ess_instance_rov_server.py 5.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130
  1. import sys
  2. import os
  3. import asyncio
  4. import logging
  5. import time
  6. import requests
  7. import utils
  8. import rov_server_config
  9. from concurrent.futures import ThreadPoolExecutor
  10. health_instances = []
  11. ess_instances = []
  12. def rov_server_health_check(client, instance_id, max_wait_time=None):
  13. """
  14. 服务健康检查
  15. :param client: 客户端连接
  16. :param instance_id: instanceId
  17. :param max_wait_time: 最长等待时间,单位:s
  18. :return:
  19. """
  20. global health_instances
  21. start_time = time.time()
  22. ip_address = utils.get_ip_address(client=client, instance_id=instance_id)
  23. while True:
  24. health_check_url = f"http://{ip_address}:5001/healthcheck"
  25. try:
  26. http_code = requests.get(health_check_url).status_code
  27. except:
  28. logging.info("images is downloading")
  29. http_code = 0
  30. if http_code == 200:
  31. health_instances.append((instance_id, ip_address))
  32. logging.info(f"health check success, instance: {instance_id}/{ip_address}")
  33. break
  34. elif max_wait_time is not None:
  35. now = time.time()
  36. if (now - start_time) >= max_wait_time:
  37. logging.info(f"health check error, instance: {instance_id}/{ip_address}")
  38. break
  39. else:
  40. time.sleep(10)
  41. else:
  42. time.sleep(10)
  43. async def ess_instance(create_client, slb_client, ess_count, max_workers):
  44. """
  45. 扩容机器并运行新服务
  46. :param create_client: 购买机器客户端连接
  47. :param slb_client: 修改负载均衡权限
  48. :param ess_count: 扩容数量
  49. :param max_workers: 线程数
  50. :return:
  51. """
  52. # 1. 购买机器并启动
  53. ess_instance_ids = utils.create_multiple_instances(
  54. amount=ess_count,
  55. client=create_client,
  56. **rov_server_config.instance_config,
  57. )
  58. time.sleep(60)
  59. # 2. 发送启动脚本到机器上
  60. utils.send_file_to_ecs(client=create_client, instance_id_list=ess_instance_ids, **rov_server_config.start_sh)
  61. logging.info(f"send start shell file finished, instances: {ess_instance_ids}")
  62. # 3. 启动服务
  63. server_start_sh = os.path.join(rov_server_config.start_sh['target_dir'], rov_server_config.start_sh['name'])
  64. server_start_commend = f"sh {server_start_sh}"
  65. utils.run_command(client=create_client, instance_ids=ess_instance_ids, command=server_start_commend)
  66. # 4. 异步探活
  67. global health_instances
  68. health_instances = []
  69. max_wait_time = 120
  70. loop = asyncio.get_running_loop()
  71. executor = ThreadPoolExecutor(max_workers=max_workers)
  72. tasks = [
  73. loop.run_in_executor(executor, rov_server_health_check, *args) for args in
  74. [(slb_client, instance_id, max_wait_time) for instance_id in ess_instance_ids]
  75. ]
  76. await asyncio.wait(tasks)
  77. logging.info(f"health instances count: {len(health_instances)}, {health_instances}")
  78. # 5. 挂载流量
  79. if len(health_instances) > 0:
  80. # 所有机器探活成功
  81. time.sleep(60)
  82. add_weight_list = [(10, 30), (20, 20), (40, 10), (60, 10), (80, 10), (100, 10)]
  83. # utils.set_instance_weight_process(client=slb_client,
  84. # slb_id=rov_server_config.slb_id,
  85. # instance_id_list=ess_instance_ids,
  86. # weight_list=add_weight_list)
  87. global ess_instances
  88. ess_instances.extend(ess_instance_ids)
  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. slb_client = utils.connect_client(access_key_id=rov_server_config.slb_client_params['access_key_id'],
  100. access_key_secret=rov_server_config.slb_client_params['access_key_secret'],
  101. region_id=rov_server_config.slb_client_params['region_id'])
  102. create_client = utils.connect_client(access_key_id=rov_server_config.create_client_params['access_key_id'],
  103. access_key_secret=rov_server_config.create_client_params['access_key_secret'],
  104. region_id=rov_server_config.create_client_params['region_id'])
  105. # 获取批量创建ECS实例的数量
  106. ess_instance_count = int(sys.argv[1])
  107. # 扩容机器并启动服务
  108. logging.info(f"ess instances start ...")
  109. logging.info(f"ess instance count: {ess_instance_count}")
  110. asyncio.run(ess_instance(create_client=create_client, slb_client=slb_client,
  111. ess_count=ess_instance_count, max_workers=2))
  112. logging.info(f"ess instances end!")
  113. except Exception as e:
  114. logging.error(e)
  115. if __name__ == '__main__':
  116. main()