longvideoapi_grayscale_update.py 7.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187
  1. import sys
  2. import time
  3. import requests
  4. import utils
  5. import logging
  6. import os
  7. import docker
  8. import longvideoapi_config
  9. health_instances = []
  10. ess_instances = []
  11. remove_container_instances = []
  12. def server_health_check(ecs_client, instance_id):
  13. """
  14. 服务健康检查
  15. :param ecs_client: 客户端连接
  16. :param instance_id: instanceId
  17. :return:
  18. """
  19. global health_instances
  20. ip_address = utils.get_ip_address(ecs_client=ecs_client, instance_id=instance_id)
  21. while True:
  22. health_check_url = f"http://{ip_address}:8080/longvideoapi/test"
  23. try:
  24. http_code = requests.get(health_check_url).status_code
  25. except:
  26. logging.info(f"images is downloading ip:{ip_address}")
  27. http_code = 0
  28. if http_code == 200:
  29. health_instances.append((instance_id, ip_address))
  30. logging.info(f"health check success, instance: {instance_id}/{ip_address}")
  31. break
  32. else:
  33. time.sleep(10)
  34. def remove_container_image(ecs_client, instance_id, container_name_list):
  35. """
  36. 移除旧容器并删除旧镜像
  37. :param ecs_client: 客户端连接
  38. :param instance_id: instanceId type-string
  39. :param container_name_list: 容器名称 type-list
  40. :return:
  41. """
  42. ip_address = utils.get_ip_address(ecs_client=ecs_client, instance_id=instance_id)
  43. logging.info(f"服务器信息:{instance_id}/{ip_address}")
  44. client = docker.DockerClient(base_url=f'tcp://{ip_address}:2375', timeout=60)
  45. # 移除旧的容器
  46. container_remove_retry = 3
  47. i = 0
  48. while True:
  49. if i >= container_remove_retry:
  50. logging.error(f"容器不存在或者无法删除当前容器, instance = {instance_id}/{ip_address}")
  51. sys.exit()
  52. try:
  53. flag = False
  54. for container_name in container_name_list:
  55. try:
  56. container_id = client.containers.get(container_name)
  57. container_id.remove(force=True)
  58. flag = True
  59. break
  60. except:
  61. continue
  62. if flag:
  63. break
  64. except Exception as e:
  65. i += 1
  66. # 删除旧镜像
  67. images_remove_retry = 3
  68. j = 0
  69. while True:
  70. if j >= images_remove_retry:
  71. logging.error(f"镜像不存在,无法获取到镜像ID, instance = {instance_id}/{ip_address}")
  72. sys.exit()
  73. try:
  74. images = client.images.list()
  75. for image in images:
  76. client.images.remove(force=True, image=image.tags[0])
  77. time.sleep(2)
  78. global remove_container_instances
  79. remove_container_instances.append(instance_id)
  80. break
  81. except Exception as e:
  82. j += 1
  83. def update_instance(ecs_client, alb_client, instance_id, version, port):
  84. """
  85. 线上机器更新
  86. :param ecs_client:
  87. :param alb_client: alb客户端连接
  88. :param instance_id: instanceId
  89. :param version: 版本标记
  90. :param port: 后端服务器使用的端口
  91. :return:
  92. """
  93. logging.info(f"update instance: {instance_id}")
  94. # 1. 摘流量,将权重降为0
  95. utils.update_server_group_servers_attribute(alb_client=alb_client,
  96. server_group_id_list=longvideoapi_config.server_group_id_list,
  97. instance_id_list=[instance_id],
  98. weight_list=[(0, 20)],
  99. port=port)
  100. logging.info(f"Set weight to 0 for instance: {instance_id}")
  101. # 2. 移除旧容器并删除旧镜像
  102. global remove_container_instances
  103. remove_container_instances = []
  104. container_name_list = ['vlogapi', 'longvideoapi']
  105. remove_container_image(ecs_client, instance_id, container_name_list)
  106. logging.info(f"remove container & images finished, instance: {remove_container_instances},"
  107. f" count: {len(remove_container_instances)}")
  108. if len(remove_container_instances) == 0:
  109. logging.error(f"remove container image failed|")
  110. sys.exit()
  111. # 3. 发送启动脚本到机器上
  112. utils.send_file_to_ecs(ecs_client=ecs_client, instance_id_list=[instance_id], **longvideoapi_config.start_sh)
  113. logging.info(f"send start shell file finished, instance: {instance_id}")
  114. # 4. 启动服务
  115. server_start_sh = os.path.join(longvideoapi_config.start_sh['target_dir'], longvideoapi_config.start_sh['name'])
  116. server_start_command = f"sh {server_start_sh} {version}"
  117. utils.run_command(ecs_client=ecs_client, instance_ids=[instance_id], command=server_start_command)
  118. # 5. 探活
  119. global health_instances
  120. health_instances = []
  121. server_health_check(ecs_client, instance_id)
  122. logging.info(f"health instances: {health_instances}, count: {len(health_instances)}")
  123. # 6. 挂载流量
  124. if len(health_instances) == 0:
  125. logging.info(f"health instances: {health_instances}, count: {len(health_instances)}")
  126. sys.exit()
  127. # 机器探活成功
  128. time.sleep(20)
  129. health_instance_ids = [instance_id for instance_id, _ in health_instances]
  130. add_weight_list = [(10, 5), (20, 5), (40, 5), (60, 5), (80, 5), (100, 5)]
  131. # add_weight_list = [(10, 10), (20, 10), (40, 10), (60, 10), (80, 10), (100, 10)]
  132. utils.update_server_group_servers_attribute(alb_client=alb_client,
  133. server_group_id_list=longvideoapi_config.server_group_id_list,
  134. instance_id_list=health_instance_ids,
  135. weight_list=add_weight_list,
  136. port=port)
  137. logging.info(f"finished instances: {health_instances}, count: {len(health_instances)}")
  138. def main():
  139. try:
  140. version = sys.argv[1]
  141. alb_client = utils.connect_alb_client(
  142. access_key_id=longvideoapi_config.alb_client_params['access_key_id'],
  143. access_key_secret=longvideoapi_config.alb_client_params['access_key_secret'],
  144. endpoint=longvideoapi_config.alb_client_params['endpoint']
  145. )
  146. ecs_client = utils.connect_client(access_key_id=longvideoapi_config.ecs_client_params['access_key_id'],
  147. access_key_secret=longvideoapi_config.ecs_client_params[
  148. 'access_key_secret'],
  149. region_id=longvideoapi_config.ecs_client_params['region_id'])
  150. # 1. 获取ALB下所有机器
  151. online_instance_ids = utils.list_server_group_servers(alb_client=alb_client,
  152. server_group_id=longvideoapi_config.server_group_id_list[0])
  153. online_instance_count = len(online_instance_ids)
  154. logging.info(f"online instance count: {online_instance_count}.")
  155. logging.info(f"online instance ids: {online_instance_ids}")
  156. # 2. 对其中一台机器进行更新
  157. instance_id = online_instance_ids[0]
  158. logging.info(f"update instance start ...")
  159. update_instance(ecs_client=ecs_client, alb_client=alb_client, instance_id=instance_id, version=version, port=longvideoapi_config.port)
  160. logging.info(f"update instances end!")
  161. except Exception as e:
  162. logging.error(e)
  163. sys.exit()
  164. if __name__ == '__main__':
  165. main()