alb_gateway_scaling_j_count.py 5.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131
  1. import sys
  2. import os
  3. import asyncio
  4. import logging
  5. import time
  6. import requests
  7. import alb_utils
  8. import gateway_config
  9. from concurrent.futures import ThreadPoolExecutor
  10. health_instances = []
  11. def gateway_health_check(ecs_client, instance_id, max_wait_time=None):
  12. """
  13. 服务健康检查
  14. :param ecs_client: 客户端连接
  15. :param instance_id: instanceId
  16. :param max_wait_time: 最长等待时间,单位:s
  17. :return:
  18. """
  19. global health_instances
  20. start_time = time.time()
  21. ip_address = alb_utils.get_ip_address(ecs_client=ecs_client, instance_id=instance_id)
  22. while True:
  23. health_check_url = f"http://{ip_address}:9000/healthcheck"
  24. try:
  25. http_code = requests.get(health_check_url).status_code
  26. except:
  27. logging.info(f"images is downloading ip: {ip_address}")
  28. http_code = 0
  29. if http_code == 200:
  30. health_instances.append((instance_id, ip_address))
  31. logging.info(f"health check success, instance: {instance_id}/{ip_address}")
  32. break
  33. elif max_wait_time is not None:
  34. now = time.time()
  35. if (now - start_time) >= max_wait_time:
  36. logging.info(f"health check error, instance: {instance_id}/{ip_address}")
  37. break
  38. else:
  39. time.sleep(10)
  40. else:
  41. time.sleep(10)
  42. async def ess_instance(ecs_client, alb_client, ess_count, max_workers, port):
  43. """
  44. 扩容机器并运行新服务
  45. :param ecs_client: 购买机器客户端连接
  46. :param alb_client: 修改负载均衡权限
  47. :param ess_count: 扩容数量
  48. :param max_workers: 线程数
  49. :param port: 后端服务器使用的端口
  50. :return:
  51. """
  52. # 1. 购买机器并启动
  53. ess_instance_ids = alb_utils.create_multiple_instances(
  54. amount=ess_count,
  55. ecs_client=ecs_client,
  56. **gateway_config.instance_config_j_alb,
  57. )
  58. time.sleep(60)
  59. # 2. 发送启动脚本到机器上
  60. alb_utils.send_file_to_ecs(ecs_client=ecs_client, instance_id_list=ess_instance_ids, **gateway_config.start_sh)
  61. logging.info(f"send start shell file finished, count: {len(ess_instance_ids)} instances: {ess_instance_ids}")
  62. # 3. 启动服务
  63. start_sh_param = "latest"
  64. server_start_sh = os.path.join(gateway_config.start_sh['target_dir'], gateway_config.start_sh['name'])
  65. server_start_commend = f"sh {server_start_sh} {start_sh_param}"
  66. alb_utils.run_command(ecs_client=ecs_client, instance_ids=ess_instance_ids, command=server_start_commend)
  67. # 4. 异步探活
  68. global health_instances
  69. health_instances = []
  70. max_wait_time = 360
  71. loop = asyncio.get_running_loop()
  72. executor = ThreadPoolExecutor(max_workers=max_workers)
  73. tasks = [
  74. loop.run_in_executor(executor, gateway_health_check, *args) for args in
  75. [(ecs_client, instance_id, max_wait_time) for instance_id in ess_instance_ids]
  76. ]
  77. await asyncio.wait(tasks)
  78. logging.info(f"health instances count: {len(health_instances)}, {health_instances}")
  79. # 5. 挂载流量到ALB
  80. if len(health_instances) > 0:
  81. # 所有机器探活成功
  82. time.sleep(20)
  83. # instance_id_list = []
  84. # for instance_id, ip in health_instances:
  85. # instance_id_list.append(instance_id)
  86. # # for server_group_id in gateway_config.server_group_id_list:
  87. health_instance_ids = [instance_id for instance_id, _ in health_instances]
  88. alb_utils.add_servers_to_server_group(alb_client, gateway_config.server_group_id_list, health_instance_ids, weight=100, port=port)
  89. logging.info(f"ess count: {ess_count}, "
  90. f"create count: {len(ess_instance_ids)}, "
  91. f"finished count: {len(health_instances)}")
  92. else:
  93. logging.info(f"ess count: {ess_count}, "
  94. f"create count: {len(ess_instance_ids)}, "
  95. f"health count: {len(health_instances)}")
  96. sys.exit()
  97. def main():
  98. try:
  99. ecs_client = alb_utils.connect_client(access_key_id=gateway_config.ecs_client_params['access_key_id'],
  100. access_key_secret=gateway_config.ecs_client_params['access_key_secret'],
  101. region_id=gateway_config.ecs_client_params['region_id'])
  102. alb_client = alb_utils.connect_alb_client(
  103. access_key_id=gateway_config.alb_client_params['access_key_id'],
  104. access_key_secret=gateway_config.alb_client_params['access_key_secret'],
  105. endpoint=gateway_config.alb_client_params['endpoint']
  106. )
  107. # 获取批量创建ECS实例的数量
  108. ess_instance_count = int(sys.argv[1])
  109. # 扩容机器并启动服务
  110. logging.info(f"ess instances start ...")
  111. logging.info(f"ess instance count: {ess_instance_count}")
  112. asyncio.run(ess_instance(ecs_client=ecs_client,
  113. alb_client=alb_client,
  114. ess_count=ess_instance_count, max_workers=2, port=gateway_config.port))
  115. logging.info(f"ess instances end!")
  116. except Exception as e:
  117. logging.error(e)
  118. if __name__ == '__main__':
  119. main()