proxmox/scripts/audit-proxmox-rpc-storage.sh

#!/usr/bin/env bash
# Audit storage node restrictions vs RPC VMID placement.
# Runs per-host: each VMID is checked on its actual Proxmox host (no cross-node config lookup).
#
# Usage:
#   ./scripts/audit-proxmox-rpc-storage.sh
#   ./scripts/audit-proxmox-rpc-storage.sh --vmid 2301
#   PROXMOX_HOST=192.168.11.10 ./scripts/audit-proxmox-rpc-storage.sh   # single-host mode (legacy)

set -euo pipefail

# Load shared project environment and VMID placement helper.
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
PROJECT_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
source "${PROJECT_ROOT}/scripts/lib/load-project-env.sh"

# SSH user for shell (PROXMOX_USER in .env may be root@pam for API)
PROXMOX_SSH_USER="${PROXMOX_SSH_USER:-${PROXMOX_USER:-root}}"
[[ "$PROXMOX_SSH_USER" == *"@"* ]] && PROXMOX_SSH_USER="root"

RPC_VMIDS=(2101 2201 2301 2303 2304 2305 2306 2307 2308 2400 2401 2402 2403)
RPC_NODES=()
for vmid in "${RPC_VMIDS[@]}"; do
  RPC_NODES+=("${vmid}:$(get_host_for_vmid "$vmid")")
done

# If PROXMOX_HOST is set, run single-host legacy mode (one host only)
PROXMOX_HOST="${PROXMOX_HOST:-}"
TARGET_VMIDS=()

usage() {
  cat <<'EOF'
Usage: ./scripts/audit-proxmox-rpc-storage.sh [--vmid <N>]

Options:
  --vmid <N>   Limit the VMID section to one VMID; repeatable
EOF
}

while [[ $# -gt 0 ]]; do
  case "$1" in
    --vmid)
      [[ $# -ge 2 ]] || { usage >&2; exit 2; }
      TARGET_VMIDS+=("$2")
      shift 2
      ;;
    -h|--help)
      usage
      exit 0
      ;;
    *)
      echo "Unknown argument: $1" >&2
      usage >&2
      exit 2
      ;;
  esac
done

selected_vmid() {
  local vmid="$1"
  [[ ${#TARGET_VMIDS[@]} -eq 0 ]] && return 0
  local wanted
  for wanted in "${TARGET_VMIDS[@]}"; do
    [[ "$vmid" == "$wanted" ]] && return 0
  done
  return 1
}

ssh_pve() {
  local host="$1"
  shift
  ssh -o StrictHostKeyChecking=no -o BatchMode=yes -o ConnectTimeout=5 "${PROXMOX_SSH_USER}@${host}" "$@"
}

if [[ -n "${PROXMOX_HOST:-}" ]]; then
  echo "=== Proxmox RPC Storage Audit (single host: ${PROXMOX_HOST}) ==="
  echo ""
  NODE="$(ssh_pve "$PROXMOX_HOST" "hostname")"
  echo "Node name: ${NODE}"
  echo ""
  echo "=== Storages active on this node (pvesm) ==="
  ssh_pve "$PROXMOX_HOST" "pvesm status" | sed 's/^/  /'
  echo ""
  echo "=== storage.cfg: storages with node restrictions ==="
  ssh_pve "$PROXMOX_HOST" "awk '
    /^dir: /{s=\$2; t=\"dir\"; nodes=\"\"}
    /^lvmthin: /{s=\$2; t=\"lvmthin\"; nodes=\"\"}
    /^[[:space:]]*nodes /{nodes=\$2}
    /^[[:space:]]*nodes /{print t \":\" s \" nodes=\" nodes}
  ' /etc/pve/storage.cfg" | sed 's/^/  /'
  echo ""
  # Single host: only VMIDs that belong on this node (we don't have full map in legacy mode, so list all and skip missing)
  for entry in "${RPC_NODES[@]}"; do
    IFS=: read -r vmid host <<< "$entry"
    selected_vmid "$vmid" || continue
    [[ "$host" != "$PROXMOX_HOST" ]] && continue
    echo "--- VMID ${vmid} ---"
    ssh_pve "$host" "pct status ${vmid} 2>&1; pct config ${vmid} 2>&1 | egrep -i 'hostname:|rootfs:|memory:|swap:|cores:|net0:'" | sed 's/^/  /'
    echo ""
  done
else
  echo "=== Proxmox RPC Storage Audit (all RPC hosts) ==="
  echo ""

  # Collect unique hosts and show storage per host
  declare -A seen_host
  for entry in "${RPC_NODES[@]}"; do
    IFS=: read -r vmid host <<< "$entry"
    [[ -n "${seen_host[$host]:-}" ]] && continue
    seen_host[$host]=1
    NODE="$(ssh_pve "$host" "hostname" 2>/dev/null || echo "$host")"
    echo "--- Host $host (node: $NODE) ---"
    echo "Storages:"
    ssh_pve "$host" "pvesm status" 2>/dev/null | sed 's/^/  /'
    echo ""
  done

  echo "=== storage.cfg (from first host) ==="
  first_host="${PROXMOX_HOST_R630_01}"
  ssh_pve "$first_host" "awk '
    /^dir: /{s=\$2; t=\"dir\"; nodes=\"\"}
    /^lvmthin: /{s=\$2; t=\"lvmthin\"; nodes=\"\"}
    /^[[:space:]]*nodes /{nodes=\$2}
    /^[[:space:]]*nodes /{print t \":\" s \" nodes=\" nodes}
  ' /etc/pve/storage.cfg" 2>/dev/null | sed 's/^/  /'
  echo ""

  echo "=== RPC VMID -> rootfs storage mapping (per host) ==="
  for entry in "${RPC_NODES[@]}"; do
    IFS=: read -r vmid host <<< "$entry"
    selected_vmid "$vmid" || continue
    echo "--- VMID ${vmid} (host $host) ---"
    ssh_pve "$host" "pct status ${vmid} 2>&1; pct config ${vmid} 2>&1 | egrep -i 'hostname:|rootfs:|memory:|swap:|cores:|net0:'" | sed 's/^/  /'
    echo ""
  done
fi

cat <<'NOTE'
NOTE:
- If a VMID uses rootfs "local-lvm:*" on a node where storage.cfg restricts "local-lvm" to other nodes,
  the container may fail to start after a shutdown/reboot.
- Fix is to update /etc/pve/storage.cfg nodes=... for that storage to include the node hosting the VMID,
  or migrate the VMID to an allowed node/storage.
NOTE