loc_az_hci/scripts/deploy/recreate-vms-smaller-disks.sh

#!/bin/bash
source ~/.bashrc
# Recreate VMs with Smaller Disk Sizes
# Stops, deletes, and recreates VMs with optimized disk sizes

set -e

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
PROJECT_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"

# Load environment variables
if [ -f "$PROJECT_ROOT/.env" ]; then
    set -a
    source <(grep -v '^#' "$PROJECT_ROOT/.env" | grep -v '^$' | sed 's/#.*$//' | grep '=')
    set +a
fi

# Colors
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'

log_info() {
    echo -e "${GREEN}[INFO]${NC} $1"
}

log_warn() {
    echo -e "${YELLOW}[WARN]${NC} $1"
}

log_error() {
    echo -e "${RED}[ERROR]${NC} $1"
}

log_step() {
    echo -e "\n${BLUE}=== $1 ===${NC}"
}

# Proxmox configuration
PROXMOX_URL="${PROXMOX_ML110_URL:-https://192.168.1.206:8006}"
PVE_USERNAME="${PVE_USERNAME:-root@pam}"
PVE_PASSWORD="${PVE_ROOT_PASS:-}"
PROXMOX_NODE="${PROXMOX_NODE:-pve}"
TEMPLATE_VMID="${TEMPLATE_VMID:-9000}"

get_api_token() {
    local response=$(curl -s -k --connect-timeout 10 --max-time 15 \
        -d "username=$PVE_USERNAME&password=$PVE_PASSWORD" \
        "$PROXMOX_URL/api2/json/access/ticket" 2>&1)
    
    if echo "$response" | grep -q '"data"'; then
        local ticket=$(echo "$response" | grep -o '"ticket":"[^"]*' | cut -d'"' -f4)
        local csrf_token=$(echo "$response" | grep -o '"CSRFPreventionToken":"[^"]*' | cut -d'"' -f4)
        echo "$ticket|$csrf_token"
    else
        echo ""
    fi
}

stop_and_delete_vm() {
    local vmid=$1
    local name=$2
    
    log_info "Stopping VM $vmid ($name)..."
    
    local tokens=$(get_api_token)
    local ticket=$(echo "$tokens" | cut -d'|' -f1)
    local csrf_token=$(echo "$tokens" | cut -d'|' -f2)
    
    # Check if VM exists
    local vm_status=$(curl -s -k -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/status/current" 2>&1)
    
    if ! echo "$vm_status" | grep -q '"data"'; then
        log_warn "VM $vmid does not exist, skipping"
        return 0
    fi
    
    # Stop VM if running
    local status=$(echo "$vm_status" | python3 -c "import sys, json; print(json.load(sys.stdin).get('data', {}).get('status', 'unknown'))" 2>/dev/null)
    
    if [ "$status" = "running" ]; then
        log_info "Stopping VM $vmid..."
        curl -s -k -X POST -H "Cookie: PVEAuthCookie=$ticket" \
            -H "CSRFPreventionToken: $csrf_token" \
            "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/status/stop" > /dev/null
        
        # Wait for VM to stop
        local wait_count=0
        while [ $wait_count -lt 30 ]; do
            sleep 2
            local current_status=$(curl -s -k -H "Cookie: PVEAuthCookie=$ticket" \
                -H "CSRFPreventionToken: $csrf_token" \
                "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/status/current" | \
                python3 -c "import sys, json; print(json.load(sys.stdin).get('data', {}).get('status', 'unknown'))" 2>/dev/null)
            
            if [ "$current_status" = "stopped" ]; then
                break
            fi
            wait_count=$((wait_count + 1))
        done
        
        if [ $wait_count -ge 30 ]; then
            log_error "VM $vmid did not stop in time"
            return 1
        fi
    fi
    
    # Delete VM
    log_info "Deleting VM $vmid..."
    local delete_response=$(curl -s -k -X DELETE \
        -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid" 2>&1)
    
    if echo "$delete_response" | grep -q '"data"'; then
        log_info "VM $vmid deleted successfully"
        return 0
    else
        log_error "Failed to delete VM $vmid: $delete_response"
        return 1
    fi
}

create_vm_with_smaller_disk() {
    local vmid=$1
    local name=$2
    local ip=$3
    local gateway=$4
    local cores=$5
    local memory=$6
    local disk=$7
    local bridge=$8
    
    log_info "Creating VM $vmid ($name) with ${disk} disk..."
    
    local tokens=$(get_api_token)
    local ticket=$(echo "$tokens" | cut -d'|' -f1)
    local csrf_token=$(echo "$tokens" | cut -d'|' -f2)
    
    if [ -z "$ticket" ] || [ -z "$csrf_token" ]; then
        log_error "Failed to get API tokens"
        return 1
    fi
    
    # Clone VM from template
    log_info "Cloning from template $TEMPLATE_VMID..."
    local clone_response=$(curl -s -k -X POST \
        -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        -d "newid=$vmid" \
        -d "name=$name" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$TEMPLATE_VMID/clone" 2>&1)
    
    if ! echo "$clone_response" | grep -q '"data"'; then
        log_error "Failed to clone VM: $clone_response"
        return 1
    fi
    
    log_info "VM cloned successfully, waiting for clone to complete..."
    sleep 5
    
    # Wait for clone to finish
    local wait_count=0
    while [ $wait_count -lt 30 ]; do
        local vm_check=$(curl -s -k -H "Cookie: PVEAuthCookie=$ticket" \
            -H "CSRFPreventionToken: $csrf_token" \
            "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" 2>&1)
        
        if echo "$vm_check" | grep -q '"data"'; then
            break
        fi
        sleep 2
        wait_count=$((wait_count + 1))
    done
    
    # Configure VM with smaller disk
    log_info "Configuring VM $vmid (CPU: $cores, RAM: ${memory}MB, Disk: $disk)..."
    
    # Stop VM if it started automatically
    curl -s -k -X POST -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/status/stop" > /dev/null 2>&1
    sleep 3
    
    # Get current disk configuration
    local current_config=$(curl -s -k -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config")
    
    local current_disk=$(echo "$current_config" | python3 -c "import sys, json; d=json.load(sys.stdin).get('data', {}); print(d.get('scsi0', ''))" 2>/dev/null)
    
    # Extract storage pool from current disk or use default
    local storage_pool="local-lvm"
    if echo "$current_disk" | grep -q ':'; then
        storage_pool=$(echo "$current_disk" | cut -d':' -f1)
    fi
    
    # Delete old disk and create new smaller one
    log_info "Removing old disk and creating new ${disk} disk..."
    
    # Remove old disk
    curl -s -k -X PUT -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        -d "scsi0=" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null 2>&1
    
    sleep 2
    
    # Create new disk with smaller size using the disk creation endpoint
    log_info "Creating new ${disk} disk..."
    local disk_create=$(curl -s -k -X POST \
        -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        -d "size=$disk" \
        -d "format=raw" \
        -d "storage=$storage_pool" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" 2>&1)
    
    # If that doesn't work, try setting it directly in config
    if ! echo "$disk_create" | grep -q '"data"'; then
        log_info "Trying alternative method: setting disk in config..."
        curl -s -k -X PUT -H "Cookie: PVEAuthCookie=$ticket" \
            -H "CSRFPreventionToken: $csrf_token" \
            -d "scsi0=$storage_pool:vm-$vmid-disk-0,iothread=1,size=$disk" \
            "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null 2>&1
    fi
    
    # Configure CPU and memory
    curl -s -k -X PUT -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        -d "cores=$cores" \
        -d "memory=$memory" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null
    
    # Configure network
    curl -s -k -X PUT -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        -d "net0=virtio,bridge=$bridge,firewall=1" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null
    
    # Configure QEMU Guest Agent
    curl -s -k -X PUT -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        -d "agent=1" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null
    
    # Configure cloud-init
    log_info "Configuring cloud-init (user: ubuntu, IP: $ip/24)..."
    curl -s -k -X PUT -H "Cookie: PVEAuthCookie=$ticket" \
        -H "CSRFPreventionToken: $csrf_token" \
        --data-urlencode "ipconfig0=ip=$ip/24,gw=$gateway" \
        -d "ciuser=ubuntu" \
        "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null
    
    log_info "VM $vmid configured successfully with ${disk} disk"
    return 0
}

main() {
    log_step "Recreating VMs with Smaller Disk Sizes"
    
    if [ -z "$PVE_PASSWORD" ]; then
        log_error "PVE_ROOT_PASS not set in .env"
        exit 1
    fi
    
    log_warn "This will DELETE and RECREATE all VMs with smaller disks!"
    log_warn "All data on these VMs will be lost!"
    
    # Check for --yes flag to skip confirmation
    if [ "$1" != "--yes" ] && [ "$1" != "-y" ]; then
        echo ""
        read -p "Are you sure you want to continue? (yes/no): " confirm
        
        if [ "$confirm" != "yes" ]; then
            log_info "Cancelled by user"
            exit 0
        fi
    else
        log_info "Auto-confirmed (--yes flag provided)"
    fi
    
    # VM definitions with smaller disk sizes
    # Format: vmid name ip gateway cores memory_mb disk_size bridge
    local vms=(
        "100 cloudflare-tunnel 192.168.1.60 192.168.1.254 2 4096 20G vmbr0"
        "101 k3s-master 192.168.1.188 192.168.1.254 4 8192 40G vmbr0"
        "102 git-server 192.168.1.121 192.168.1.254 4 8192 50G vmbr0"
        "103 observability 192.168.1.82 192.168.1.254 4 8192 100G vmbr0"
    )
    
    # Step 1: Stop and delete existing VMs
    log_step "Step 1: Stopping and Deleting Existing VMs"
    for vm_spec in "${vms[@]}"; do
        read -r vmid name ip gateway cores memory disk bridge <<< "$vm_spec"
        stop_and_delete_vm "$vmid" "$name"
        sleep 2
    done
    
    # Step 2: Recreate VMs with smaller disks
    log_step "Step 2: Creating VMs with Smaller Disks"
    for vm_spec in "${vms[@]}"; do
        read -r vmid name ip gateway cores memory disk bridge <<< "$vm_spec"
        create_vm_with_smaller_disk "$vmid" "$name" "$ip" "$gateway" "$cores" "$memory" "$disk" "$bridge"
        sleep 3
    done
    
    # Step 3: Start all VMs
    log_step "Step 3: Starting All VMs"
    local tokens=$(get_api_token)
    local ticket=$(echo "$tokens" | cut -d'|' -f1)
    local csrf_token=$(echo "$tokens" | cut -d'|' -f2)
    
    for vm_spec in "${vms[@]}"; do
        read -r vmid name ip gateway cores memory disk bridge <<< "$vm_spec"
        log_info "Starting VM $vmid ($name)..."
        curl -s -k -X POST -H "Cookie: PVEAuthCookie=$ticket" \
            -H "CSRFPreventionToken: $csrf_token" \
            "$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/status/start" > /dev/null
        sleep 2
    done
    
    log_step "Recreation Complete!"
    log_info "All VMs recreated with smaller disk sizes:"
    log_info "  VM 100: 20G (was 40G)"
    log_info "  VM 101: 40G (was 80G)"
    log_info "  VM 102: 50G (was 100G)"
    log_info "  VM 103: 100G (was 200G)"
    log_info "Total saved: 210GB"
}

main "$@"