Files
loc_az_hci/scripts/fix/recreate-vms-with-ssh-keys.sh
defiQUG c39465c2bd
Some checks failed
Test / test (push) Has been cancelled
Initial commit: loc_az_hci (smom-dbis-138 excluded via .gitignore)
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-02-08 09:04:46 -08:00

270 lines
8.2 KiB
Bash
Executable File

#!/bin/bash
source ~/.bashrc
# Recreate VMs with Proper SSH Key Configuration
# Destroys existing VMs and recreates them with cloud-init SSH keys
set -euo pipefail
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} $1"
}
log_step() {
echo ""
echo -e "${BLUE}========================================${NC}"
echo -e "${BLUE}$1${NC}"
echo -e "${BLUE}========================================${NC}"
echo ""
}
PVE_USERNAME="${PVE_USERNAME:-root@pam}"
PVE_PASSWORD="${PVE_ROOT_PASS:-}"
PROXMOX_URL="${PROXMOX_ML110_URL:-https://192.168.1.206:8006}"
PROXMOX_NODE="${PROXMOX_NODE:-pve}"
TEMPLATE_VMID="${TEMPLATE_VMID:-9000}"
SSH_KEY_FILE="$HOME/.ssh/id_ed25519_proxmox.pub"
GATEWAY="${GATEWAY:-192.168.1.254}"
# VM definitions: vmid name ip cores memory disk_size
VMS=(
"100 cloudflare-tunnel 192.168.1.188 2 2048 20"
"101 k3s-master 192.168.1.60 4 4096 40"
"102 git-server 192.168.1.121 2 2048 30"
"103 observability 192.168.1.82 2 2048 30"
)
get_api_token() {
local response=$(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
}
destroy_vm() {
local vmid=$1
local name=$2
log_info "Destroying VM $vmid ($name)..."
local tokens=$(get_api_token)
if [ -z "$tokens" ]; then
log_error "Failed to authenticate with Proxmox"
return 1
fi
local ticket=$(echo "$tokens" | cut -d'|' -f1)
local csrf_token=$(echo "$tokens" | cut -d'|' -f2)
# Stop VM if running
local 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', 'stopped'))" 2>/dev/null || echo "stopped")
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
sleep 5
fi
# Delete VM
log_info "Deleting VM $vmid..."
curl -s -k -X DELETE \
-H "Cookie: PVEAuthCookie=$ticket" \
-H "CSRFPreventionToken: $csrf_token" \
"$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid" > /dev/null
log_info "VM $vmid destroyed"
}
create_vm_with_ssh() {
local vmid=$1
local name=$2
local ip=$3
local cores=$4
local memory=$5
local disk_size=$6
log_info "Creating VM $vmid: $name with SSH keys"
local tokens=$(get_api_token)
if [ -z "$tokens" ]; then
log_error "Failed to authenticate with Proxmox"
return 1
fi
local ticket=$(echo "$tokens" | cut -d'|' -f1)
local csrf_token=$(echo "$tokens" | cut -d'|' -f2)
# Read SSH public key
if [ ! -f "$SSH_KEY_FILE" ]; then
log_error "SSH key file not found: $SSH_KEY_FILE"
return 1
fi
local ssh_key_content=$(cat "$SSH_KEY_FILE")
local ssh_key_b64=$(echo "$ssh_key_content" | base64 -w 0)
# Check if template exists
local template_check=$(curl -s -k -H "Cookie: PVEAuthCookie=$ticket" \
-H "CSRFPreventionToken: $csrf_token" \
"$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$TEMPLATE_VMID/status/current" 2>&1)
if ! echo "$template_check" | grep -q '"data"'; then
log_error "Template VM $TEMPLATE_VMID not found"
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, waiting for clone to complete..."
sleep 10
# Configure VM
log_info "Configuring VM $vmid..."
# Set resources
curl -s -k -X POST \
-H "Cookie: PVEAuthCookie=$ticket" \
-H "CSRFPreventionToken: $csrf_token" \
-d "cores=$cores" \
-d "memory=$memory" \
"$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null
# Configure cloud-init with SSH keys
log_info "Configuring cloud-init with SSH keys..."
curl -s -k -X POST \
-H "Cookie: PVEAuthCookie=$ticket" \
-H "CSRFPreventionToken: $csrf_token" \
--data-urlencode "ipconfig0=ip=$ip/24,gw=$GATEWAY" \
--data-urlencode "ciuser=ubuntu" \
--data-urlencode "sshkeys=$ssh_key_b64" \
--data-urlencode "agent=1" \
"$PROXMOX_URL/api2/json/nodes/$PROXMOX_NODE/qemu/$vmid/config" > /dev/null
log_info "✓ VM $vmid configured with SSH keys"
# Start VM
log_info "Starting 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/start" > /dev/null
log_info "✓ VM $vmid started"
}
main() {
log_step "Recreate VMs with SSH Key Configuration"
if [ ! -f "$SSH_KEY_FILE" ]; then
log_error "SSH key file not found: $SSH_KEY_FILE"
exit 1
fi
log_warn "This will DESTROY and RECREATE all 4 VMs (100-103)"
log_warn "All data on these VMs will be lost!"
echo ""
read -p "Are you sure you want to continue? (yes/no): " confirm
if [ "$confirm" != "yes" ]; then
log_info "Cancelled"
exit 0
fi
# Destroy existing VMs
log_step "Step 1: Destroying Existing VMs"
for vm_spec in "${VMS[@]}"; do
read -r vmid name ip cores memory disk_size <<< "$vm_spec"
destroy_vm "$vmid" "$name" || log_warn "Failed to destroy VM $vmid"
done
sleep 5
# Create new VMs with SSH keys
log_step "Step 2: Creating VMs with SSH Keys"
for vm_spec in "${VMS[@]}"; do
read -r vmid name ip cores memory disk_size <<< "$vm_spec"
create_vm_with_ssh "$vmid" "$name" "$ip" "$cores" "$memory" "$disk_size" || {
log_error "Failed to create VM $vmid"
continue
}
done
log_step "Step 3: Waiting for VMs to Boot"
log_info "Waiting 60 seconds for VMs to boot and apply cloud-init..."
sleep 60
log_step "Step 4: Testing SSH Access"
log_info "Testing SSH access to VMs..."
local all_ok=true
for vm_spec in "${VMS[@]}"; do
read -r vmid name ip cores memory disk_size <<< "$vm_spec"
if ssh -i "${SSH_KEY_FILE%.pub}" -o ConnectTimeout=10 -o StrictHostKeyChecking=no ubuntu@$ip "echo 'SSH OK' && hostname" &>/dev/null; then
log_info " ✓ VM $vmid ($name) at $ip: SSH working"
else
log_error " ✗ VM $vmid ($name) at $ip: SSH not working"
all_ok=false
fi
done
if [ "$all_ok" = true ]; then
log_info ""
log_info "✓ All VMs recreated successfully with SSH access!"
log_info "You can now run: ./scripts/deploy/complete-all-next-steps.sh"
else
log_warn ""
log_warn "Some VMs may need more time to boot. Wait a few minutes and test again."
fi
}
main "$@"