longvideoapi_reduce_with_count.py 4.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118
  1. import sys
  2. import time
  3. import utils
  4. import longvideoapi_config
  5. import logging
  6. logging.basicConfig(level=logging.INFO,
  7. format='%(asctime)s %(filename)s[line:%(lineno)d] %(levelname)s %(message)s',
  8. datefmt='%a, %d %b %Y %H:%M:%S')
  9. def remove_instances(ecs_client, alb_client, instance_ids):
  10. """
  11. 停止并释放机器
  12. :param ecs_client: 创建客户端
  13. :param alb_client: ALB 客户端
  14. :param instance_ids: instanceId 类型列表
  15. :return: None
  16. """
  17. # 1. 摘除流量
  18. # todo 20
  19. weight_list = [(0, 20)] # 设置权重为0,等待20秒
  20. try:
  21. utils.set_instance_weight_process_with_alb(alb_client,
  22. longvideoapi_config.server_group_id_list,
  23. instance_ids,
  24. weight_list)
  25. except Exception as e:
  26. logging.error(f"Failed to set instance weight: {e}")
  27. sys.exit()
  28. # 等待 ALB 更新权重
  29. time.sleep(10)
  30. # 2. 从 ALB 服务器组中移除实例
  31. for server_group_id in longvideoapi_config.server_group_id_list:
  32. try:
  33. utils.remove_servers_from_server_group(alb_client=alb_client, server_group_id=server_group_id,
  34. instance_id=instance_ids)
  35. logging.info(f"Successfully removed instances from server group {server_group_id}.")
  36. except Exception as e:
  37. logging.error(f"Failed to remove instances from server group {server_group_id}: {e}")
  38. logging.info(f"Remove from ALB finished, instances: {instance_ids}")
  39. # 3. 停止机器
  40. stop_response = utils.stop_instances(ecs_client=ecs_client, instance_ids=instance_ids)
  41. if stop_response.get('Code') is None:
  42. logging.info(f"Instances stop finished, instances: {instance_ids}")
  43. else:
  44. logging.error(f"Failed to stop instances: {stop_response}")
  45. sys.exit()
  46. # 4. 判断机器运行状态是否为 Stopped
  47. while True:
  48. response = utils.get_instances_status(ecs_client=ecs_client, instance_ids=instance_ids)
  49. if response.get('Code') is None:
  50. instances_list = response.get('InstanceStatuses').get('InstanceStatus')
  51. stopped_instances = [instance.get('InstanceId') for instance in instances_list if
  52. instance.get('Status') == 'Stopped']
  53. if len(stopped_instances) == len(instance_ids):
  54. logging.info(f"Instances stopped status set success, instances: {stopped_instances}")
  55. break
  56. else:
  57. logging.info(f"Stopped instances count = {len(stopped_instances)}, instances: {stopped_instances}")
  58. time.sleep(5)
  59. else:
  60. logging.error(response)
  61. sys.exit()
  62. # 5. 释放机器
  63. release_response = utils.release_instances(ecs_client=ecs_client, instance_ids=stopped_instances)
  64. if release_response.get('Code') is None:
  65. logging.info(f"Release instances finished, instances: {stopped_instances}")
  66. else:
  67. logging.error(f"Release instances fail!!!")
  68. sys.exit()
  69. def main():
  70. try:
  71. alb_client = utils.connect_alb_client(
  72. access_key_id=longvideoapi_config.alb_client_params['access_key_id'],
  73. access_key_secret=longvideoapi_config.alb_client_params['access_key_secret']
  74. )
  75. # todo
  76. ecs_client = utils.connect_client(
  77. access_key_id=longvideoapi_config.ecs_client_params['access_key_id'],
  78. access_key_secret=longvideoapi_config.ecs_client_params['access_key_secret'],
  79. region_id=longvideoapi_config.ecs_client_params['region_id']
  80. )
  81. # 获取指定释放的机器数量
  82. reduce_count = int(sys.argv[1])
  83. logging.info(f"reduce instances count: {reduce_count}")
  84. # 获取 ALB 下所有机器
  85. online_instance_ids = utils.get_instance_ids(alb_client=alb_client, server_group_id=longvideoapi_config.server_group_id_list[0])
  86. online_instance_count = len(online_instance_ids)
  87. logging.info(f"online instance count: {online_instance_count}.")
  88. logging.info(f"online instance ids: {online_instance_ids}")
  89. # 获取前 count 台机器进行释放
  90. reduce_instance_ids = online_instance_ids[:reduce_count]
  91. logging.info(f"reduce instances: {reduce_instance_ids}")
  92. # 停止并释放机器
  93. remove_instances(ecs_client=ecs_client, alb_client=alb_client, instance_ids=reduce_instance_ids)
  94. logging.info(f"stop & release instances end!")
  95. except Exception as e:
  96. logging.error(e)
  97. sys.exit()
  98. if __name__ == '__main__':
  99. main()