rov_sever_remove_failed_instances.py 2.4 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859
  1. import requests
  2. import utils
  3. import rov_server_config
  4. import logging
  5. logging.basicConfig(level=logging.INFO,
  6. format='%(asctime)s %(filename)s[line:%(lineno)d] %(levelname)s %(message)s',
  7. datefmt='%a, %d %b %Y %H:%M:%S')
  8. def get_failed_sever_instances(slb_client, instance_ids):
  9. """获取探活失败的机器"""
  10. failed_instances = []
  11. for instance_id in instance_ids:
  12. ip_address = utils.get_ip_address(client=slb_client, instance_id=instance_id)
  13. health_check_url = f"http://{ip_address}:5001/healthcheck"
  14. try:
  15. http_code = requests.get(health_check_url).status_code
  16. except:
  17. failed_instances.append(instance_id)
  18. logging.info(f"failed instances: {instance_id}/{ip_address}")
  19. continue
  20. if http_code == 200:
  21. continue
  22. else:
  23. failed_instances.append(instance_id)
  24. logging.info(f"failed instances: {instance_id}/{ip_address}")
  25. return failed_instances
  26. def main():
  27. """移除服务挂掉的机器"""
  28. slb_client = utils.connect_client(access_key_id=rov_server_config.slb_client_params['access_key_id'],
  29. access_key_secret=rov_server_config.slb_client_params['access_key_secret'],
  30. region_id=rov_server_config.slb_client_params['region_id'])
  31. # 1. 获取slb下所有机器
  32. online_instance_ids = utils.get_instance_ids(client=slb_client, slb_id=rov_server_config.slb_id)
  33. online_instance_count = len(online_instance_ids)
  34. logging.info(f"online instance count: {online_instance_count}.")
  35. logging.info(f"online instance ids: {online_instance_ids}")
  36. # 2. 获取探活失败的机器
  37. failed_instances = get_failed_sever_instances(slb_client=slb_client, instance_ids=online_instance_ids)
  38. logging.info(f"failed instances count: {len(failed_instances)}")
  39. logging.info(f"failed instance ids: {failed_instances}")
  40. # 3. 摘流量
  41. if len(failed_instances) > 0:
  42. utils.set_instance_weight_process(client=slb_client,
  43. slb_id=rov_server_config.slb_id,
  44. instance_id_list=failed_instances,
  45. weight_list=[(0, 60)])
  46. logging.info(f"set weight with 0 finished.")
  47. else:
  48. logging.info("not have failed instances!")
  49. if __name__ == '__main__':
  50. main()