AWatch-rus/ops/detmir-ai/detmir-auto

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

state_dir="${DETMIR_AI_STATE_DIR:-/var/lib/detmir-ai}"
lock_dir="${DETMIR_AI_RUN_DIR:-${XDG_RUNTIME_DIR:-/tmp}}"
auto_heal="${DETMIR_AUTO_HEAL:-1}"
retain_days="${DETMIR_AI_RETAIN_DAYS:-14}"

mkdir -p "$state_dir"/runs "$state_dir"/reports "$state_dir"/logs

lock_file="$lock_dir/detmir-auto.lock"
exec 9>"$lock_file"
if ! flock -n 9; then
  echo "detmir-auto: another run is active"
  exit 0
fi

stamp="$(date -u +%Y%m%d-%H%M%S)"
run_dir_full="$state_dir/runs/$stamp"
mkdir -p "$run_dir_full"

check_file="$run_dir_full/detmir-check.json"
dlp_file="$run_dir_full/detmir-dlp.json"
heal_log="$run_dir_full/heal.log"
bundle_file="$run_dir_full/bundle.txt"
report_file="$state_dir/reports/detmir-report-$stamp.md"
state_file="$state_dir/state-$stamp.json"

run_check() {
  local rc=0
  detmir-check --json >"$check_file" || rc=$?
  echo "$rc" >"$run_dir_full/check.rc"
}

run_dlp() {
  local rc=0
  detmir-dlp >"$dlp_file" || rc=$?
  echo "$rc" >"$run_dir_full/dlp.rc"
}

summarize() {
  python3 - "$check_file" "$dlp_file" "$run_dir_full/check.rc" "$run_dir_full/dlp.rc" <<'PY'
import json
import sys
from pathlib import Path

check_path, dlp_path, check_rc_path, dlp_rc_path = map(Path, sys.argv[1:])
check_rc = int(check_rc_path.read_text().strip())
dlp_rc = int(dlp_rc_path.read_text().strip())
summary = {
    "check_rc": check_rc,
    "dlp_rc": dlp_rc,
    "check_ok": False,
    "dlp_ok": False,
    "severity": "FAIL" if check_rc or dlp_rc else "OK",
    "needs_heal": bool(check_rc or dlp_rc),
    "reasons": [],
}
try:
    check = json.loads(check_path.read_text())
    summary["check_ok"] = bool(check.get("ok"))
    cs = check.get("summary") or {}
    summary["detmir_summary"] = cs
    if cs.get("bucket_dead", 0) or cs.get("bucket_stale", 0) or cs.get("service_failures", 0):
        summary["reasons"].append("detmir-check has stale/dead bucket or required service failure")
except Exception as exc:
    summary["reasons"].append(f"detmir-check parse failed: {exc}")

try:
    dlp = json.loads(dlp_path.read_text())
    summary["dlp_ok"] = bool(dlp.get("ok"))
    summary["dlp_counts"] = dlp.get("counts")
    counts = dlp.get("counts") or {}
    if counts.get("fail", 0) or counts.get("warn", 0):
        summary["reasons"].append("dlp-health-check has warn/fail")
except Exception as exc:
    summary["reasons"].append(f"detmir-dlp parse failed: {exc}")

if summary["check_ok"] and summary["dlp_ok"]:
    summary["severity"] = "OK"
    summary["needs_heal"] = False
elif not summary["reasons"]:
    summary["severity"] = "WARN"
else:
    summary["severity"] = "FAIL"

print(json.dumps(summary, ensure_ascii=False, indent=2))
PY
}

run_check
run_dlp
summary_before="$(summarize)"
printf '%s\n' "$summary_before" >"$run_dir_full/summary-before.json"

if [ "$auto_heal" = "1" ] && python3 -c 'import json,sys; print("yes" if json.load(sys.stdin).get("needs_heal") else "no")' <<<"$summary_before" | grep -qx yes; then
  {
    echo "detmir-heal-safe started at $(date -u --iso-8601=seconds)"
    detmir-heal-safe
    echo "detmir-heal-safe finished at $(date -u --iso-8601=seconds)"
  } >"$heal_log" 2>&1 || true
  sleep 10
  run_check
  run_dlp
else
  echo "auto-heal skipped" >"$heal_log"
fi

summary_after="$(summarize)"
printf '%s\n' "$summary_after" >"$state_file"

{
  echo "Ты операторский AI-помощник DetMir. По фактам ниже дай короткий русский отчет."
  echo "Структура ответа:"
  echo "1. Состояние: OK/WARN/FAIL"
  echo "2. Что важно"
  echo "3. Что уже сделал автомат"
  echo "4. Что сделать человеку, если нужно"
  echo
  echo "Правила:"
  echo "- Не предлагай рестарты, если факты чистые."
  echo "- Отличай event-driven bucket от dead/stale."
  echo "- DLP sendFailures важны только при новом sendFailuresDelta или warn/fail."
  echo "- Auto-heal умеет только серверные systemd-сервисы AW/DLP; Windows/RDP не трогает."
  echo
  echo "=== summary-before ==="
  cat "$run_dir_full/summary-before.json"
  echo
  echo "=== summary-after ==="
  cat "$state_file"
  echo
  echo "=== heal-log ==="
  sed -n '1,300p' "$heal_log"
  echo
  echo "=== detmir-check ==="
  sed -n '1,1600p' "$check_file"
  echo
  echo "=== detmir-dlp ==="
  sed -n '1,1600p' "$dlp_file"
} >"$bundle_file"

{
  echo "# DetMir Autonomous Report"
  echo
  echo "- generated_at_utc: $(date -u --iso-8601=seconds)"
  echo "- run_dir: $run_dir_full"
  echo
  polli-chat --model text.daily --max-tokens 900 <"$bundle_file" || {
    echo "Pollinations report failed; raw summary follows."
    cat "$state_file"
  }
} >"$report_file"

ln -sfn "$run_dir_full" "$state_dir/latest-run"
ln -sfn "$report_file" "$state_dir/latest-report.md"
ln -sfn "$state_file" "$state_dir/latest-state.json"

find "$state_dir/runs" -mindepth 1 -maxdepth 1 -type d -mtime +"$retain_days" -exec rm -rf {} +
find "$state_dir/reports" -type f -name 'detmir-report-*.md' -mtime +"$retain_days" -delete
find "$state_dir" -maxdepth 1 -type f -name 'state-*.json' -mtime +"$retain_days" -delete

cat "$report_file"

python3 - "$state_file" <<'PY'
import json
import sys
state = json.load(open(sys.argv[1]))
raise SystemExit(0 if state.get("severity") == "OK" else 2)
PY