_WAN" = true ]; then echo -e " ${DIM}(keeping existing)${NC}" fi echo " vmbr1 (LAN): $LAN_IFACE" if [ "$LAN_IP_MODE" == "1" ]; then echo " IP: $LAN_IP" else echo " IP: DHCP" fi if [[ "$ADD_LAN_ROUTE" =~ ^[Yy]$ ]]; then echo " Route: 10.0.0.0/8 via $LAN_ROUTE_GW" fi if [ "$SKIP_LAN" = true ]; then echo -e " ${DIM}(keeping existing)${NC}" fi echo "" if ! confirm "Apply this network configuration?"; then log_warn "Network configuration skipped." return 0 fi cp $INTERFACES_FILE "${INTERFACES_FILE}.bak.$(date +%Y%m%d%H%M%S)" log_info "Backed up current network configuration." generate_standard_interfaces "$WAN_IFACE" "$WAN_IP_MODE" "$WAN_IP" "$WAN_GW" "$LAN_IFACE" "$LAN_IP_MODE" "$LAN_IP" "$ADD_LAN_ROUTE" "$LAN_ROUTE_GW" log_success "Network bridges configured!" log_warn "Network changes will take effect after reboot or 'ifreload -a'" # Apply network changes now if confirm "Apply network changes now (ifreload -a)?"; then log_step "Applying network configuration..." ifreload -a 2>/dev/null || ifdown -a && ifup -a 2>/dev/null || true sleep 2 # Run connectivity checks check_network_connectivity fi NETWORK_CONFIGURED=true } function check_network_connectivity() { echo "" log_step "Checking network connectivity..." echo "" # Check WAN connectivity (1.1.1.1) echo -n " WAN (1.1.1.1): " if ping -c 2 -W 3 1.1.1.1 &>/dev/null; then echo -e "${GREEN}OK${NC}" else echo -e "${RED}FAILED${NC}" log_warn "WAN connectivity check failed. Check vmbr0 configuration." fi # Check LAN connectivity (10.1.0.5) echo -n " LAN (10.1.0.5): " if ping -c 2 -W 3 10.1.0.5 &>/dev/null; then echo -e "${GREEN}OK${NC}" else echo -e "${RED}FAILED${NC}" log_warn "LAN connectivity check failed. Check vmbr1 configuration and 10.0.0.0/8 route." fi echo "" } function generate_standard_interfaces() { local WAN_IFACE="$1" local WAN_IP_MODE="$2" local WAN_IP="$3" local WAN_GW="$4" local LAN_IFACE="$5" local LAN_IP_MODE="$6" local LAN_IP="$7" local ADD_ROUTE="$8" local ROUTE_GW="$9" cat > $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE </dev/null | grep bridge-ports | awk '{print $2}' || echo "") EXISTING_VMBR0_IP=$(grep -A5 "^iface vmbr0" $INTERFACES_FILE 2>/dev/null | grep "address" | awk '{print $2}' || echo "") EXISTING_VMBR1=$(grep -A5 "^iface vmbr1" $INTERFACES_FILE 2>/dev/null | grep bridge-ports | awk '{print $2}' || echo "") EXISTING_VMBR1_IP=$(grep -A5 "^iface vmbr1" $INTERFACES_FILE 2>/dev/null | grep "address" | awk '{print $2}' || echo "") # Try to detect trunk interface from existing VLAN config (e.g., "enp1s0f0.20" -> "enp1s0f0") EXISTING_TRUNK="" if [ -n "$EXISTING_VMBR0" ] && [[ "$EXISTING_VMBR0" == *.* ]]; then EXISTING_TRUNK="${EXISTING_VMBR0%.*}" fi SKIP_WAN=false SKIP_LAN=false TRUNK_IFACE=$(prompt_choice "Enter trunk interface (physical port)" "${EXISTING_TRUNK:-}") if [ -z "$TRUNK_IFACE" ]; then log_error "Trunk interface is required!" return 1 fi if [ ! -d "/sys/class/net/$TRUNK_IFACE" ]; then log_error "Interface $TRUNK_IFACE does not exist!" if ! confirm "Continue anyway?" "n"; then return 1 fi fi echo "" log_info "VLAN Configuration" echo -e " ${DIM}Common VLANs:${NC}" echo -e " ${DIM} VLAN 10 = LAN/Private network${NC}" echo -e " ${DIM} VLAN 20 = WAN/Public network${NC}" # ===== WAN (vmbr0) Configuration ===== echo "" echo -e "${BOLD}━━━ WAN Bridge (vmbr0) ━━━${NC}" if [ -n "$EXISTING_VMBR0" ]; then echo "" log_success "vmbr0 already configured:" echo -e " ${DIM}Bridge Port:${NC} $EXISTING_VMBR0" echo -e " ${DIM}IP:${NC} ${EXISTING_VMBR0_IP:-DHCP}" if ! confirm "Reconfigure vmbr0 (WAN)?"; then SKIP_WAN=true # Extract VLAN from existing config if [[ "$EXISTING_VMBR0" == *.* ]]; then WAN_VLAN="${EXISTING_VMBR0##*.}" fi log_info "Keeping existing vmbr0 configuration." fi fi if [ "$SKIP_WAN" = false ]; then WAN_VLAN=$(prompt_choice "Enter WAN VLAN ID" "20") echo "" echo -e "${BOLD}WAN (VLAN $WAN_VLAN) IP Configuration:${NC}" echo -e " ${CYAN}1)${NC} Static IP" echo -e " ${CYAN}2)${NC} DHCP" WAN_IP_MODE=$(prompt_choice "Select" "1") if [ "$WAN_IP_MODE" == "1" ]; then WAN_IP=$(prompt_choice "Enter WAN IP address (e.g., 108.165.121.14/24)" "$EXISTING_VMBR0_IP") WAN_GW=$(prompt_choice "Enter WAN gateway (e.g., 108.165.121.1)" "") fi fi # ===== LAN (vmbr1) Configuration ===== echo "" echo -e "${BOLD}━━━ LAN Bridge (vmbr1) ━━━${NC}" if [ -n "$EXISTING_VMBR1" ]; then echo "" log_success "vmbr1 already configured:" echo -e " ${DIM}Bridge Port:${NC} $EXISTING_VMBR1" echo -e " ${DIM}IP:${NC} ${EXISTING_VMBR1_IP:-DHCP}" if ! confirm "Reconfigure vmbr1 (LAN)?"; then SKIP_LAN=true # Extract VLAN from existing config if [[ "$EXISTING_VMBR1" == *.* ]]; then LAN_VLAN="${EXISTING_VMBR1##*.}" fi log_info "Keeping existing vmbr1 configuration." fi fi if [ "$SKIP_LAN" = false ]; then LAN_VLAN=$(prompt_choice "Enter LAN VLAN ID" "10") echo "" echo -e "${BOLD}LAN (VLAN $LAN_VLAN) IP Configuration:${NC}" echo -e " ${CYAN}1)${NC} Static IP" echo -e " ${CYAN}2)${NC} DHCP" LAN_IP_MODE=$(prompt_choice "Select" "2") if [ "$LAN_IP_MODE" == "1" ]; then LAN_IP=$(prompt_choice "Enter LAN IP address (e.g., 10.4.0.10/24)" "$EXISTING_VMBR1_IP") fi if confirm "Add route for 10.0.0.0/8 network?"; then ADD_LAN_ROUTE="y" LAN_ROUTE_GW=$(prompt_choice "Enter LAN gateway for 10.0.0.0/8 route" "10.4.0.1") else ADD_LAN_ROUTE="n" fi else ADD_LAN_ROUTE="n" fi # ===== Summary and Apply ===== if [ "$SKIP_WAN" = true ] && [ "$SKIP_LAN" = true ]; then log_info "Both bridges already configured. No changes needed." NETWORK_CONFIGURED=true return 0 fi # If skipping WAN, load existing WAN config if [ "$SKIP_WAN" = true ]; then if [ -n "$EXISTING_VMBR0_IP" ]; then WAN_IP_MODE="1" WAN_IP="$EXISTING_VMBR0_IP" WAN_GW=$(grep -A10 "^iface vmbr0" $INTERFACES_FILE 2>/dev/null | grep "gateway" | awk '{print $2}' || echo "") else WAN_IP_MODE="2" fi fi # If skipping LAN, load existing LAN config if [ "$SKIP_LAN" = true ]; then if [ -n "$EXISTING_VMBR1_IP" ]; then LAN_IP_MODE="1" LAN_IP="$EXISTING_VMBR1_IP" else LAN_IP_MODE="2" fi # Check for existing route if grep -q "10.0.0.0/8" $INTERFACES_FILE 2>/dev/null; then ADD_LAN_ROUTE="y" LAN_ROUTE_GW=$(grep -A15 "^iface vmbr1" $INTERFACES_FILE 2>/dev/null | grep "10.0.0.0/8" | grep -oP 'via \K[\d.]+' || echo "10.4.0.1") fi fi echo "" log_info "Configuration summary:" echo " Trunk Interface: $TRUNK_IFACE" echo "" echo " vmbr0 (WAN - VLAN $WAN_VLAN): ${TRUNK_IFACE}.${WAN_VLAN}" if [ "$WAN_IP_MODE" == "1" ]; then echo " IP: $WAN_IP, Gateway: $WAN_GW" else echo " IP: DHCP" fi if [ "$SKIP_WAN" = true ]; then echo -e " ${DIM}(keeping existing)${NC}" fi echo "" echo " vmbr1 (LAN - VLAN $LAN_VLAN): ${TRUNK_IFACE}.${LAN_VLAN}" if [ "$LAN_IP_MODE" == "1" ]; then echo " IP: $LAN_IP" else echo " IP: DHCP" fi if [[ "$ADD_LAN_ROUTE" =~ ^[Yy]$ ]]; then echo " Route: 10.0.0.0/8 via $LAN_ROUTE_GW" fi if [ "$SKIP_LAN" = true ]; then echo -e " ${DIM}(keeping existing)${NC}" fi echo "" if ! confirm "Apply this network configuration?"; then log_warn "Network configuration skipped." return 0 fi cp $INTERFACES_FILE "${INTERFACES_FILE}.bak.$(date +%Y%m%d%H%M%S)" log_info "Backed up current network configuration." generate_vlan_interfaces "$TRUNK_IFACE" "$WAN_VLAN" "$WAN_IP_MODE" "$WAN_IP" "$WAN_GW" "$LAN_VLAN" "$LAN_IP_MODE" "$LAN_IP" "$ADD_LAN_ROUTE" "$LAN_ROUTE_GW" log_success "VLAN network bridges configured!" log_warn "Network changes will take effect after reboot or 'ifreload -a'" # Apply network changes now if confirm "Apply network changes now (ifreload -a)?"; then log_step "Applying network configuration..." ifreload -a 2>/dev/null || ifdown -a && ifup -a 2>/dev/null || true sleep 2 # Run connectivity checks check_network_connectivity fi NETWORK_CONFIGURED=true } function generate_vlan_interfaces() { local TRUNK_IFACE="$1" local WAN_VLAN="$2" local WAN_IP_MODE="$3" local WAN_IP="$4" local WAN_GW="$5" local LAN_VLAN="$6" local LAN_IP_MODE="$7" local LAN_IP="$8" local ADD_LAN_ROUTE="$9" local LAN_ROUTE_GW="${10}" cat > $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE <> $INTERFACES_FILE < /dev/null; then log_info "Installing mdadm..." apt-get install -y mdadm # Verify installation if ! command -v mdadm &> /dev/null; then log_error "Failed to install mdadm. Please install it manually: apt-get install mdadm" return 1 fi log_success "mdadm installed successfully" fi EXISTING_ARRAYS=$(cat /proc/mdstat | grep "^md" | awk '{print "/dev/"$1}' || echo "") if [ -n "$EXISTING_ARRAYS" ]; then log_warn "Existing RAID arrays detected:" cat /proc/mdstat echo "" if ! confirm "Continue with RAID setup? (Existing arrays will not be modified)"; then return 0 fi fi list_available_disks echo -e "${BOLD}RAID Level Options:${NC}" echo "" echo -e " ${CYAN}1)${NC} RAID-1 ${DIM}(Mirroring, 2 disks, 50% capacity)${NC}" echo -e " ${CYAN}2)${NC} RAID-10 ${DIM}(Striped mirrors, 4+ disks, 50% capacity)${NC}" echo -e " ${CYAN}3)${NC} RAID-5 ${DIM}(Striping with parity, 3+ disks, ~66-87% capacity)${NC}" echo -e " ${CYAN}4)${NC} RAID-6 ${DIM}(Striping with double parity, 4+ disks)${NC}" echo -e " ${CYAN}5)${NC} Skip RAID setup" RAID_CHOICE=$(prompt_choice "Select RAID level" "1") case $RAID_CHOICE in 1) RAID_LEVEL="1" MIN_DISKS=2 ;; 2) RAID_LEVEL="10" MIN_DISKS=4 ;; 3) RAID_LEVEL="5" MIN_DISKS=3 ;; 4) RAID_LEVEL="6" MIN_DISKS=4 ;; 5) log_info "RAID setup skipped." return 0 ;; *) log_error "Invalid selection." return 1 ;; esac log_info "Selected RAID level: $RAID_LEVEL (minimum $MIN_DISKS disks required)" echo "" echo -e "${DIM}Enter the disks to use for the RAID array.${NC}" echo -e "${DIM}Example: /dev/sda /dev/sdb /dev/sdc /dev/sdd${NC}" echo -e "${DIM} or: sda sdb sdc sdd${NC}" DISK_INPUT=$(prompt_choice "Disks" "") DISKS=() for disk in $DISK_INPUT; do if [[ ! "$disk" == /dev/* ]]; then disk="/dev/$disk" fi DISKS+=("$disk") done DISK_COUNT=${#DISKS[@]} if [ $DISK_COUNT -lt $MIN_DISKS ]; then log_error "RAID-$RAID_LEVEL requires at least $MIN_DISKS disks. You provided $DISK_COUNT." return 1 fi for disk in "${DISKS[@]}"; do if [ ! -b "$disk" ]; then log_error "Disk $disk does not exist!" return 1 fi done echo "" log_warn "The following disks will be used:" for disk in "${DISKS[@]}"; do lsblk -d -o NAME,SIZE,MODEL "$disk" done echo "" log_error "WARNING: ALL DATA ON THESE DISKS WILL BE DESTROYED!" echo "" if ! confirm "Are you absolutely sure you want to proceed?" "n"; then log_info "RAID setup cancelled." return 0 fi CONFIRM_DESTROY=$(prompt_choice "Type 'DESTROY' to confirm data destruction" "") if [ "$CONFIRM_DESTROY" != "DESTROY" ]; then log_info "RAID setup cancelled." return 0 fi EXISTING_MD=$(ls /dev/md* 2>/dev/null | grep -oP 'md\d+' | sort -V | tail -1 || echo "") if [ -n "$EXISTING_MD" ]; then NEXT_MD_NUM=$((${EXISTING_MD#md} + 1)) else NEXT_MD_NUM=0 fi MD_DEVICE="/dev/md${NEXT_MD_NUM}" MD_DEVICE=$(prompt_choice "RAID device name" "$MD_DEVICE") echo "" log_info "Creating RAID-$RAID_LEVEL array on $MD_DEVICE..." for disk in "${DISKS[@]}"; do log_info "Wiping $disk..." wipefs -a "$disk" 2>/dev/null || true dd if=/dev/zero of="$disk" bs=1M count=100 2>/dev/null || true done DISK_STRING="${DISKS[*]}" log_info "Running: mdadm --create $MD_DEVICE --level=$RAID_LEVEL --raid-devices=$DISK_COUNT $DISK_STRING" mdadm --create "$MD_DEVICE" \ --level="$RAID_LEVEL" \ --raid-devices="$DISK_COUNT" \ --metadata=1.2 \ ${DISKS[@]} log_info "Waiting for array to initialize..." sleep 2 cat /proc/mdstat mkdir -p /etc/mdadm mdadm --detail --scan >> $MDADM_CONF update-initramfs -u log_success "RAID array $MD_DEVICE created successfully!" RAID_DEVICE="$MD_DEVICE" RAID_CONFIGURED=true } # ============================================ # LVM Configuration # ============================================ function setup_lvm() { echo "" echo -e "${BOLD}=== LVM Configuration ===${NC}" echo "" if ! command -v pvcreate &> /dev/null; then log_info "Installing lvm2..." apt-get update && apt-get install -y lvm2 fi log_info "Current LVM configuration:" echo "" echo "Physical Volumes:" pvs 2>/dev/null || echo " (none)" echo "" echo "Volume Groups:" vgs 2>/dev/null || echo " (none)" echo "" if [ -n "$RAID_DEVICE" ]; then log_info "Using newly created RAID device: $RAID_DEVICE" PV_DEVICE="$RAID_DEVICE" else echo -e "${BOLD}Available block devices:${NC}" lsblk -d -o NAME,SIZE,TYPE | grep -E "(disk|raid)" echo "" PV_DEVICE=$(prompt_choice "Enter device for Physical Volume (e.g., /dev/md0 or /dev/sda)" "") if [ -z "$PV_DEVICE" ]; then log_warn "LVM setup skipped." return 0 fi if [[ ! "$PV_DEVICE" == /dev/* ]]; then PV_DEVICE="/dev/$PV_DEVICE" fi fi if [ ! -b "$PV_DEVICE" ]; then log_error "Device $PV_DEVICE does not exist!" return 1 fi if pvs "$PV_DEVICE" &>/dev/null; then log_warn "$PV_DEVICE is already a Physical Volume." EXISTING_VG=$(pvs --noheadings -o vg_name "$PV_DEVICE" | tr -d ' ') if [ -n "$EXISTING_VG" ]; then log_info "It belongs to Volume Group: $EXISTING_VG" if [ "$EXISTING_VG" == "vg0" ]; then log_success "Already configured as expected (vg0). LVM setup complete!" LVM_CONFIGURED=true return 0 fi fi if ! confirm "Continue and add to vg0?"; then return 0 fi else log_info "Creating Physical Volume on $PV_DEVICE..." # Wait for udev to settle and device to be fully ready log_info "Waiting for device to settle..." udevadm settle --timeout=30 2>/dev/null || true # If this is a RAID device, check sync status and warn user if [[ "$PV_DEVICE" == /dev/md* ]]; then SYNC_STATUS=$(cat /proc/mdstat 2>/dev/null | grep -A1 "$(basename $PV_DEVICE)" | grep -oP '\d+%' || echo "") if [ -n "$SYNC_STATUS" ]; then log_warn "RAID array is still syncing ($SYNC_STATUS). LVM operations may be slow." log_info "You can continue, but operations will be faster after sync completes." fi fi pvcreate "$PV_DEVICE" log_success "Physical Volume created." fi if vgs vg0 &>/dev/null; then log_info "Volume Group 'vg0' already exists." log_info "Extending vg0 with $PV_DEVICE..." vgextend vg0 "$PV_DEVICE" 2>/dev/null || log_warn "Device may already be in vg0" else log_info "Creating Volume Group 'vg0' with $PV_DEVICE..." # Ensure udev has settled before vgcreate udevadm settle --timeout=30 2>/dev/null || true vgcreate vg0 "$PV_DEVICE" fi log_success "LVM configuration complete!" echo "" log_info "Final LVM status:" echo "" echo "Physical Volumes:" pvs echo "" echo "Volume Groups:" vgs echo "" LVM_CONFIGURED=true } # ============================================ # Proxmox Storage Configuration # ============================================ function setup_proxmox_storage() { echo "" echo -e "${BOLD}=== Proxmox Storage Configuration ===${NC}" echo "" log_info "Configuring Proxmox storage..." log_step "Updating 'local' storage to support images and ISO..." pvesm set local --content iso,vztmpl,snippets,backup,images 2>/dev/null || \ log_warn "Could not update local storage (may already be configured)" log_step "Updating 'local-lvm' storage to support images..." pvesm set local-lvm --content images,rootdir 2>/dev/null || \ log_warn "Could not update local-lvm storage (may already be configured)" echo "" if vgs vg0 &>/dev/null; then log_info "Setting up LVM storage on vg0..." if pvesm status | grep -q "VM_LVM"; then log_info "VM_LVM storage already exists." pvesm set VM_LVM --content images,rootdir 2>/dev/null || true else LVM_STORAGE_NAME=$(prompt_choice "Enter LVM storage name" "VM_LVM") log_step "Creating '$LVM_STORAGE_NAME' LVM storage on vg0..." # Use standard LVM (not LVM-thin) log_step "Adding LVM storage to Proxmox..." pvesm add lvm "$LVM_STORAGE_NAME" \ --vgname vg0 \ --content images,rootdir \ --nodes $(hostname) 2>/dev/null || { log_error "Failed to create LVM storage" } if pvesm status | grep -q "$LVM_STORAGE_NAME"; then log_success "LVM storage '$LVM_STORAGE_NAME' created!" # Store for API registration STORAGE_NAME="$LVM_STORAGE_NAME" fi fi else log_warn "vg0 not found. Skipping VM_LVM storage setup." log_info "Run the LVM configuration step first to create vg0." fi log_success "Storage configuration complete!" echo "" log_info "Current storage configuration:" pvesm status echo "" STORAGE_CONFIGURED=true } # ============================================ # Restore Mode (non-destructive storage recovery) # ============================================ # Used when the boot drive (OS + /etc/pve) was lost but the customer data # disks (RAID members / LVM PVs holding vm-*-disk-* volumes) are intact. # These functions only ASSEMBLE and ACTIVATE what already exists. function assemble_raid_array() { echo "" echo -e "${BOLD}=== RAID Recovery (assemble existing array) ===${NC}" echo "" if ! command -v mdadm &> /dev/null; then log_info "Installing mdadm..." apt-get install -y mdadm if ! command -v mdadm &> /dev/null; then log_error "Failed to install mdadm. Please install it manually: apt-get install mdadm" return 1 fi fi log_info "Scanning for existing RAID arrays (no disks will be modified)..." # Assemble anything with intact superblocks. Non-zero exit just means # "nothing new assembled" (already running or no arrays present). mdadm --assemble --scan 2>/dev/null || true udevadm settle --timeout=30 2>/dev/null || true ACTIVE_ARRAYS=$(grep "^md" /proc/mdstat 2>/dev/null | awk '{print "/dev/"$1}' || echo "") if [ -z "$ACTIVE_ARRAYS" ]; then log_warn "No RAID arrays found. If this node used RAID, the members were not detected." log_warn "Check 'mdadm --examine /dev/sdX' on the data disks before proceeding." log_info "If this node used plain disks (no RAID), this is expected — continuing." return 0 fi log_success "Active RAID arrays:" cat /proc/mdstat echo "" # Warn loudly about degraded arrays but keep going — data is still readable. for md in $ACTIVE_ARRAYS; do DEGRADED=$(mdadm --detail "$md" 2>/dev/null | grep -cE "degraded|faulty|removed" || true) if [ "$DEGRADED" -gt 0 ]; then log_warn "$md appears DEGRADED — restore can continue, but replace the failed member ASAP:" mdadm --detail "$md" | grep -E "State :|faulty|removed" || true fi done # Persist assembly so the array survives the next reboot of the new OS. mkdir -p /etc/mdadm touch "$MDADM_CONF" mdadm --detail --scan 2>/dev/null | while read -r line; do grep -qF "$line" "$MDADM_CONF" 2>/dev/null || echo "$line" >> "$MDADM_CONF" done update-initramfs -u RAID_DEVICE=$(echo "$ACTIVE_ARRAYS" | head -1) RAID_CONFIGURED=true log_success "RAID array(s) assembled and persisted to $MDADM_CONF" } function import_existing_lvm() { echo "" echo -e "${BOLD}=== LVM Recovery (activate existing volume group) ===${NC}" echo "" if ! command -v pvscan &> /dev/null; then log_info "Installing lvm2..." apt-get update && apt-get install -y lvm2 fi log_info "Rescanning LVM metadata (no volumes will be created or wiped)..." pvscan --cache 2>/dev/null || true vgscan --mknodes 2>/dev/null || true echo "" echo "Physical Volumes:" pvs 2>/dev/null || echo " (none)" echo "" echo "Volume Groups:" vgs 2>/dev/null || echo " (none)" echo "" VG_LIST=$(vgs --noheadings -o vg_name 2>/dev/null | tr -d ' ' || echo "") if [ -z "$VG_LIST" ]; then log_error "No volume groups found!" log_error "Either the RAID array is not assembled yet, or the data disks are not attached." log_error "Do NOT run the normal 'lvm' setup — pvcreate would destroy surviving customer data." return 1 fi if echo "$VG_LIST" | grep -qx "vg0"; then DISCOVERED_VG="vg0" elif [ "$(echo "$VG_LIST" | wc -l)" -eq 1 ]; then DISCOVERED_VG="$VG_LIST" log_warn "Volume group is named '$DISCOVERED_VG' (expected 'vg0')." else log_info "Multiple volume groups found:" echo "$VG_LIST" DISCOVERED_VG=$(prompt_choice "Which VG holds the VM disks?" "vg0") fi log_info "Activating volume group '$DISCOVERED_VG'..." vgchange -ay "$DISCOVERED_VG" udevadm settle --timeout=30 2>/dev/null || true VM_DISK_COUNT=$(lvs --noheadings -o lv_name "$DISCOVERED_VG" 2>/dev/null | grep -c "vm-.*-disk-" || true) echo "" if [ "$VM_DISK_COUNT" -gt 0 ]; then log_success "Found ${VM_DISK_COUNT} VM disk volume(s) in '$DISCOVERED_VG':" lvs -o lv_name,lv_size,lv_attr "$DISCOVERED_VG" | grep -E "LV|vm-" || true else log_warn "Volume group '$DISCOVERED_VG' is active but contains NO vm-*-disk-* volumes." log_warn "Verify you activated the right VG before restoring VMs from the admin panel." fi LVM_CONFIGURED=true log_success "LVM recovery complete — '$DISCOVERED_VG' is active." } function restore_proxmox_storage() { echo "" echo -e "${BOLD}=== Proxmox Storage Recovery (re-register existing VG) ===${NC}" echo "" if [ -z "$DISCOVERED_VG" ]; then log_error "No volume group discovered — run the LVM recovery step first." return 1 fi log_step "Updating 'local' storage to support images, snippets and ISO..." pvesm set local --content iso,vztmpl,snippets,backup,images 2>/dev/null || \ log_warn "Could not update local storage (may already be configured)" # The storage ID must be EXACTLY what the DartNode DB has for this node — # every service's disk config references ":vm--disk-N". DEFAULT_STORAGE_NAME="${TOKEN_STORAGE_NAME:-VM_LVM}" if [ -n "$TOKEN_STORAGE_NAME" ]; then log_info "DartNode DB expects storage name: ${BOLD}${TOKEN_STORAGE_NAME}${NC}" else log_warn "DartNode DB has no existing storage name for this node (new node?)." fi RESTORE_STORAGE_NAME=$(prompt_choice "Proxmox storage ID to map onto '$DISCOVERED_VG'" "$DEFAULT_STORAGE_NAME") if pvesm status 2>/dev/null | awk '{print $1}' | grep -qx "$RESTORE_STORAGE_NAME"; then log_info "Storage '$RESTORE_STORAGE_NAME' already registered." pvesm set "$RESTORE_STORAGE_NAME" --content images,rootdir 2>/dev/null || true else log_step "Registering LVM storage '$RESTORE_STORAGE_NAME' on VG '$DISCOVERED_VG'..." pvesm add lvm "$RESTORE_STORAGE_NAME" \ --vgname "$DISCOVERED_VG" \ --content images,rootdir \ --nodes "$(hostname)" || { log_error "Failed to register LVM storage" return 1 } fi if pvesm status | grep -q "$RESTORE_STORAGE_NAME"; then log_success "Storage '$RESTORE_STORAGE_NAME' is registered." STORAGE_NAME="$RESTORE_STORAGE_NAME" fi echo "" log_info "Existing VM volumes visible to Proxmox:" pvesm list "$RESTORE_STORAGE_NAME" 2>/dev/null | head -40 || true echo "" STORAGE_CONFIGURED=true log_success "Storage recovery complete!" } function setup_nfs_storage() { echo "" echo -e "${BOLD}=== NFS Storage Configuration ===${NC}" echo "" log_info "Configure NFS shares for cloud-init images, backups, and VM storage." echo "" if ! confirm "Configure NFS storage?"; then log_info "NFS configuration skipped." return 0 fi # Check existing NFS storage EXISTING_NFS=$(pvesm status --enabled 2>/dev/null | grep "nfs" | awk '{print $1}' || echo "") if [ -n "$EXISTING_NFS" ]; then log_info "Existing NFS storage:" echo "$EXISTING_NFS" | while read storage; do echo " - $storage" done echo "" fi # Preferred path: mirror the NFS shares the rest of the fleet mounts # (pulled live from an online node via /node/config). No guessed defaults. if [ -n "$TOKEN_NFS_STORAGES" ]; then log_info "NFS shares in use across the DartNode fleet:" while IFS='|' read -r name server export path content options; do [ -z "$name" ] && continue echo " - ${name} -> ${server}:${export}" done <<< "$TOKEN_NFS_STORAGES" echo "" if confirm "Add these NFS shares (existing ones are skipped)?"; then while IFS='|' read -r name server export path content options; do [ -z "$name" ] && continue if pvesm status 2>/dev/null | awk '{print $1}' | grep -qx "$name"; then log_info "${name} storage already exists. Skipping..." continue fi log_step "Adding '${name}' (${server}:${export})..." PVESM_CMD=(pvesm add nfs "$name" --server "$server" --export "$export" --content "$content") [ -n "$path" ] && PVESM_CMD+=(--path "$path") [ -n "$options" ] && PVESM_CMD+=(--options "$options") PVESM_OUT=$("${PVESM_CMD[@]}" 2>&1) && \ log_success "${name} storage added!" || { log_warn "Failed to add ${name} storage:" echo "$PVESM_OUT" | sed 's/^/ /' } done <<< "$TOKEN_NFS_STORAGES" log_success "NFS storage configuration complete!" echo "" log_info "Current NFS storage:" pvesm status --enabled 2>/dev/null | grep -E "(Name|nfs)" || echo " (none)" return 0 fi log_info "Falling back to manual NFS configuration..." echo "" else log_warn "Could not fetch the fleet's NFS shares from the API — using manual configuration." echo "" fi # Cloud-init images NFS share echo "" log_step "1/3: Cloud-init Images NFS Share" if pvesm status | grep -q "cloud-init-img"; then log_info "cloud-init-img storage already exists. Skipping..." else CI_NFS_SERVER=$(prompt_choice "NFS Server IP for cloud-init images" "10.1.0.5") CI_NFS_EXPORT=$(prompt_choice "NFS Export path" "/volume1/CloudInitImages") PVESM_OUT=$(pvesm add nfs cloud-init-img \ --server "$CI_NFS_SERVER" \ --export "$CI_NFS_EXPORT" \ --path /mnt/pve/cloud-init-img \ --content snippets,iso,images 2>&1) && \ log_success "cloud-init-img storage added!" || { log_warn "Failed to add cloud-init-img storage:" echo "$PVESM_OUT" | sed 's/^/ /' } fi # Backups NFS share echo "" log_step "2/3: Backups NFS Share" if pvesm status | grep -q "dn-backups"; then log_info "dn-backups storage already exists. Skipping..." else BK_NFS_SERVER=$(prompt_choice "NFS Server IP for backups" "10.1.0.5") BK_NFS_EXPORT=$(prompt_choice "NFS Export path" "/volume1/PXEBackups") PVESM_OUT=$(pvesm add nfs dn-backups \ --server "$BK_NFS_SERVER" \ --export "$BK_NFS_EXPORT" \ --path /mnt/pve/dn-backups \ --content images,backup,iso 2>&1) && \ log_success "dn-backups storage added!" || { log_warn "Failed to add dn-backups storage:" echo "$PVESM_OUT" | sed 's/^/ /' } fi # Additional VM storage (store-1) echo "" log_step "3/3: Additional VM Storage NFS Share (optional)" if pvesm status | grep -q "store-1"; then log_info "store-1 storage already exists. Skipping..." else if confirm "Add store-1 NFS share for additional VM storage?"; then ST_NFS_SERVER=$(prompt_choice "NFS Server IP for store-1" "10.1.0.214") ST_NFS_EXPORT=$(prompt_choice "NFS Export path" "/mnt/HOU-1-S1-P1/store-1-vms") PVESM_OUT=$(pvesm add nfs store-1 \ --server "$ST_NFS_SERVER" \ --export "$ST_NFS_EXPORT" \ --path /mnt/pve/store-1 \ --content images 2>&1) && \ log_success "store-1 storage added!" || { log_warn "Failed to add store-1 storage:" echo "$PVESM_OUT" | sed 's/^/ /' } fi fi log_success "NFS storage configuration complete!" echo "" log_info "Current NFS storage:" pvesm status --enabled 2>/dev/null | grep -E "(Name|nfs)" || echo " (none)" } # ============================================ # DartNode Daemon Installation # ============================================ function install_dartnode_daemon() { echo "" echo -e "${BOLD}=== DartNode Stats Daemon ===${NC}" echo "" if command -v dartnode &> /dev/null; then log_info "DartNode daemon is already installed." CURRENT_VERSION=$(dartnode version 2>/dev/null | head -1 || echo "unknown") log_info "Current version: $CURRENT_VERSION" if confirm "Update to latest version?"; then log_info "Updating DartNode daemon..." dartnode update fi if confirm "Reconfigure daemon settings?"; then if [ -n "$REGISTRATION_TOKEN" ]; then # Use token-based configuration configure_daemon_with_token else dartnode config fi fi if ! systemctl is-active --quiet dn-daemon; then log_info "Starting DartNode daemon..." dartnode start fi DAEMON_INSTALLED=true return 0 fi if ! confirm "Install DartNode PVE Stats Daemon?"; then log_info "Daemon installation skipped." return 0 fi log_info "Downloading and installing DartNode daemon..." echo "" # Download and run the installer WITHOUT the token flag # Registration with the API is handled in Step 7 once all node data is collected curl -sSL https://pkg.dev.snaju.com/dn/dartnode.sh | bash -s install if ! command -v dartnode &> /dev/null; then log_error "DartNode daemon installation failed!" return 1 fi log_success "DartNode daemon installed!" # Configure the daemon with token settings if [ -n "$REGISTRATION_TOKEN" ]; then log_step "Configuring daemon with Redis and node settings..." configure_daemon_with_token # Start the daemon log_step "Starting DartNode daemon service..." systemctl enable dn-daemon 2>/dev/null || true systemctl restart dn-daemon 2>/dev/null || dartnode start 2>/dev/null || true if systemctl is-active --quiet dn-daemon; then log_success "DartNode daemon is running!" else log_warn "Daemon may not have started. Check with: systemctl status dn-daemon" fi fi DAEMON_INSTALLED=true } function configure_daemon_with_token() { # Fetch config from API if not already fetched if [ -z "$TOKEN_NODE_NAME" ]; then if ! fetch_token_config "$REGISTRATION_TOKEN"; then log_error "Failed to fetch configuration from API" return 1 fi fi local config_dir="/etc/dn-daemon" mkdir -p "$config_dir" # Use database ID if available (re-setup), otherwise use node name temporarily # The node_id will be updated to the DB ID after registration in Step 7 local DAEMON_NODE_ID="${TOKEN_NODE_DB_ID:-${TOKEN_NODE_NAME}}" log_info "Writing daemon configuration..." echo "" echo " Node ID: ${DAEMON_NODE_ID}" echo " Node Name: ${TOKEN_NODE_NAME}" echo " Mode: ${TOKEN_MODE:-pve}" echo " Redis Host: ${TOKEN_REDIS_HOST}" echo " Redis Port: ${TOKEN_REDIS_PORT}" echo " Redis DB: ${TOKEN_REDIS_DB}" echo " Redis Password: ${TOKEN_REDIS_PASSWORD:0:8}..." echo " VNC/SSH Secret: ${TOKEN_AUTH_SECRET:0:12}..." echo "" # Write main config cat > "${config_dir}/config.json" <<-EOF { "mode": "${TOKEN_MODE:-pve}", "redis_host": "${TOKEN_REDIS_HOST}", "redis_port": ${TOKEN_REDIS_PORT}, "redis_password": "${TOKEN_REDIS_PASSWORD}", "redis_db": ${TOKEN_REDIS_DB}, "node_id": "${DAEMON_NODE_ID}", "auto_update": true, "update_interval": 300, "health_reporting": true, "debug": false } EOF # Write PVE-specific config cat > "${config_dir}/pve.json" <<-EOF { "poll_interval": 5, "stats_ttl": 120, "storage_ttl": 300, "qmp_socket_dir": "/var/run/qemu-server", "backup_path": "/mnt/pve/dn-backups", "backup_interval": 300, "worker_count": 4, "vm_timeout": 8, "collect_timeout": 60, "vnc_enabled": true, "vnc_port": 5700, "vnc_socket_dir": "/var/run/qemu-server", "vnc_auth_secret": "${TOKEN_AUTH_SECRET}", "ssh_enabled": true, "ssh_auth_secret": "${TOKEN_AUTH_SECRET}" } EOF log_success "Daemon configuration saved to ${config_dir}/" log_info "Config files: config.json, pve.json" } # ============================================ # API Registration # ============================================ function register_with_api() { echo "" echo -e "${BOLD}=== Register Node with DartNode ===${NC}" echo "" # If we have a token, use it if [ -n "$REGISTRATION_TOKEN" ]; then register_with_token return $? fi log_info "No registration token provided." log_info "You can get a node token from the admin panel." echo "" NODE_TOKEN=$(prompt_choice "Enter node registration token (or press Enter to skip)" "") if [ -z "$NODE_TOKEN" ]; then log_info "Node registration skipped. You can register manually later." return 0 fi REGISTRATION_TOKEN="$NODE_TOKEN" register_with_token } function register_with_token() { log_info "Registering node with DartNode..." # Validate token first if ! validate_token "$REGISTRATION_TOKEN"; then return 1 fi # Get system information log_step "Gathering system information..." HOSTNAME=$(hostname) WAN_IP=$(ip -4 addr show vmbr0 2>/dev/null | grep -oP 'inet \K[\d.]+' | head -1 || echo "") LAN_IP=$(ip -4 addr show vmbr1 2>/dev/null | grep -oP 'inet \K[\d.]+' | head -1 || echo "") if [ -z "$WAN_IP" ]; then # Try to get any IP WAN_IP=$(hostname -I | awk '{print $1}') fi log_info "Detected WAN IP: $WAN_IP" WAN_IP=$(prompt_choice "Confirm node IP" "$WAN_IP") # Create Proxmox API user and token log_step "Creating Proxmox API credentials..." API_PASSWORD=$(openssl rand -hex 12) # Remove existing user/token if present, then create fresh pveum user token remove dn-api@pve dartnode-web 2>/dev/null || true pveum user delete dn-api@pve 2>/dev/null || true pveum user add dn-api@pve --password "$API_PASSWORD" 2>/dev/null || true pveum aclmod / -user dn-api@pve -role Administrator 2>/dev/null || true log_step "Creating API token..." TOKEN_JSON=$(pveum user token add dn-api@pve dartnode-web --comment 'DartNode API Token' --privsep=0 --output-format json 2>/dev/null) # Parse the token value from JSON - try multiple methods if command -v python3 &> /dev/null; then API_SECRET=$(echo "$TOKEN_JSON" | python3 -c "import sys,json; print(json.load(sys.stdin).get('value',''))" 2>/dev/null || echo "") fi # Fallback to grep if python didn't work if [ -z "$API_SECRET" ]; then API_SECRET=$(echo "$TOKEN_JSON" | grep -oP '"value"\s*:\s*"\K[^"]+' 2>/dev/null || echo "") fi # Another fallback using sed if [ -z "$API_SECRET" ]; then API_SECRET=$(echo "$TOKEN_JSON" | sed -n 's/.*"value"[[:space:]]*:[[:space:]]*"\([^"]*\)".*/\1/p' 2>/dev/null || echo "") fi # Debug output echo "[DEBUG] TOKEN_JSON: $TOKEN_JSON" echo "[DEBUG] Extracted API_SECRET: ${API_SECRET:0:20}..." if [ -z "$API_SECRET" ]; then log_error "Failed to create API token!" log_error "Token JSON output: $TOKEN_JSON" return 1 fi pveum aclmod / -token 'dn-api@pve!dartnode-web' -role Administrator 2>/dev/null || true log_success "API credentials created!" # Install SSH public key log_step "Installing DartNode SSH key..." # Rotated automation key, 4096-bit (SHA256:8tD9kxY6VpvLlDqIgCCGfG8wsarcr0wHEbn2GmNVFrs). # Must stay in sync with ProxmoxService::SSH_PUB_KEY / SSH_KEY_B64. DN_SSH_KEY="ssh-rsa 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 dartnode-automation-2026" # Unique body of the CURRENT key — match on key material, not the comment. # The old check grepped for "dartnode", so a host still carrying the REVOKED # "dartnode"-commented key would be treated as already-provisioned and never # get the rotated key (and the revoked key would never be removed). DN_SSH_KEY_BODY="$(echo "$DN_SSH_KEY" | awk '{print $2}')" mkdir -p ~/.ssh chmod 700 ~/.ssh touch ~/.ssh/authorized_keys chmod 600 ~/.ssh/authorized_keys # Purge the revoked key (2048-bit, ...Byq2P) if present, then install the # current key if its exact body isn't already authorized. sed -i '/AAABAQCfetdLfmwhiKEZ/d' ~/.ssh/authorized_keys 2>/dev/null || true if ! grep -qF "$DN_SSH_KEY_BODY" ~/.ssh/authorized_keys 2>/dev/null; then echo "$DN_SSH_KEY" >> ~/.ssh/authorized_keys fi log_success "SSH key installed!" # Register with API log_step "Completing registration with DartNode API..." local storage_name="${STORAGE_NAME:-VM_LVM}" local storage_type="raw" # nodeUsername is the Proxmox API user (dn-api@pve) # nodePassword is the password for that user # apiTokenId is the full token ID (user!tokenname) # apiSecret is the token value from Proxmox local node_username="dn-api@pve" local api_token_id="dn-api@pve!dartnode-web" # Debug output log_info "Credentials being sent:" echo " Node Username (API User): $node_username" echo " Node Password: ${API_PASSWORD:0:8}..." echo " API Token ID: $api_token_id" echo " API Secret: ${API_SECRET:0:20}..." echo " Storage Name: $storage_name" local response response=$(curl -sS -X POST "${API_BASE}/node/register" \ -H "Content-Type: application/json" \ -d "{\"token\":\"${REGISTRATION_TOKEN}\",\"nodeIp\":\"${WAN_IP}\",\"apiTokenId\":\"${api_token_id}\",\"apiSecret\":\"${API_SECRET}\",\"nodeUsername\":\"${node_username}\",\"nodePassword\":\"${API_PASSWORD}\",\"storageName\":\"${storage_name}\",\"storageType\":\"${storage_type}\"}" 2>/dev/null) if echo "$response" | grep -q '"success":true'; then local node_id node_name node_id=$(json_get "$response" "nodeId") node_name=$(json_get "$response" "nodeName") log_success "Node registered successfully!" echo "" log_info "Node DB ID: $node_id" log_info "Node Name: $node_name" # Update daemon config with the database ID as node_id # This ensures the daemon uses the DB ID for Redis keys if [ -n "$node_id" ] && [ -f "/etc/dn-daemon/config.json" ]; then log_step "Updating daemon config with database ID..." if command -v python3 &> /dev/null; then python3 -c " import json with open('/etc/dn-daemon/config.json', 'r') as f: cfg = json.load(f) cfg['node_id'] = '$node_id' with open('/etc/dn-daemon/config.json', 'w') as f: json.dump(cfg, f, indent=4) " 2>/dev/null else sed -i "s/\"node_id\": \"[^\"]*\"/\"node_id\": \"$node_id\"/" /etc/dn-daemon/config.json 2>/dev/null fi log_success "Daemon config updated with node_id: $node_id" # Restart daemon to pick up new node_id systemctl restart dn-daemon 2>/dev/null || dartnode restart 2>/dev/null || true log_info "Daemon restarted with new node ID" fi REGISTERED_WITH_API=true else local error_msg error_msg=$(json_get "$response" "error") log_error "Registration failed: ${error_msg:-Unknown error}" return 1 fi } # ============================================ # Summary # ============================================ function show_summary() { echo "" echo -e "${BOLD}============================================${NC}" echo -e "${BOLD} Deployment Summary${NC}" echo -e "${BOLD}============================================${NC}" echo "" if [ "$NETWORK_CONFIGURED" == "true" ]; then log_success "Network bridges configured (vmbr0, vmbr1)" else log_warn "Network bridges: Not configured" fi if [ "$RAID_CONFIGURED" == "true" ]; then log_success "RAID array created: $RAID_DEVICE" else log_warn "RAID array: Not configured" fi if [ "$LVM_CONFIGURED" == "true" ]; then log_success "LVM configured with vg0" else log_warn "LVM: Not configured" fi if [ "$STORAGE_CONFIGURED" == "true" ]; then log_success "Proxmox storage types configured" else log_warn "Proxmox storage: Not configured" fi if [ "$DAEMON_INSTALLED" == "true" ]; then log_success "DartNode daemon installed" else log_warn "DartNode daemon: Not installed" fi if [ "$REGISTERED_WITH_API" == "true" ]; then log_success "Registered with DartNode" else log_warn "DartNode registration: Not completed" fi echo "" echo -e "${BOLD}============================================${NC}" echo -e "${BOLD} Next Steps${NC}" echo -e "${BOLD}============================================${NC}" echo "" if [ "$NETWORK_CONFIGURED" == "true" ]; then echo "1. Reboot the node or run 'ifreload -a' to apply network changes" echo "" fi if [ "$DAEMON_INSTALLED" == "true" ] && [ "$REGISTERED_WITH_API" != "true" ]; then echo "2. Configure the DartNode daemon:" echo " dartnode config" echo "" fi if [ "$REGISTERED_WITH_API" != "true" ]; then echo -e "${CYAN}3. Add this node to DartNode Admin Panel:${NC}" echo "" echo -e " ${BOLD}https://vader.dartnode.net${NC}" echo "" echo " - Navigate to Infrastructure > Nodes" echo " - Click 'Provision New Node'" echo " - Enter this node's details and connection info" echo "" fi if [ "$NETWORK_CONFIGURED" == "true" ] || [ "$RAID_CONFIGURED" == "true" ]; then echo -e "${YELLOW}NOTE: A reboot is recommended to apply all changes.${NC}" echo "" if confirm "Reboot now?"; then log_info "Rebooting in 5 seconds..." sleep 5 reboot fi fi } # ============================================ # Interactive Deployment # ============================================ function run_interactive() { header log_info "This script will configure a Proxmox node for DartNode." echo "" # If no token provided, ask for one if [ -z "$REGISTRATION_TOKEN" ]; then echo -e "${BOLD}A registration token is required to proceed.${NC}" echo -e "${DIM}Get your token from: https://vader.dartnode.net/infra/vps/nodes/new${NC}" echo "" REGISTRATION_TOKEN=$(prompt_choice "Enter your registration token" "") if [ -z "$REGISTRATION_TOKEN" ]; then log_error "No token provided. Cannot continue." exit 1 fi fi log_success "Registration token: ${REGISTRATION_TOKEN:0:16}..." echo "" if ! validate_token "$REGISTRATION_TOKEN"; then log_error "Invalid token. Exiting." exit 1 fi fetch_token_config "$REGISTRATION_TOKEN" echo "" echo -e "${BOLD}┌─────────────────────────────────────────────────────────────┐${NC}" echo -e "${BOLD}│ Deployment Steps │${NC}" echo -e "${BOLD}├─────────────────────────────────────────────────────────────┤${NC}" echo -e "${BOLD}│${NC} ${CYAN}1.${NC} Configure network bridges (vmbr0, vmbr1) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}2.${NC} Set up RAID array (RAID-1, RAID-5, RAID-6, RAID-10) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}3.${NC} Configure LVM (Physical Volume + Volume Group) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}4.${NC} Configure Proxmox storage types ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}5.${NC} Configure NFS storage (optional) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}6.${NC} Install DartNode stats daemon ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}7.${NC} Register with DartNode API ${BOLD}│${NC}" echo -e "${BOLD}└─────────────────────────────────────────────────────────────┘${NC}" echo "" echo -e "${DIM}Each step is optional - you can skip any step during the process.${NC}" if ! confirm "Ready to begin deployment?"; then log_info "Deployment cancelled." exit 0 fi # Step 1: Network bridges echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 1 of 7: Network Configuration${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" setup_network_bridges # Step 2: RAID array echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 2 of 7: RAID Configuration${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" setup_raid_array # Step 3: LVM echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 3 of 7: LVM Configuration${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" setup_lvm # Step 4: Proxmox storage echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 4 of 7: Proxmox Storage Configuration${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" setup_proxmox_storage # Step 5: NFS storage echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 5 of 7: NFS Storage Configuration${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" setup_nfs_storage # Step 6: DartNode daemon echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 6 of 7: DartNode Daemon Installation${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" install_dartnode_daemon # Step 7: Register with API echo "" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" echo -e "${BOLD} STEP 7 of 7: DartNode Registration${NC}" echo -e "${BOLD}${CYAN}━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━${NC}" register_with_api # Show summary show_summary } # ============================================ # Restore Deployment (lost boot drive, data intact) # ============================================ function run_restore() { header echo -e "${BOLD}${YELLOW}RESTORE MODE${NC}" echo "" log_info "For a node that lost its OS/boot drive (Proxmox configs gone) while the" log_info "customer data disks (RAID/LVM with vm-*-disk-* volumes) survived." echo "" log_warn "This mode NEVER creates/wipes RAID arrays or LVM volumes — it only" log_warn "assembles and re-registers what already exists." echo "" if [ -z "$REGISTRATION_TOKEN" ]; then echo -e "${BOLD}A registration token is required to proceed.${NC}" echo -e "${DIM}Create a re-setup token from the node's page: https://vader.dartnode.net/infra/vps/nodes${NC}" echo "" REGISTRATION_TOKEN=$(prompt_choice "Enter your registration token" "") if [ -z "$REGISTRATION_TOKEN" ]; then log_error "No token provided. Cannot continue." exit 1 fi fi if ! validate_token "$REGISTRATION_TOKEN"; then log_error "Invalid token. Exiting." exit 1 fi fetch_token_config "$REGISTRATION_TOKEN" echo "" # Re-registration matches on the Proxmox hostname (VPSHost.nodeId), and every # service pins data['nodeid'] to that hostname. A mismatch would register a # brand-new node and orphan every service. CURRENT_HOSTNAME=$(hostname) if [ -n "$TOKEN_NODE_NAME" ] && [ "$CURRENT_HOSTNAME" != "$TOKEN_NODE_NAME" ]; then log_error "Hostname mismatch: this host is '$CURRENT_HOSTNAME' but the token is for '$TOKEN_NODE_NAME'." log_error "Set the hostname to match before restoring:" log_error " hostnamectl set-hostname $TOKEN_NODE_NAME (then re-check /etc/hosts and pvecm status)" exit 1 fi if [ -z "$TOKEN_NODE_DB_ID" ] || [ "$TOKEN_NODE_DB_ID" == "null" ]; then log_warn "No existing node record found in DartNode for '$TOKEN_NODE_NAME'." log_warn "Restore expects a node that is already in the DB (its services reference it)." if ! confirm "Continue anyway?" "n"; then exit 1 fi fi echo -e "${BOLD}┌─────────────────────────────────────────────────────────────┐${NC}" echo -e "${BOLD}│ Restore Steps │${NC}" echo -e "${BOLD}├─────────────────────────────────────────────────────────────┤${NC}" echo -e "${BOLD}│${NC} ${CYAN}1.${NC} Configure network bridges (vmbr0, vmbr1) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}2.${NC} Assemble existing RAID array (non-destructive) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}3.${NC} Activate existing LVM volume group ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}4.${NC} Re-register Proxmox storage on the existing VG ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}5.${NC} Configure NFS storage (cloud-init images, backups) ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}6.${NC} Install DartNode stats daemon ${BOLD}│${NC}" echo -e "${BOLD}│${NC} ${CYAN}7.${NC} Re-register with DartNode API ${BOLD}│${NC}" echo -e "${BOLD}└─────────────────────────────────────────────────────────────┘${NC}" echo "" echo -e "${DIM}After this completes, restore the VMs from the admin panel:${NC}" echo -e "${DIM} vader.dartnode.net > Infra > VPS Nodes > ${TOKEN_NODE_NAME:-} > Tools > Restore VMs${NC}" if ! confirm "Ready to begin restore?"; then log_info "Restore cancelled." exit 0 fi echo "" echo -e "${BOLD}${CYAN}━━ STEP 1 of 7: Network Configuration ━━${NC}" setup_network_bridges echo "" echo -e "${BOLD}${CYAN}━━ STEP 2 of 7: RAID Recovery ━━${NC}" assemble_raid_array echo "" echo -e "${BOLD}${CYAN}━━ STEP 3 of 7: LVM Recovery ━━${NC}" import_existing_lvm echo "" echo -e "${BOLD}${CYAN}━━ STEP 4 of 7: Proxmox Storage Recovery ━━${NC}" restore_proxmox_storage echo "" echo -e "${BOLD}${CYAN}━━ STEP 5 of 7: NFS Storage Configuration ━━${NC}" setup_nfs_storage echo "" echo -e "${BOLD}${CYAN}━━ STEP 6 of 7: DartNode Daemon Installation ━━${NC}" install_dartnode_daemon echo "" echo -e "${BOLD}${CYAN}━━ STEP 7 of 7: DartNode Registration ━━${NC}" register_with_api show_summary echo "" echo -e "${BOLD}${GREEN}Node restore (host side) complete.${NC}" echo -e "Next: open ${BOLD}https://vader.dartnode.net/infra/vps/nodes${NC} > this node > ${BOLD}Tools > Restore VMs${NC}" echo -e "to scan for surviving disks and rebuild the VM configurations." } function usage() { echo "Usage: dn-deploy.sh [command] [options]" echo "" echo "Commands:" echo " (none) - Run interactive deployment" echo " restore - Restore a node after boot-drive loss (non-destructive:" echo " assembles existing RAID, activates existing LVM VG," echo " re-registers storage — never wipes data disks)" echo " network - Configure network bridges only" echo " raid - Configure RAID array only" echo " raid-assemble - Assemble an existing RAID array only (non-destructive)" echo " lvm - Configure LVM only" echo " lvm-import - Activate an existing LVM VG only (non-destructive)" echo " storage - Configure Proxmox storage types only" echo " nfs - Configure NFS storage only" echo " daemon - Install DartNode daemon only" echo " register - Register with DartNode only" echo " help - Show this help" echo "" echo "Options:" echo " --token - Registration token from DartNode admin panel" echo " --restore - Same as the 'restore' command" echo "" echo "Restore a node whose boot drive died but data disks survived:" echo " curl -sSL https://pkg.dev.snaju.com/dn/dn-deploy.sh | bash -s -- --restore --token YOUR_TOKEN" echo "" echo "Interactive deployment (recommended):" echo " curl -sSL https://pkg.dev.snaju.com/dn/dn-deploy.sh | bash" echo "" echo "With registration token:" echo " curl -sSL https://pkg.dev.snaju.com/dn/dn-deploy.sh | bash -s -- --token YOUR_TOKEN" echo "" } # ============================================ # Parse Arguments # ============================================ COMMAND="" while [[ $# -gt 0 ]]; do case $1 in --token) REGISTRATION_TOKEN="$2" shift 2 ;; --token=*) REGISTRATION_TOKEN="${1#*=}" shift ;; --restore) RESTORE_MODE=true shift ;; --) shift ;; *) if [ -z "$COMMAND" ]; then COMMAND="$1" fi shift ;; esac done COMMAND=${COMMAND:-interactive} # --restore flag overrides the default interactive flow if [ "$RESTORE_MODE" == "true" ] && [ "$COMMAND" == "interactive" ]; then COMMAND="restore" fi # ============================================ # Main # ============================================ check_root check_proxmox fix_proxmox_repos # Standalone commands still benefit from API-provided config (fleet NFS shares, # expected storage name) when a token was passed — fetch it up front. The # interactive/restore flows fetch it themselves after validating. case ${COMMAND} in interactive|restore|help|--help|-h) ;; *) if [ -n "$REGISTRATION_TOKEN" ]; then fetch_token_config "$REGISTRATION_TOKEN" || true fi ;; esac case ${COMMAND} in interactive) run_interactive ;; restore) RESTORE_MODE=true run_restore ;; network) header setup_network_bridges ;; raid) header setup_raid_array ;; raid-assemble) header assemble_raid_array ;; lvm) header setup_lvm ;; lvm-import) header import_existing_lvm ;; storage) header setup_proxmox_storage ;; nfs) header setup_nfs_storage ;; daemon) header install_dartnode_daemon ;; register) header register_with_api ;; help|--help|-h) header usage ;; *) header echo "Unknown command: $COMMAND" echo "" usage exit 1 ;; esac