sanguo_vnpy_v2/sanguo_trader/shadow/supervisor.py

"""影子柜台账户主管（P1-d）。

轮询 paper 库:发现 ``mode='shadow' & status='running'`` 的组合账户 → 为每个账户
拉起一个 ``python -m sanguo_trader.shadow --account N`` 子进程(独立虚拟账户);
账户停止/删除 → 终止对应子进程;子进程崩溃 → 重启(告警日志)。

用法(VPS,常驻):
    set SANGUO_SHADOW_DB=C:\\sanguo_vnpy_v2\\data\\backtest_results.db
    python -m sanguo_trader.shadow --auto

与 sanguo_live.supervisor 同思路:主管只管进程生命周期,不碰撮合。
"""
from __future__ import annotations

import json
import logging
import os
import sqlite3
import subprocess
import sys
import time
from typing import Any, Dict, Optional

logger = logging.getLogger(__name__)

POLL_SEC = 60


def load_shadow_accounts(db_path: str) -> list[dict[str, Any]]:
    """读所有该拉起影子柜台的账户(mode=shadow & running & portfolio)。"""
    with sqlite3.connect(db_path) as conn:
        conn.row_factory = sqlite3.Row
        rows = conn.execute(
            "SELECT * FROM paper_accounts "
            "WHERE mode='shadow' AND status='running' AND strategy_type='portfolio'"
        ).fetchall()
    return [dict(r) for r in rows]


def _num(v: Any) -> str:
    """数字转 env 字符串,整数不带小数点(500000 而非 500000.0)。"""
    f = float(v)
    return str(int(f)) if f.is_integer() else repr(f)


def account_env(acc: dict[str, Any], db_path: str) -> Dict[str, str]:
    """paper 账户行 → 影子柜台子进程 env(复用 SANGUO_LIVE_* 契约)。"""
    strategies = json.loads(acc.get("strategies") or "[]")
    name = (strategies[0].get("name") if strategies else None) or "all_weather"
    params = (strategies[0].get("params") if strategies else {}) or {}
    env = dict(os.environ)
    env.update({
        "SANGUO_LIVE_STRATEGY": name,
        "SANGUO_LIVE_MAX_POOL": _num(params.get("max_pool", 0)),
        "SANGUO_LIVE_BENCHMARK": str(params.get("benchmark", "000300.XSHG")),
        "SANGUO_LIVE_CASH": _num(acc.get("initial_capital") or 1_000_000),
        "SANGUO_SHADOW_DB": db_path,
        "SANGUO_SHADOW_ACCOUNT_ID": str(acc["id"]),
        "SANGUO_SHADOW_COMMISSION": repr(float(acc.get("rate") or 0.0003)),
        "SANGUO_SHADOW_STAMP": repr(float(acc.get("stamp_duty_rate") or 0.001)),
        "SANGUO_SHADOW_MIN_COMM": _num(acc.get("min_commission") or 5),
        "SANGUO_SHADOW_SLIPPAGE": repr(float(acc.get("slippage") or 0.001)),
    })
    return env


def spawn_child(acc: dict[str, Any], db_path: str) -> subprocess.Popen:
    env = account_env(acc, db_path)
    argv = [sys.executable, "-X", "utf8", "-m", "sanguo_trader.shadow",
            "--account", str(acc["id"])]
    logger.info("[shadow-supervisor] 拉起账户 #%s(%s) 影子柜台",
                acc["id"], env["SANGUO_LIVE_STRATEGY"])
    # #88 可观测性:子进程 stdout/stderr 落 logs/shadow_{aid}.log(>5MB 轮转截断)。
    # 原实现继承 schtask 控制台(=黑洞),子进程秒崩时零线索——2026-08-16
    # VPS 实况 41/42 拉不起,连一行报错都没有,只能瞎猜。
    from pathlib import Path as _P
    log_dir = _P(__file__).resolve().parents[2] / "logs"
    log_dir.mkdir(parents=True, exist_ok=True)
    log_path = log_dir / f"shadow_{acc['id']}.log"
    try:
        if log_path.exists() and log_path.stat().st_size > 5 * 1024 * 1024:
            log_path.write_text("", encoding="utf-8")  # 超限重置,避免崩溃循环刷爆盘
        fh = open(log_path, "ab")
        fh.write(f"\n==== spawn {time.strftime('%Y-%m-%d %H:%M:%S')} ====\n".encode())
        return subprocess.Popen(argv, env=env, stdout=fh, stderr=subprocess.STDOUT)
    except Exception:
        logger.warning("[shadow-supervisor] 日志重定向失败,退回继承控制台", exc_info=True)
        return subprocess.Popen(argv, env=env)


def _maybe_daily_reconcile(db_path: str, done_dates: set,
                           now: Optional["datetime.datetime"] = None) -> None:
    """日终对账(设计 §8.2):收盘后(≥15:10)每日一次全配对跑报表落库。

    API/CLI 也可随时现算;这里只是自动化兜底,done_dates 防当日重复。
    """
    import datetime as _dt

    from .reconcile_report import (
        build_identity_report, build_reconcile_report, find_dual_track_pairs,
        save_identity_report, save_reconcile_report,
    )

    now = now or _dt.datetime.now()
    if now.hour < 15 or (now.hour == 15 and now.minute < 10):
        return
    today = now.strftime("%Y-%m-%d")
    if today in done_dates:
        return
    try:
        # B5 恒等式先行:全账户 = Σ实例账本 + 未归因;再逐对 live↔shadow 行为对比
        identity = build_identity_report(db_path, today)
        save_identity_report(db_path, identity)
        for row in identity["rows"]:
            logger.info("[shadow-supervisor] 恒等式 %s %s: 未归因=%.0f → %s",
                        today, row["account"],
                        row["unattributed_mv"] or 0, row["status"])
        for pair in find_dual_track_pairs(db_path):
            report = build_reconcile_report(
                db_path, pair["live_account_id"], pair["shadow_account_id"], today)
            save_reconcile_report(db_path, report)
            logger.info("[shadow-supervisor] 日终对账 %s live#%s vs shadow#%s: %s",
                        today, pair["live_account_id"], pair["shadow_account_id"],
                        "PASS" if report["passed"] else "FAIL")
        done_dates.add(today)
    except Exception as exc:  # noqa: BLE001 - 对账失败不退出主管,下一轮重试
        logger.warning("[shadow-supervisor] 日终对账失败(下轮重试): %s", exc)


def run_auto_supervisor(db_path: Optional[str] = None, poll_sec: float = POLL_SEC) -> None:
    """常驻主循环:同步账户 ↔ 子进程。"""
    db_path = db_path or os.environ.get("SANGUO_SHADOW_DB") \
        or r"C:\sanguo_vnpy_v2\data\backtest_results.db"
    logger.info("[shadow-supervisor] 启动 db=%s", db_path)
    children: Dict[int, subprocess.Popen] = {}
    reconcile_done: set = set()

    while True:
        try:
            accounts = load_shadow_accounts(db_path)
        except Exception as exc:  # noqa: BLE001 - db 抖动不退出主管
            logger.warning("[shadow-supervisor] 读账户失败: %s", exc)
            accounts = []
        want = {a["id"]: a for a in accounts}

        # 1) 终止不再需要的
        for aid in list(children):
            if aid not in want:
                child = children.pop(aid)
                if child.poll() is None:
                    child.terminate()
                    logger.info("[shadow-supervisor] 账户 #%s 停止,子进程已终止", aid)

        # 2) 重启崩溃的 / 拉起新的
        for aid, acc in want.items():
            child = children.get(aid)
            if child is not None and child.poll() is None:
                continue
            if child is not None:
                logger.warning("[shadow-supervisor] 账户 #%s 子进程退出(rc=%s),重启",
                               aid, child.returncode)
            try:
                children[aid] = spawn_child(acc, db_path)
            except Exception as exc:  # noqa: BLE001
                logger.error("[shadow-supervisor] 账户 #%s 拉起失败: %s", aid, exc)

        # 3) 日终双轨对账(收盘后每日一次)
        _maybe_daily_reconcile(db_path, reconcile_done)

        time.sleep(poll_sec)