process_manager.py

"""
process_manager.py
──────────────────
Thread-safe singleton that manages the bot.py subprocess.

Why a separate module?
  Streamlit reruns app.py on every interaction (button click, page refresh).
  A naive approach loses the subprocess reference on each rerun.
  This module is imported *once* by the Python interpreter, so module-level
  globals survive across Streamlit reruns — giving us a true singleton.

Public API
  start_bot()   → (success: bool, message: str)
  stop_bot()    → (success: bool, message: str)
  is_running()  → bool
  get_stats()   → dict | None
"""

import os
import sys
import signal
import subprocess
import threading
import time
from datetime import datetime
from pathlib import Path

import psutil

# ── Paths ─────────────────────────────────────────────────────────────────────
LOG_FILE = Path("/tmp/logs/bot.log")
PID_FILE = Path("/tmp/logs/bot.pid")
BOT_SCRIPT = str(Path(__file__).parent.parent / "bot.py")

# ── Module-level state (survives Streamlit reruns) ────────────────────────────
_process: subprocess.Popen | None = None
_lock = threading.Lock()

# ── Internal helpers ──────────────────────────────────────────────────────────

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


def _read_pid() -> int | None:
    try:
        return int(PID_FILE.read_text().strip())
    except (FileNotFoundError, ValueError):
        return None


def _clear_pid() -> None:
    PID_FILE.unlink(missing_ok=True)


def _proc_alive(pid: int) -> bool:
    """Return True if OS process `pid` exists and is not a zombie."""
    try:
        p = psutil.Process(pid)
        return p.is_running() and p.status() != psutil.STATUS_ZOMBIE
    except (psutil.NoSuchProcess, psutil.AccessDenied):
        return False


def _append_log(text: str) -> None:
    LOG_FILE.parent.mkdir(parents=True, exist_ok=True)
    with open(LOG_FILE, "a", encoding="utf-8") as fh:
        fh.write(text)


# ── Public API ────────────────────────────────────────────────────────────────

def is_running() -> bool:
    """Check whether the trading bot is alive."""
    global _process

    # 1. In-memory reference (fastest path)
    if _process is not None and _process.poll() is None:
        return True

    # 2. PID file — handles page refreshes / new browser sessions
    pid = _read_pid()
    if pid and _proc_alive(pid):
        return True

    # Stale PID file cleanup
    if pid and not _proc_alive(pid):
        _clear_pid()

    return False


def start_bot() -> tuple[bool, str]:
    """Launch bot.py as a detached subprocess. Idempotent."""
    global _process

    with _lock:
        if is_running():
            return False, "Bot is already running."

        if not Path(BOT_SCRIPT).exists():
            return False, f"bot.py not found at: {BOT_SCRIPT}"

        try:
            LOG_FILE.parent.mkdir(parents=True, exist_ok=True)

            # Separator so the log is easy to scan
            _append_log(
                f"\n{'═' * 60}\n"
                f"[{datetime.now().isoformat(timespec='seconds')}] ▶  BOT STARTED\n"
                f"{'═' * 60}\n"
            )

            log_fh = open(LOG_FILE, "a", encoding="utf-8", buffering=1)

            env = os.environ.copy()

            _process = subprocess.Popen(
                [sys.executable, "-u", BOT_SCRIPT],  # -u = unbuffered stdout
                stdout=log_fh,
                stderr=subprocess.STDOUT,
                env=env,
                # start_new_session prevents SIGINT propagation from Streamlit
                start_new_session=True,
            )

            _write_pid(_process.pid)
            return True, f"Bot started — PID {_process.pid}"

        except Exception as exc:  # noqa: BLE001
            return False, f"Failed to start bot: {exc}"


def stop_bot() -> tuple[bool, str]:
    """Gracefully terminate the bot. Falls back to SIGKILL after 10 s."""
    global _process

    with _lock:
        if not is_running():
            _clear_pid()
            return False, "Bot is not running."

        pid = _read_pid()
        stopped = False

        # -- Try in-memory handle first --
        if _process is not None:
            try:
                _process.terminate()
                _process.wait(timeout=10)
                stopped = True
            except subprocess.TimeoutExpired:
                _process.kill()
                _process.wait()
                stopped = True
            except Exception:
                pass
            finally:
                _process = None

        # -- Fall back to PID file (cross-session stop) --
        if not stopped and pid:
            try:
                p = psutil.Process(pid)
                p.terminate()
                p.wait(timeout=10)
                stopped = True
            except psutil.NoSuchProcess:
                stopped = True          # Already gone — that's fine
            except psutil.TimeoutExpired:
                p.kill()
                stopped = True
            except Exception:
                pass

        _clear_pid()

        _append_log(
            f"\n{'═' * 60}\n"
            f"[{datetime.now().isoformat(timespec='seconds')}] ⏹  BOT STOPPED\n"
            f"{'═' * 60}\n"
        )

        return stopped, "Bot stopped." if stopped else "Stop attempted (process may have already exited)."


def get_stats() -> dict | None:
    """Return live resource metrics for the bot process, or None."""
    pid = _read_pid()
    if not pid:
        return None
    try:
        p = psutil.Process(pid)
        with p.oneshot():
            mem = p.memory_info()
            created = datetime.fromtimestamp(p.create_time())
            uptime = datetime.now() - created
            hours, remainder = divmod(int(uptime.total_seconds()), 3600)
            minutes, seconds = divmod(remainder, 60)
            return {
                "pid": pid,
                "status": p.status(),
                "cpu_pct": p.cpu_percent(interval=0.2),
                "rss_mb": mem.rss / (1024 ** 2),
                "vms_mb": mem.vms / (1024 ** 2),
                "uptime_str": f"{hours:02d}h {minutes:02d}m {seconds:02d}s",
                "started_at": created.strftime("%Y-%m-%d %H:%M:%S"),
            }
    except (psutil.NoSuchProcess, psutil.AccessDenied):
        _clear_pid()
        return None


def get_logs(n_lines: int = 200) -> str:
    """Return the last *n_lines* from the log file."""
    if not LOG_FILE.exists():
        return "(No log output yet — start the bot to see activity.)"
    try:
        lines = LOG_FILE.read_text(encoding="utf-8", errors="replace").splitlines()
        return "\n".join(lines[-n_lines:]) if lines else "(Log file is empty.)"
    except OSError as exc:
        return f"(Error reading log: {exc})"


def clear_logs() -> None:
    """Truncate the log file."""
    if LOG_FILE.exists():
        LOG_FILE.write_text("", encoding="utf-8")