优化

arlo.py

@@ -1,5 +1,16 @@
+"""Entrypoint for the scheduler process.
+
+This module starts the APScheduler scheduler and ensures a graceful
+shutdown on SIGINT/SIGTERM. It also improves job error logging to
+include exception tracebacks for easier debugging.
+"""
+
 import datetime
+import signal
+import sys
+import traceback
 from functools import partial
+from typing import Any
 
 from apscheduler.events import EVENT_JOB_ERROR
 from apscheduler.schedulers.blocking import BlockingScheduler
@@ -9,28 +20,92 @@ from log.log_manager import log, logger
 from task.manager_task import manager_task
 
 
-def job_error_listener(event):
-    if event.exception:
-        logger.error(f"Job {event.job_id} crashed: {str(event.exception)}")
-        # 可添加邮件/钉钉告警逻辑
+def _format_exception(exc: BaseException) -> str:
+    """Return a nicely formatted exception with traceback."""
+    return "".join(traceback.format_exception(type(exc), exc, exc.__traceback__))
 
 
-if __name__ == '__main__':
+def job_error_listener(event: Any) -> None:
+    """Listener for job errors that logs the exception and traceback.
+
+    Uses the APScheduler event object. This is defensive: if an exception
+    is present it logs its string representation and the full traceback to
+    the configured logger. Keep this lightweight to avoid raising inside
+    the listener.
+    """
+    try:
+        if getattr(event, "exception", None):
+            exc = event.exception
+            logger.error(f"Job {getattr(event, 'job_id', '<unknown>')} crashed: {exc}")
+            logger.error(_format_exception(exc))
+            # 可添加邮件告警逻辑（如果需要且已配置）
+    except Exception:
+        # We must not let the listener raise — log and continue.
+        logger.exception("Exception in job_error_listener")
+
+
+def _validate_interval(value: Any) -> int:
+    """Return a valid positive integer interval in seconds.
+
+    If the provided value is invalid, returns a safe default (300).
+    """
+    DEFAULT = 300
+    try:
+        iv = int(value)
+        if iv <= 0:
+            raise ValueError("interval must be > 0")
+        return iv
+    except Exception:
+        logger.warning(
+            "Invalid config.scheduler_interval=%r, falling back to %s seconds",
+            value,
+            DEFAULT,
+        )
+        return DEFAULT
+
+
+def main() -> None:
+    """Create and start the scheduler with graceful shutdown handling."""
     scheduler = BlockingScheduler()
-    # 每隔config.scheduler_interval秒执行一次任务，同时设定第一次执行在程序启动后10秒后执行
+
+    interval_seconds = _validate_interval(getattr(config, "scheduler_interval", None))
+
+    # 每隔 interval_seconds 秒执行一次任务，同时设定第一次执行在程序启动10秒后执行
     scheduler.add_job(
         partial(manager_task, scheduler),
-        'interval',
-        seconds=config.scheduler_interval,
-        jitter=30,  # 添加随机抖动避免任务雪崩
-        next_run_time=datetime.datetime.now() + datetime.timedelta(seconds=10)  # 替代 date 触发器
+        "interval",
+        seconds=interval_seconds,
+        jitter=30,
+        next_run_time=datetime.datetime.now() + datetime.timedelta(seconds=10),
     )
 
     # 添加任务错误监听器
     scheduler.add_listener(job_error_listener, EVENT_JOB_ERROR)
 
+    # Graceful shutdown handlers
+    def _shutdown(signum, frame):
+        log(f"Received signal {signum}. Shutting down scheduler...")
+        try:
+            scheduler.shutdown(wait=False)
+        except Exception:
+            logger.exception("Error while shutting down scheduler")
+        # Ensure process exits after scheduler shutdown
+        # sys.exit(0)
+
+    signal.signal(signal.SIGINT, _shutdown)
+    signal.signal(signal.SIGTERM, _shutdown)
+
     try:
         log("started successfully.")
         scheduler.start()  # 阻塞运行
     except (KeyboardInterrupt, SystemExit):
         log("Shutting down ...")
+        if scheduler.state == 1:
+            try:
+                scheduler.shutdown(wait=False)
+            except Exception:
+                logger.exception("Error while shutting down scheduler")
+
+
+if __name__ == "__main__":
+    main()