alb_gateway_reduce_with_count.py 5.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130
  1. import sys
  2. import time
  3. import alb_utils
  4. import gateway_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, port):
  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. :param port: 后端服务器使用的端口
  16. :return: None
  17. """
  18. # 1. 摘除流量
  19. weight_list = [(0, 20)] # 设置权重为0,等待20秒
  20. try:
  21. alb_utils.update_server_group_servers_attribute(alb_client,
  22. gateway_config.server_group_id_list,
  23. instance_id_list=instance_ids,
  24. weight_list=weight_list,
  25. port=port)
  26. except Exception as e:
  27. logging.error(f"Failed to set instance weight: {e}")
  28. sys.exit()
  29. # 等待 ALB 更新权重
  30. time.sleep(20)
  31. # 2. 从 ALB 服务器组中移除实例
  32. # for server_group_id in gateway_config.server_group_id_list:
  33. try:
  34. alb_utils.remove_servers_from_server_group(alb_client=alb_client,
  35. server_group_id_list=gateway_config.server_group_id_list,
  36. instance_ids=instance_ids, port=port)
  37. logging.info(
  38. f"Successfully removed count:{len(instance_ids)} instance_ids: {instance_ids} from server group {gateway_config.server_group_id_list}.")
  39. except Exception as e:
  40. logging.error(
  41. f"Failed to remove count:{len(instance_ids)} instance_ids: {instance_ids} from server group {gateway_config.server_group_id_list}: {e}")
  42. logging.info(f"Remove from ALB finished, count:{len(instance_ids)} instances: {instance_ids}")
  43. # 3. 停止机器
  44. alb_utils.stop_instances(ecs_client=ecs_client, instance_ids=instance_ids)
  45. # stop_response = alb_utils.stop_instances(ecs_client=ecs_client, instance_ids=instance_ids)
  46. # if stop_response.get('Code') is None:
  47. # logging.info(f"Instances stop finished, instances: {instance_ids}")
  48. # else:
  49. # logging.error(f"Failed to stop instances: {stop_response}")
  50. # sys.exit()
  51. # 4. 判断机器运行状态是否为 Stopped
  52. stopped_instances = alb_utils.get_instances_status(ecs_client=ecs_client, instance_ids=instance_ids)
  53. # while True:
  54. # response = alb_utils.get_instances_status(ecs_client=ecs_client, instance_ids=instance_ids)
  55. # if response.get('Code') is None:
  56. # instances_list = response.get('InstanceStatuses').get('InstanceStatus')
  57. # stopped_instances = [instance.get('InstanceId') for instance in instances_list if
  58. # instance.get('Status') == 'Stopped']
  59. # if len(stopped_instances) == len(instance_ids):
  60. # logging.info(f"Instances stopped status set success, instances: {stopped_instances}")
  61. # break
  62. # else:
  63. # logging.info(f"Stopped instances count = {len(stopped_instances)}, instances: {stopped_instances}")
  64. # time.sleep(5)
  65. # else:
  66. # logging.error(response)
  67. # sys.exit()
  68. # 5. 释放机器
  69. alb_utils.release_instances(ecs_client=ecs_client, instance_ids=stopped_instances)
  70. # release_response = alb_utils.release_instances(ecs_client=ecs_client, instance_ids=stopped_instances)
  71. # if release_response.get('Code') is None:
  72. # logging.info(f"Release instances finished, instances: {stopped_instances}")
  73. # else:
  74. # logging.error(f"Release instances fail!!!")
  75. # sys.exit()
  76. def main():
  77. try:
  78. alb_client = alb_utils.connect_alb_client(
  79. access_key_id=gateway_config.alb_client_params['access_key_id'],
  80. access_key_secret=gateway_config.alb_client_params['access_key_secret'],
  81. endpoint=gateway_config.alb_client_params['endpoint']
  82. )
  83. ecs_client = alb_utils.connect_client(
  84. access_key_id=gateway_config.ecs_client_params['access_key_id'],
  85. access_key_secret=gateway_config.ecs_client_params['access_key_secret'],
  86. region_id=gateway_config.ecs_client_params['region_id']
  87. )
  88. # 获取指定释放的机器数量
  89. reduce_count = int(sys.argv[1])
  90. logging.info(f"reduce_count: {reduce_count}")
  91. # 获取 ALB 下所有机器
  92. online_instance_ids = alb_utils.list_server_group_servers(alb_client=alb_client,
  93. server_group_id=gateway_config.server_group_id_list[
  94. 0])
  95. online_instance_count = len(online_instance_ids)
  96. logging.info(f"online instance count: {online_instance_count} instance_ids: {online_instance_ids}")
  97. # if online_instance_count - reduce_count < 10:
  98. # logging.error("缩容后服务器数量不能小于10台")
  99. # sys.exit()
  100. # 获取前 count 台机器进行释放
  101. reduce_instance_ids = online_instance_ids[:reduce_count]
  102. logging.info(f"reduce instances count = {len(reduce_instance_ids)} instances: {reduce_instance_ids}")
  103. # 停止并释放机器
  104. remove_instances(ecs_client=ecs_client, alb_client=alb_client, instance_ids=reduce_instance_ids,
  105. port=gateway_config.port)
  106. logging.info(f"stop & release instances end!")
  107. except Exception as e:
  108. logging.error(e)
  109. sys.exit()
  110. if __name__ == '__main__':
  111. main()