vps/runbooks/scripts/ha-maintenance.sh

#!/usr/bin/env bash
#
# ha-maintenance.sh — Home Assistant maintenance via the `ha` supervisor CLI.
#
# Target : hass.windy.lan (HAOS / Core), SSH user `hassio`.
# Access : non-interactive `sudo -n -i ha <cmd>`; the root login environment
#          carries the supervisor API token (interactive login loads
#          ~hassio/.zprofile which runs `exec sudo -i`). The whole remote
#          procedure runs in ONE `sudo -n -i bash -s` login so the SSH & Web
#          Terminal MOTD is printed only once, then stripped locally.
#          See hosts/hass.windy.lan.md and runbooks/home-assistant-maintenance.md.
#
# Usage:
#   ./ha-maintenance.sh                                read-only health snapshot
#   ./ha-maintenance.sh --check-config                 validate core configuration
#   ./ha-maintenance.sh --logs core|supervisor|host|apps:<slug> [N]
#   ./ha-maintenance.sh --update --yes                 refresh + update core(--backup)/supervisor/os
#   ./ha-maintenance.sh --restart-core [--safe-mode] --yes
#   ./ha-maintenance.sh --rebuild-core --yes           rebuild core image (after options change)
#   ./ha-maintenance.sh --rollback-os --yes            boot the previous OS slot (A/B rollback)
#   ./ha-maintenance.sh --backup [NAME] --yes          full backup (optionally named)
#   ./ha-maintenance.sh --restore <slug> --yes         restore a backup (DESTRUCTIVE)
#   ./ha-maintenance.sh --app start|stop|restart|update <slug> --yes
#
# Environment: HA_HOST (default hass.windy.lan), HA_SSH_USER (default hassio).
# Mutating actions always require --yes; this script never prompts.

set -uo pipefail

HOST="${HA_HOST:-hass.windy.lan}"
SSH_USER="${HA_SSH_USER:-hassio}"

MODE="health"
TARGET="core"
LINES=100
SAFE_MODE=0
CONFIRM=0
ARG1=""
ARG2=""

while [ $# -gt 0 ]; do
  case "$1" in
    --check-config) MODE="check-config" ;;
    --logs)
      MODE="logs"
      if [ $# -gt 1 ]; then TARGET="$2"; shift; fi
      if [ $# -gt 1 ] && [[ "$2" =~ ^[0-9]+$ ]]; then LINES="$2"; shift; fi
      ;;
    --update) MODE="update" ;;
    --restart-core) MODE="restart-core" ;;
    --safe-mode) SAFE_MODE=1 ;;
    --rebuild-core) MODE="rebuild-core" ;;
    --rollback-os) MODE="rollback-os" ;;
    --reboot) MODE="reboot" ;;
    --backup)
      MODE="backup"
      if [ $# -gt 1 ] && [[ "$2" != --* ]]; then ARG1="$2"; shift; fi
      ;;
    --restore)
      MODE="restore"
      if [ $# -gt 1 ] && [[ "$2" != --* ]]; then ARG1="$2"; shift
      else echo "ha-maintenance: --restore needs a backup slug" >&2; exit 2; fi
      ;;
    --app)
      MODE="app"
      if [ $# -gt 1 ] && [[ "$2" != --* ]]; then ARG1="$2"; shift; fi
      if [ $# -gt 1 ] && [[ "$2" != --* ]]; then ARG2="$2"; shift; fi
      ;;
    --yes) CONFIRM=1 ;;
    *)
      echo "ha-maintenance: unknown option '$1'" >&2
      echo "usage: $0 [--check-config|--logs TARGET [N]|--update|--restart-core [--safe-mode]|--rebuild-core|--rollback-os|--reboot|--backup [NAME]|--restore SLUG|--app ACTION SLUG] [--yes]" >&2
      exit 2
      ;;
  esac
  shift
done

if [ "$MODE" = "app" ]; then
  case "$ARG1" in
    start|stop|restart|update) ;;
    *) echo "ha-maintenance: --app action must be start|stop|restart|update (got '$ARG1')" >&2; exit 2 ;;
  esac
  if [ -z "$ARG2" ]; then
    echo "ha-maintenance: --app needs an app slug, e.g. --app restart core_mosquitto" >&2
    exit 2
  fi
fi

case "$MODE" in
  update|restart-core|rebuild-core|rollback-os|reboot|backup|restore|app)
    if [ "$CONFIRM" -ne 1 ]; then
      echo "ha-maintenance: refusing '$MODE' without --yes (mutating action)" >&2
      exit 1
    fi
    ;;
esac

REMOTE_SCRIPT=$(cat <<'EOF'
set -u
ha() { command ha --no-progress "$@"; }

case "$1" in
  health)
    printf '\n=== CORE ===\n'
    ha core info
    printf '\n=== SUPERVISOR ===\n'
    ha supervisor info | grep -E '^(healthy|supported|version|version_latest|update_available|auto_update|timezone|logging):'
    printf '\n=== ADDON STATES ===\n'
    ha supervisor info | grep -E '^  (name|state):'
    printf '\n=== OS ===\n'
    ha os info
    printf '\n=== HOST ===\n'
    ha host info
    printf '\n=== NETWORK ===\n'
    ha network info | grep -E '^(address|ip_address|supervisor_internet|vlan|wifi):'
    printf '\n=== AVAILABLE UPDATES ===\n'
    ha available-updates
    printf '\n=== JOBS ===\n'
    ha jobs info
    printf '\n=== RESOLUTION ===\n'
    ha resolution info
    printf '\n=== CORE STATS ===\n'
    ha core stats
    ;;
  check-config)
    ha core check
    ;;
  logs)
    case "$2" in
      core)       ha core logs -n "$3" ;;
      supervisor) ha supervisor logs -n "$3" ;;
      host)       ha host logs -n "$3" ;;
      apps:*)     ha apps logs -n "$3" "${2#apps:}" ;;
      *)          echo "unknown log target: $2" >&2; exit 2 ;;
    esac
    ;;
  update)
    ha refresh-updates
    printf '\n=== UPDATE CORE (with backup) ===\n';  ha core update --backup
    printf '\n=== UPDATE SUPERVISOR ===\n';          ha supervisor update
    printf '\n=== UPDATE OS ===\n';                  ha os update
    printf '\n=== PENDING AFTER ===\n';              ha available-updates
    ;;
  restart-core)
    if [ "$4" = "1" ]; then ha core restart --safe-mode; else ha core restart; fi
    ;;
  rebuild-core)
    ha core rebuild
    ;;
  rollback-os)
    ha os boot-slot other
    ;;
  reboot)
    ha host reboot
    ;;
  backup)
    if [ -n "${2:-}" ]; then ha backups new --name "$2"; else ha backups new; fi
    ;;
  restore)
    echo "WARNING: restoring backup $2"
    ha backups restore "$2"
    ;;
  app)
    ha apps "$2" "$3"
    ;;
esac
EOF
)

if [ "$MODE" = "logs" ]; then
  ARG1="$TARGET"
  ARG2="$LINES"
fi

# One remote login shell (MOTD printed once), then drop the MOTD locally.
printf '%s\n' "$REMOTE_SCRIPT" \
  | ssh -o BatchMode=yes -o ConnectTimeout=10 "${SSH_USER}@${HOST}" \
      'sudo -n -i bash -s' "$MODE" "$ARG1" "$ARG2" "$SAFE_MODE" 2>/dev/null \
  | awk 'BEGIN{on=0} /^System is ready/{on=1; next} on'