kis_bot/kis_trader/utils/ops_alert.py

"""
운영 치명 알림 — 예외 삼킴/로그만 하던 인프라 장애를 MM(+선택 TG)로 올린다.

- 체결·기동 알림과 분리. 채널 기본 = KIS_SYSTEM_MM_CHANNEL.
- 코드별 쿨다운 + 전역 최소간격으로 스팸 방지.
- 장중 전용 코드는 session gate (주말·장외 오탐 완화).
"""
from __future__ import annotations

import threading
import time
from datetime import datetime
from pathlib import Path
from typing import Any, Dict, Optional

from .env import get_env_bool, get_env_from_db, get_env_int
from .logger import atomic_load_json, atomic_save_json, get_logger, msg_mm, msg_tg

logger = get_logger("kis_trader.ops_alert")

_ROOT = Path(__file__).resolve().parents[2]
_STATE_PATH = _ROOT / "logs" / "ops_alert_state.json"
_LOCK = threading.Lock()

# 장중 세션에서만 의미 있는 코드 (토큰·PANIC·VI 는 상시)
_SESSION_CODES = frozenset({
    "ws_kis_down",
    "ws_kiwoom_down",
    "ws_ls_down",
    "ws_tick_silence",
    "universe_zero",
    "universe_wipe",
    "history_stale",
    "kwcond_off",
    "order_buy_reject",
    "order_sell_reject",
    "rate_limit",
})


def _channel() -> str:
    ch = (get_env_from_db("OPS_ALERT_MM_CHANNEL", "") or "").strip()
    if not ch:
        ch = (get_env_from_db("KIS_SYSTEM_MM_CHANNEL", "default") or "default").strip()
    return ch or "default"


def _cooldown_sec(code: str) -> int:
    per = get_env_int(f"OPS_ALERT_COOLDOWN_{code.upper()}_SEC", 0)
    if per > 0:
        return max(30, int(per))
    return max(30, get_env_int("OPS_ALERT_COOLDOWN_SEC", 300))


def _in_kr_session(now: Optional[datetime] = None) -> bool:
    now = now or datetime.now()
    if now.weekday() >= 5:
        return False
    hm = now.hour * 100 + now.minute
    start = int(get_env_int("OPS_ALERT_SESSION_START_HM", 900) or 900)
    end = int(get_env_int("OPS_ALERT_SESSION_END_HM", 1535) or 1535)
    return start <= hm <= end


def _load_state() -> Dict[str, Any]:
    st = atomic_load_json(_STATE_PATH, default={})
    return st if isinstance(st, dict) else {}


def _save_state(st: Dict[str, Any]) -> None:
    try:
        _STATE_PATH.parent.mkdir(parents=True, exist_ok=True)
        atomic_save_json(_STATE_PATH, st)
    except Exception as e:
        logger.debug("ops_alert state save: %s", e)


def ops_alert(
    code: str,
    title: str,
    *,
    detail: str = "",
    level: str = "critical",
    force: bool = False,
    session_only: Optional[bool] = None,
) -> bool:
    """운영 알림 1건. True=발송됨.

    ``level``: critical | warn
    ``session_only``: None 이면 코드 기본(장중 전용 집합), True/False 강제.
    """
    if not get_env_bool("OPS_ALERT_ENABLED", True):
        return False
    code = str(code or "").strip().lower() or "unknown"
    title = str(title or "").strip() or code
    level = str(level or "critical").strip().lower()
    if level not in ("critical", "warn"):
        level = "critical"

    need_session = (
        bool(session_only) if session_only is not None
        else (code in _SESSION_CODES)
    )
    if need_session and not _in_kr_session():
        return False

    now = time.time()
    with _LOCK:
        st = _load_state()
        by_code = st.setdefault("by_code", {})
        if not isinstance(by_code, dict):
            by_code = {}
            st["by_code"] = by_code
        last = float(by_code.get(code) or 0.0)
        cd = _cooldown_sec(code)
        if not force and (now - last) < cd:
            return False
        global_gap = float(get_env_int("OPS_ALERT_GLOBAL_MIN_GAP_SEC", 20) or 20)
        last_any = float(st.get("last_any_ts") or 0.0)
        if not force and (now - last_any) < global_gap:
            return False

        icon = "🚨" if level == "critical" else "⚠️"
        lines = [
            f"{icon} **[운영알림/{level.upper()}] {title}**",
            f"- 코드: `{code}`",
            f"- 시각: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}",
        ]
        if detail:
            clipped = str(detail).strip()
            if len(clipped) > 1200:
                clipped = clipped[:1200] + "…"
            lines.append(f"- 상세: {clipped}")
        body = "\n".join(lines)

        ok = False
        try:
            ok = bool(msg_mm(body, channel_alias=_channel(), jitter=False))
        except Exception as e:
            logger.debug("ops_alert MM 실패: %s", e)
        if level == "critical" and get_env_bool("OPS_ALERT_TG_ON_CRITICAL", True):
            try:
                msg_tg(body, jitter=False)
            except Exception as e:
                logger.debug("ops_alert TG 실패: %s", e)

        by_code[code] = now
        st["last_any_ts"] = now
        st["last_code"] = code
        _save_state(st)
        logger.warning("[ops_alert] sent code=%s ok_mm=%s title=%s", code, ok, title)
        return ok


def note_counter(key: str, *, reset: bool = False) -> int:
    """연속 실패 카운터 (주문거부·유량 등). reset=True 이면 0."""
    with _LOCK:
        st = _load_state()
        counters = st.setdefault("counters", {})
        if not isinstance(counters, dict):
            counters = {}
            st["counters"] = counters
        if reset:
            counters[key] = 0
            _save_state(st)
            return 0
        n = int(counters.get(key) or 0) + 1
        counters[key] = n
        _save_state(st)
        return n