darkforge/tests/proxmox/create-vm.sh

#!/bin/bash
# ============================================================================
# DarkForge — Proxmox VM Creation Script
# ============================================================================
# Run this on the Proxmox host to create an Arch Linux test VM.
#
# Requirements:
#   - Proxmox VE 8.x or 9.x
#   - ~100GB free on a storage pool
#   - Internet access
#
# Usage:
#   bash create-vm.sh                    # use defaults
#   bash create-vm.sh --vmid 200         # custom VM ID
#   bash create-vm.sh --storage local-lvm  # custom storage
# ============================================================================

set -euo pipefail

# --- Configuration (override via environment or flags) -----------------------
VMID="${VMID:-900}"
VM_NAME="${VM_NAME:-darkforge-test}"
STORAGE="${STORAGE:-local-lvm}"
DISK_SIZE="${DISK_SIZE:-100G}"
RAM="${RAM:-28672}"       # 16GB
CORES="${CORES:-18}"
BRIDGE="${BRIDGE:-vmbr0}"

# Arch Linux cloud image
ARCH_IMG_URL="https://geo.mirror.pkgbuild.com/images/latest/Arch-Linux-x86_64-cloudimg.qcow2"
ARCH_IMG_FILE="/var/lib/vz/template/iso/arch-cloudimg.qcow2"

# Git repo to clone inside the VM
DARKFORGE_REPO="gitea@git.dannyhaslund.dk:danny8632/darkforge.git"
# Fallback if SSH key isn't available in the VM
DARKFORGE_REPO_HTTPS="https://git.dannyhaslund.dk/danny8632/darkforge.git"

# Parse args
for arg in "$@"; do
    case "$arg" in
        --vmid=*)    VMID="${arg#*=}" ;;
        --storage=*) STORAGE="${arg#*=}" ;;
        --cores=*)   CORES="${arg#*=}" ;;
        --ram=*)     RAM="${arg#*=}" ;;
        --bridge=*)  BRIDGE="${arg#*=}" ;;
    esac
done

echo "═══════════════════════════════════════════════"
echo " DarkForge Test VM Creator"
echo " VMID: ${VMID} | Cores: ${CORES} | RAM: ${RAM}MB"
echo " Storage: ${STORAGE} | Disk: ${DISK_SIZE}"
echo "═══════════════════════════════════════════════"
echo ""

# --- Check if VM already exists ----------------------------------------------
if qm status "${VMID}" &>/dev/null; then
    echo "VM ${VMID} already exists."
    read -p "Destroy and recreate? [y/N] " confirm
    if [[ "${confirm}" =~ ^[Yy]$ ]]; then
        qm stop "${VMID}" 2>/dev/null || true
        sleep 3
        qm destroy "${VMID}" --purge
        echo "Old VM destroyed."
    else
        echo "Aborted."
        exit 0
    fi
fi

# --- Download Arch cloud image if not cached ---------------------------------
if [ ! -f "${ARCH_IMG_FILE}" ]; then
    echo ">>> Downloading Arch Linux cloud image..."
    mkdir -p "$(dirname "${ARCH_IMG_FILE}")"
    wget -q --show-progress -O "${ARCH_IMG_FILE}" "${ARCH_IMG_URL}"
    echo ">>> Downloaded: ${ARCH_IMG_FILE}"
else
    echo ">>> Using cached image: ${ARCH_IMG_FILE}"
fi

# --- Create the VM -----------------------------------------------------------
echo ">>> Creating VM ${VMID}..."

qm create "${VMID}" \
    --name "${VM_NAME}" \
    --ostype l26 \
    --machine q35 \
    --bios ovmf \
    --cpu host \
    --cores "${CORES}" \
    --memory "${RAM}" \
    --net0 "virtio,bridge=${BRIDGE}" \
    --agent enabled=1 \
    --serial0 socket \
    --vga serial0

# Add EFI disk (required for OVMF BIOS)
qm set "${VMID}" --efidisk0 "${STORAGE}:1,efitype=4m,pre-enrolled-keys=0"

# Import the cloud image as the boot disk
echo ">>> Importing cloud image as boot disk..."
qm importdisk "${VMID}" "${ARCH_IMG_FILE}" "${STORAGE}" --format qcow2 2>/dev/null
qm set "${VMID}" --scsi0 "${STORAGE}:vm-${VMID}-disk-1,size=${DISK_SIZE}"
qm set "${VMID}" --boot order=scsi0
qm set "${VMID}" --scsihw virtio-scsi-single

# --- Configure cloud-init ----------------------------------------------------
echo ">>> Configuring cloud-init..."

qm set "${VMID}" --ide2 "${STORAGE}:cloudinit"

# Don't use --ciuser/--cipassword — they conflict with the snippet on Arch.
# We handle ALL user creation in the cloud-init snippet runcmd instead.
qm set "${VMID}" --ipconfig0 "ip=dhcp"

# Enable nested virtualization (for QEMU-in-QEMU boot tests)
qm set "${VMID}" --args "-cpu host,+vmx"

# Resize the disk to our desired size
echo ">>> Resizing disk to ${DISK_SIZE}..."
qm resize "${VMID}" scsi0 "${DISK_SIZE}" 2>/dev/null || true

# --- Generate cloud-init user-data snippet -----------------------------------
# NOTE: The Arch Linux cloud image has quirks with user/password handling.
# We do EVERYTHING via runcmd to guarantee it works regardless of cloud-init
# version or Proxmox's cloud-init integration behavior.
SNIPPET_DIR="/var/lib/vz/snippets"
mkdir -p "${SNIPPET_DIR}"

cat > "${SNIPPET_DIR}/darkforge-test-init.yaml" << 'CLOUDINIT'
#cloud-config

# Disable the default user module to avoid conflicts
# We create our user manually via runcmd below
users: []

ssh_pwauth: true

write_files:
  - path: /etc/ssh/sshd_config.d/99-darkforge.conf
    content: |
      PasswordAuthentication yes
      PermitRootLogin yes

runcmd:
  # --- USER SETUP (do this first, before anything else) ----------------------
  # Set root password so we always have a fallback login
  - echo 'root:darkforge' | chpasswd

  # Create the darkforge user if it doesn't exist
  - id darkforge &>/dev/null || useradd -m -G wheel -s /bin/bash darkforge
  - echo 'darkforge:darkforge' | chpasswd

  # Give darkforge sudo/wheel access without password
  - echo 'darkforge ALL=(ALL) NOPASSWD:ALL' > /etc/sudoers.d/darkforge
  - chmod 440 /etc/sudoers.d/darkforge

  # Enable and restart sshd
  - systemctl enable sshd
  - systemctl restart sshd

  # --- DISK RESIZE -----------------------------------------------------------
  - growpart /dev/sda 2 || growpart /dev/vda 2 || true
  - resize2fs /dev/sda2 || resize2fs /dev/vda2 || btrfs filesystem resize max / || true

  # --- PACKAGE INSTALL -------------------------------------------------------
  - pacman-key --init
  - pacman-key --populate archlinux
  - pacman -Syu --noconfirm
  - pacman -S --noconfirm --needed base-devel git wget curl rust cargo qemu-full edk2-ovmf squashfs-tools xorriso dosfstools mtools python bc rsync openssh tmux

  # --- CLONE PROJECT ---------------------------------------------------------
  # Try HTTPS (with SSL_NO_VERIFY for self-hosted Gitea), fall back to GitHub
  - |
    su - darkforge -c '
      cd /home/darkforge
      GIT_SSL_NO_VERIFY=true git clone --recurse-submodules https://git.dannyhaslund.dk/danny8632/darkforge.git 2>/dev/null || \
        git clone --recurse-submodules https://github.com/danny8632/darkforge.git 2>/dev/null || \
        echo "CLONE FAILED — run manually: GIT_SSL_NO_VERIFY=true git clone --recurse-submodules https://git.dannyhaslund.dk/danny8632/darkforge.git ~/darkforge"
    '

  # --- INSTALL CONVENIENCE COMMAND -------------------------------------------
  # NOTE: heredoc inside cloud-init runcmd must NOT be indented or the
  # shebang gets leading spaces and the script won't execute properly.
  - |
    cat > /usr/local/bin/darkforge-test << 'DTEOF'
#!/bin/bash
SCRIPT="/home/darkforge/darkforge/tests/proxmox/run-in-vm.sh"
if [ ! -f "$SCRIPT" ]; then
    echo "ERROR: Test script not found at: $SCRIPT"
    echo ""
    echo "The git clone probably failed during provisioning."
    echo "Clone manually:"
    echo "  git clone --recurse-submodules https://git.dannyhaslund.dk/danny8632/darkforge.git ~/darkforge"
    echo ""
    echo "Then run:  darkforge-test"
    exit 1
fi
ARGS="$*"
# Kill any existing tmux session first
tmux kill-session -t darkforge 2>/dev/null || true
exec tmux new-session -d -s darkforge \
    "bash ${SCRIPT} --tmux ${ARGS}; echo ''; echo 'Tests finished. Press Enter to close.'; read" \; \
    attach-session -t darkforge
DTEOF
    chmod +x /usr/local/bin/darkforge-test

  # --- SIGNAL DONE -----------------------------------------------------------
  - touch /home/darkforge/.provisioned
  - chown darkforge:darkforge /home/darkforge/.provisioned
CLOUDINIT

qm set "${VMID}" --cicustom "user=local:snippets/darkforge-test-init.yaml"

# --- Start the VM ------------------------------------------------------------
echo ""
echo ">>> Starting VM ${VMID}..."
qm start "${VMID}"

echo ""
echo "═══════════════════════════════════════════════"
echo " VM ${VMID} created and starting."
echo ""
echo " Cloud-init will install packages and clone the repo."
echo " Wait ~5 min for provisioning, then SSH in to run tests."
echo ""
echo " Get the VM IP:"
echo "   qm guest cmd ${VMID} network-get-interfaces | grep -oP '\"ip-address\":\\s*\"\\K[0-9.]+'"
echo ""
echo " SSH in:"
echo "   ssh darkforge@<IP>        (password: darkforge)"
echo ""
echo " Run tests in a tmux session (detachable):"
echo "   darkforge-test             # starts tests in tmux"
echo "   darkforge-test --quick     # fast mode (30 min)"
echo ""
echo " Detach from tmux:  Ctrl+B then D"
echo " Reattach later:    tmux attach -t darkforge"
echo ""
echo " Collect report:"
echo "   scp darkforge@<IP>:~/darkforge/tests/report.* ./"
echo "═══════════════════════════════════════════════"