longvideoapi_grayscale_update.py 7.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191
  1. import asyncio
  2. import sys
  3. import time
  4. import requests
  5. import alb_utils
  6. import logging
  7. import os
  8. import docker
  9. import longvideoapi_config
  10. from concurrent.futures import ThreadPoolExecutor
  11. health_instances = []
  12. ess_instances = []
  13. remove_container_instances = []
  14. def server_health_check(ecs_client, instance_id):
  15. """
  16. 服务健康检查
  17. :param ecs_client: 客户端连接
  18. :param instance_id: instanceId
  19. :return:
  20. """
  21. global health_instances
  22. ip_address = alb_utils.get_ip_address(ecs_client=ecs_client, instance_id=instance_id)
  23. while True:
  24. health_check_url = f"http://{ip_address}:8080/longvideoapi/test"
  25. try:
  26. http_code = requests.get(health_check_url).status_code
  27. except:
  28. logging.info(f"images is downloading ip:{ip_address}")
  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. else:
  35. time.sleep(10)
  36. def remove_container_image(ecs_client, instance_id, container_name_list):
  37. """
  38. 移除旧容器并删除旧镜像
  39. :param ecs_client: 客户端连接
  40. :param instance_id: instanceId type-string
  41. :param container_name_list: 容器名称 type-list
  42. :return:
  43. """
  44. ip_address = alb_utils.get_ip_address(ecs_client=ecs_client, instance_id=instance_id)
  45. logging.info(f"服务器信息:{instance_id}/{ip_address}")
  46. client = docker.DockerClient(base_url=f'tcp://{ip_address}:2375', timeout=60)
  47. # 移除旧的容器
  48. container_remove_retry = 3
  49. i = 0
  50. while True:
  51. if i >= container_remove_retry:
  52. logging.error(f"容器不存在或者无法删除当前容器, instance = {instance_id}/{ip_address}")
  53. sys.exit()
  54. try:
  55. flag = False
  56. for container_name in container_name_list:
  57. try:
  58. container_id = client.containers.get(container_name)
  59. container_id.remove(force=True)
  60. flag = True
  61. break
  62. except:
  63. continue
  64. if flag:
  65. break
  66. except Exception as e:
  67. i += 1
  68. # 删除旧镜像
  69. images_remove_retry = 3
  70. j = 0
  71. while True:
  72. if j >= images_remove_retry:
  73. logging.error(f"镜像不存在,无法获取到镜像ID, instance = {instance_id}/{ip_address}")
  74. sys.exit()
  75. try:
  76. images = client.images.list()
  77. for image in images:
  78. client.images.remove(force=True, image=image.tags[0])
  79. time.sleep(2)
  80. global remove_container_instances
  81. remove_container_instances.append(instance_id)
  82. break
  83. except Exception as e:
  84. j += 1
  85. def update_instance(ecs_client, alb_client, instance_id, version, port):
  86. """
  87. 线上机器更新
  88. :param ecs_client:
  89. :param alb_client: alb客户端连接
  90. :param instance_id: instanceId
  91. :param version: 版本标记
  92. :param port: 后端服务器使用的端口
  93. :return:
  94. """
  95. logging.info(f"update instance: {instance_id}")
  96. # 1. 摘流量,将权重降为0
  97. # health_instance_ids = [instance_id]
  98. weight_list = [(0, 20)] # 设置权重为0,等待20秒
  99. alb_utils.update_server_group_servers_attribute(alb_client,
  100. longvideoapi_config.server_group_id_list,
  101. instance_id_list=instance_id,
  102. weight_list=weight_list,
  103. port=port)
  104. logging.info(f"Set weight to 0 for instance: {instance_id}")
  105. # 2. 移除旧容器并删除旧镜像
  106. global remove_container_instances
  107. remove_container_instances = []
  108. container_name_list = ['vlogapi', 'longvideoapi']
  109. remove_container_image(ecs_client, instance_id, container_name_list)
  110. logging.info(f"remove container & images finished, instance: {remove_container_instances},"
  111. f" count: {len(remove_container_instances)}")
  112. if len(remove_container_instances) == 0:
  113. logging.error(f"remove container image failed|")
  114. sys.exit()
  115. # 3. 发送启动脚本到机器上
  116. alb_utils.send_file_to_ecs(ecs_client=ecs_client, instance_id_list=[instance_id], **longvideoapi_config.start_sh)
  117. logging.info(f"send start shell file finished, instance: {instance_id}")
  118. # 4. 启动服务
  119. server_start_sh = os.path.join(longvideoapi_config.start_sh['target_dir'], longvideoapi_config.start_sh['name'])
  120. server_start_command = f"sh {server_start_sh} {version}"
  121. alb_utils.run_command(ecs_client=ecs_client, instance_ids=[instance_id], command=server_start_command)
  122. # 5. 探活
  123. global health_instances
  124. health_instances = []
  125. server_health_check(ecs_client, instance_id)
  126. logging.info(f"health instances: {health_instances}, count: {len(health_instances)}")
  127. # 6. 挂载流量
  128. if len(health_instances) == 0:
  129. logging.info(f"health instances: {health_instances}, count: {len(health_instances)}")
  130. sys.exit()
  131. # 机器探活成功
  132. time.sleep(20)
  133. add_weight_list = [(10, 5), (20, 5), (40, 5), (60, 5), (80, 5), (100, 5)]
  134. alb_utils.update_server_group_servers_attribute(alb_client,
  135. longvideoapi_config.server_group_id_list,
  136. instance_id_list=health_instances,
  137. weight_list=add_weight_list,
  138. port=port)
  139. logging.info(f"finished instances: {health_instances}, count: {len(health_instances)}")
  140. def main():
  141. try:
  142. version = sys.argv[1]
  143. alb_client = alb_utils.connect_alb_client(
  144. access_key_id=longvideoapi_config.alb_client_params['access_key_id'],
  145. access_key_secret=longvideoapi_config.alb_client_params['access_key_secret'],
  146. endpoint=longvideoapi_config.alb_client_params['endpoint']
  147. )
  148. ecs_client = alb_utils.connect_client(access_key_id=longvideoapi_config.ecs_client_params['access_key_id'],
  149. access_key_secret=longvideoapi_config.ecs_client_params[
  150. 'access_key_secret'],
  151. region_id=longvideoapi_config.ecs_client_params['region_id'])
  152. # 1. 获取ALB下所有机器
  153. online_instance_ids = alb_utils.list_server_group_servers(alb_client=alb_client,
  154. server_group_id=longvideoapi_config.server_group_id_list[0])
  155. online_instance_count = len(online_instance_ids)
  156. logging.info(f"online instance count: {online_instance_count}.")
  157. logging.info(f"online instance ids: {online_instance_ids}")
  158. # 2. 对其中一台机器进行更新
  159. instance_id = online_instance_ids[0]
  160. logging.info(f"update instance start ...")
  161. update_instance(ecs_client=ecs_client, alb_client=alb_client, instance_id=instance_id, version=version, port=longvideoapi_config.port)
  162. logging.info(f"update instances end!")
  163. except Exception as e:
  164. logging.error(e)
  165. sys.exit()
  166. if __name__ == '__main__':
  167. main()