vigilar/vigilar/main.py

"""Process supervisor — starts and monitors all Vigilar subsystems."""

import logging
import signal
import sys
import time
from multiprocessing import Process
from pathlib import Path
from typing import Any

from vigilar.config import VigilarConfig
from vigilar.storage.db import get_db_path, init_db

log = logging.getLogger(__name__)


class SubsystemProcess:
    """Wrapper for a subsystem running in a child process."""

    def __init__(self, name: str, target: Any, args: tuple = ()):
        self.name = name
        self.target = target
        self.args = args
        self.process: Process | None = None
        self.restart_count = 0
        self.max_restarts = 10
        self.last_restart: float = 0

    def start(self) -> None:
        self.process = Process(target=self.target, args=self.args, name=self.name, daemon=True)
        self.process.start()
        log.info("Started %s (pid=%s)", self.name, self.process.pid)

    def is_alive(self) -> bool:
        return self.process is not None and self.process.is_alive()

    def stop(self, timeout: float = 5.0) -> None:
        if self.process and self.process.is_alive():
            log.info("Stopping %s (pid=%s)", self.name, self.process.pid)
            self.process.terminate()
            self.process.join(timeout=timeout)
            if self.process.is_alive():
                log.warning("Force-killing %s", self.name)
                self.process.kill()
                self.process.join(timeout=2)

    def maybe_restart(self) -> bool:
        """Restart if crashed, with backoff. Returns True if restarted."""
        if self.is_alive():
            return False
        if self.restart_count >= self.max_restarts:
            log.error("%s exceeded max restarts (%d), giving up", self.name, self.max_restarts)
            return False

        backoff = min(2 ** self.restart_count, 60)
        elapsed = time.monotonic() - self.last_restart
        if elapsed < backoff:
            return False

        self.restart_count += 1
        self.last_restart = time.monotonic()
        log.warning("Restarting %s (attempt %d)", self.name, self.restart_count)
        self.start()
        return True


def _run_web(cfg: VigilarConfig) -> None:
    """Run the Flask web server in a subprocess."""
    from vigilar.web.app import create_app

    app = create_app(cfg)
    app.run(host=cfg.web.host, port=cfg.web.port, debug=False, use_reloader=False)


def _run_event_processor(cfg: VigilarConfig) -> None:
    """Run the event processor in a subprocess."""
    from vigilar.events.processor import EventProcessor

    processor = EventProcessor(cfg)
    processor.run()


def _run_ups_monitor(cfg: VigilarConfig) -> None:
    """Run the UPS monitor in a subprocess."""
    from vigilar.ups.monitor import UPSMonitor

    monitor = UPSMonitor(cfg)
    monitor.run()


def run_supervisor(cfg: VigilarConfig) -> None:
    """Main supervisor loop — starts all subsystems, monitors, restarts on crash."""
    # Initialize database
    db_path = get_db_path(cfg.system.data_dir)
    init_db(db_path)

    # Ensure directories exist
    Path(cfg.system.data_dir).mkdir(parents=True, exist_ok=True)
    Path(cfg.system.recordings_dir).mkdir(parents=True, exist_ok=True)
    Path(cfg.system.hls_dir).mkdir(parents=True, exist_ok=True)

    # Build subsystem list
    subsystems: list[SubsystemProcess] = []

    # Web server
    subsystems.append(SubsystemProcess("web", _run_web, (cfg,)))

    # Camera manager (manages its own child processes internally)
    from vigilar.camera.manager import CameraManager
    camera_mgr = CameraManager(cfg) if cfg.cameras else None

    # Event processor
    subsystems.append(SubsystemProcess("event-processor", _run_event_processor, (cfg,)))

    # Phase 7 — Sensor bridge
    if cfg.sensors:
        from vigilar.sensors.bridge import run_sensor_bridge
        subsystems.append(SubsystemProcess("sensor-bridge", run_sensor_bridge, (cfg,)))

    # Phase 8 — UPS monitor
    if cfg.ups.enabled:
        subsystems.append(SubsystemProcess("ups-monitor", _run_ups_monitor, (cfg,)))

    # Presence monitor
    if cfg.presence.enabled and cfg.presence.members:
        from vigilar.presence.monitor import run_presence_monitor
        subsystems.append(SubsystemProcess("presence-monitor", run_presence_monitor, (cfg,)))

    # Health monitor
    if cfg.health.enabled:
        from vigilar.health.monitor import run_health_monitor
        subsystems.append(SubsystemProcess("health-monitor", run_health_monitor, (cfg,)))

    # Handle signals for clean shutdown
    shutdown_requested = False

    def handle_signal(signum: int, frame: Any) -> None:
        nonlocal shutdown_requested
        if shutdown_requested:
            log.warning("Force shutdown")
            sys.exit(1)
        shutdown_requested = True
        log.info("Shutdown requested (signal %d)", signum)

    signal.signal(signal.SIGTERM, handle_signal)
    signal.signal(signal.SIGINT, handle_signal)

    # Start all subsystems
    log.info("Starting %d subsystems", len(subsystems))
    for sub in subsystems:
        sub.start()

    if camera_mgr:
        camera_mgr.start()

    log.info("Vigilar is running")

    # Monitor loop
    try:
        while not shutdown_requested:
            for sub in subsystems:
                if not sub.is_alive():
                    sub.maybe_restart()
            if camera_mgr:
                camera_mgr.check_and_restart()
            time.sleep(2)
    except KeyboardInterrupt:
        pass

    # Shutdown
    log.info("Shutting down subsystems...")
    if camera_mgr:
        camera_mgr.stop()
    for sub in reversed(subsystems):
        sub.stop()

    log.info("Vigilar stopped")