main.py 4.5 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122
  1. import time
  2. import signal # 新增:导入signal模块
  3. from multiprocessing import Process, cpu_count
  4. from typing import Dict
  5. from core.utils.log.logger_manager import LoggerManager
  6. from scheduler.process_manager import split_topics, start_worker_process
  7. from spiders.spider_registry import SPIDER_CLASS_MAP
  8. # 全局关闭标志
  9. shutdown_flag = False
  10. def handle_signal(sig, frame):
  11. """信号处理函数"""
  12. global shutdown_flag
  13. logger = LoggerManager.get_logger()
  14. logger.warning(f"[主进程] 接收到信号 {sig},开始优雅关闭...")
  15. shutdown_flag = True
  16. def main():
  17. global shutdown_flag
  18. logger = LoggerManager.get_logger()
  19. aliyun_log = LoggerManager.get_aliyun_logger()
  20. # 注册信号处理器 - 关键新增
  21. signal.signal(signal.SIGTERM, handle_signal) # 处理 kill
  22. signal.signal(signal.SIGINT, handle_signal) # 处理 Ctrl+C
  23. signal.signal(signal.SIGHUP, handle_signal) # 处理终端断开
  24. logger.info(f"[主进程] 启动,PID={os.getpid()}") # 记录PID便于管理
  25. topic_list = list(SPIDER_CLASS_MAP.keys())
  26. logger.info(f"[主进程] 监听 Topics: {topic_list}")
  27. max_processes = cpu_count()
  28. num_processes = min(len(topic_list), max_processes)
  29. # topic均分给进程
  30. topic_groups = split_topics(topic_list, num_processes)
  31. logger.info(f"[主进程] CPU 核心数: {max_processes},启动进程数: {num_processes}")
  32. process_map: Dict[int, Process] = {}
  33. for group_id, topic_group in enumerate(topic_groups):
  34. start_worker_process(group_id, topic_group, process_map)
  35. # 重启状态跟踪
  36. restart_count = {group_id: 0 for group_id in range(len(topic_groups))}
  37. last_restart_time = {group_id: 0 for group_id in range(len(topic_groups))}
  38. # 主进程持续监控子进程状态
  39. try:
  40. while not shutdown_flag: # 检查关闭标志
  41. time.sleep(5)
  42. for group_id, p in list(process_map.items()):
  43. if shutdown_flag: # 如果正在关闭,跳过检查
  44. break
  45. if not p.is_alive():
  46. current_time = time.time()
  47. # 重启限制逻辑 - 防止无限重启
  48. if restart_count[group_id] > 5 and (current_time - last_restart_time[group_id]) < 60:
  49. logger.error(f"[监控] 进程 {p.name} 频繁崩溃,停止重启!")
  50. aliyun_log.logging(
  51. code="1502",
  52. message="进程频繁崩溃已停止重启",
  53. data={
  54. "pid": p.pid,
  55. "exitcode": p.exitcode,
  56. "restart_count": restart_count[group_id],
  57. "group_id": group_id,
  58. "topics": topic_groups[group_id]
  59. }
  60. )
  61. continue
  62. logger.warning(f"[监控] 进程 {p.name} PID={p.pid} 已崩溃,正在重启...")
  63. aliyun_log.logging(
  64. code="1501",
  65. message=f"[主进程监控] 子进程 {p.name} 崩溃重启",
  66. data={
  67. "pid": p.pid,
  68. "exitcode": p.exitcode,
  69. "group_id": group_id,
  70. "topics": topic_groups[group_id],
  71. "restart_count": restart_count[group_id] + 1
  72. }
  73. )
  74. # 更新重启状态
  75. restart_count[group_id] += 1
  76. last_restart_time[group_id] = current_time
  77. time.sleep(2)
  78. start_worker_process(group_id, topic_groups[group_id], process_map)
  79. except KeyboardInterrupt:
  80. logger.warning("[主进程] 接收到退出信号,终止所有子进程...")
  81. shutdown_flag = True
  82. # 优雅终止所有子进程 - 统一处理关闭
  83. logger.info("[主进程] 终止所有子进程...")
  84. for p in process_map.values():
  85. if p.is_alive():
  86. p.terminate() # 先尝试正常终止
  87. # 等待子进程退出
  88. for p in process_map.values():
  89. p.join(timeout=5) # 最多等待5秒
  90. if p.is_alive():
  91. logger.warning(f"进程 {p.pid} 未正常退出,强制终止")
  92. p.kill() # 强制终止
  93. p.join()
  94. logger.info("[主进程] 所有子进程已终止,退出")
  95. if __name__ == '__main__':
  96. import os
  97. main()