Stock-trading-programming/scripts/_watchdog.py

import asyncio
import os
import subprocess
import sys
from datetime import datetime
from pathlib import Path


PROJECT = Path(__file__).resolve().parents[1]
PID_FILE = PROJECT / "logs" / "bot.pid"

os.chdir(PROJECT)
sys.path.insert(0, str(PROJECT))

from app.main import load_env

load_env()


def _is_process_alive(pid: int) -> bool:
    result = subprocess.run(
        ["tasklist", "/FI", f"PID eq {pid}", "/NH"],
        capture_output=True,
        text=True,
    )
    return str(pid) in result.stdout


def _find_bot_pids() -> list[int]:
    command = (
        "Get-CimInstance Win32_Process | "
        "Where-Object { $_.Name -like 'python*' -and "
        "($_.CommandLine -like '*app/main.py*' "
        "-or $_.CommandLine -like '*app\\\\main.py*') } | "
        "Select-Object -ExpandProperty ProcessId"
    )
    result = subprocess.run(
        ["powershell", "-NoProfile", "-Command", command],
        capture_output=True,
        text=True,
    )
    pids: list[int] = []
    for line in result.stdout.splitlines():
        line = line.strip()
        if line.isdigit():
            pids.append(int(line))
    return pids


def _get_pid() -> int | None:
    try:
        return int(PID_FILE.read_text(encoding="utf-8").strip())
    except Exception:
        return None


def _write_pid(pid: int) -> None:
    PID_FILE.parent.mkdir(exist_ok=True)
    PID_FILE.write_text(str(pid), encoding="utf-8")


def _restart_bot() -> int:
    env = os.environ.copy()
    env["PYTHONUNBUFFERED"] = "1"
    creationflags = 0
    if hasattr(subprocess, "DETACHED_PROCESS"):
        creationflags = subprocess.DETACHED_PROCESS | subprocess.CREATE_NEW_PROCESS_GROUP

    log_path = PROJECT / "logs" / "bot_stderr.log"
    with open(log_path, "a", encoding="utf-8") as log:
        proc = subprocess.Popen(
            [sys.executable, "-u", "app/main.py"],
            cwd=PROJECT,
            creationflags=creationflags,
            stdout=log,
            stderr=subprocess.STDOUT,
            close_fds=True,
            env=env,
        )
    _write_pid(proc.pid)
    return proc.pid


async def main():
    now = datetime.now()
    now_str = now.strftime("%H:%M")

    if not ("09:00" <= now_str <= "15:10"):
        print(f"[{now_str}] outside trading window - watchdog skipped")
        return

    from app.monitor.notifier import send

    pid = _get_pid()
    if pid is not None and _is_process_alive(pid):
        print(f"[{now_str}] bot running PID={pid}")
        return

    live_pids = _find_bot_pids()
    if live_pids:
        recovered_pid = live_pids[0]
        _write_pid(recovered_pid)
        msg = f"[복구] bot.pid corrected to running bot PID={recovered_pid} ({now_str})"
        print(msg)
        await send(msg)
        return

    msg = f"[긴급] bot process not found (pid={pid}) - restarting"
    print(msg)
    await send(msg)

    new_pid = _restart_bot()
    await send(f"[복구] bot restarted PID={new_pid} ({now_str})")
    print(f"bot restarted PID={new_pid}")


if __name__ == "__main__":
    asyncio.run(main())