#!/usr/bin/env bash
set -Euo pipefail

CONFIG_FILE="${XDG_CONFIG_HOME:-$HOME/.config}/hermes-stack-maintenance/config.env"
STATE_DIR="${XDG_STATE_HOME:-$HOME/.local/state}/hermes-stack-maintenance"
NOTIFIER="$HOME/.local/libexec/hermes-stack-notify"
mkdir -p "$STATE_DIR"
# shellcheck disable=SC1090
source "$CONFIG_FILE"

# Niemals den absichtlich gestoppten Gateway mitten im Update neu starten.
exec 9>"${XDG_RUNTIME_DIR:-/tmp}/hermes-stack-hermes-update.lock"
flock -n 9 || exit 0

state_file="$STATE_DIR/health-state"
problems=()

if ! systemctl --user is-active --quiet "$GATEWAY_SERVICE"; then
  problems+=("$GATEWAY_SERVICE war inaktiv; Neustart wurde versucht")
  systemctl --user restart "$GATEWAY_SERVICE" || true
fi

if ! hermes gateway status >/dev/null 2>&1; then
  problems+=("Hermes-Gateway antwortete nicht gesund; Neustart wurde versucht")
  systemctl --user restart "$GATEWAY_SERVICE" || true
fi

if [[ "${LLAMA_HEALTH_AUTORESTART:-0}" == 1 ]] &&
   ! systemctl --user is-active --quiet "$LLAMA_SERVICE"; then
  problems+=("$LLAMA_SERVICE war inaktiv; Neustart wurde versucht")
  systemctl --user restart "$LLAMA_SERVICE" || true
fi

if ! curl --fail --silent --max-time 10 "$LLAMA_URL/health" |
     jq -e '.status == "ok"' >/dev/null 2>&1; then
  problems+=("llama-server antwortete nicht gesund; kein HTTP-bedingter Neustart laufender Inferenz")
fi

if ((${#problems[@]} == 0)); then
  current="healthy"
  previous="$(cat "$state_file" 2>/dev/null || true)"
  if [[ "$previous" == unhealthy ]]; then
    "$NOTIFIER" "Hermes-Stack: Dienste sind wieder gesund." || true
  fi
else
  current="unhealthy"
  previous="$(cat "$state_file" 2>/dev/null || true)"
  if [[ "$previous" != unhealthy ]]; then
    message="$(IFS='; '; printf '%s' "${problems[*]}")"
    "$NOTIFIER" "Hermes-Stack: Gesundheitsprüfung fehlgeschlagen. $message. Details: journalctl --user -u $LLAMA_SERVICE -u $GATEWAY_SERVICE" || true
  fi
fi

printf '%s\n' "$current" >"$state_file"
[[ "$current" == healthy ]]