docker_helper.py 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389
  1. import os
  2. import socket
  3. import json
  4. import docker
  5. import time
  6. import re
  7. import traceback
  8. from threading import Thread
  9. from colorama import Fore, Style
  10. from toolset.utils.output_helper import log
  11. from toolset.utils.database_helper import test_database
  12. class DockerHelper:
  13. def __init__(self, benchmarker=None):
  14. self.benchmarker = benchmarker
  15. self.client = docker.DockerClient(
  16. base_url=self.benchmarker.config.client_docker_host)
  17. self.server = docker.DockerClient(
  18. base_url=self.benchmarker.config.server_docker_host)
  19. self.database = docker.DockerClient(
  20. base_url=self.benchmarker.config.database_docker_host)
  21. def __build(self, base_url, path, build_log_file, log_prefix, dockerfile,
  22. tag):
  23. '''
  24. Builds docker containers using docker-py low-level api
  25. '''
  26. self.benchmarker.time_logger.mark_build_start()
  27. with open(build_log_file, 'w') as build_log:
  28. try:
  29. client = docker.APIClient(base_url=base_url)
  30. output = client.build(
  31. path=path,
  32. dockerfile=dockerfile,
  33. tag=tag,
  34. forcerm=True,
  35. timeout=3600,
  36. pull=True,
  37. buildargs=({
  38. 'BENCHMARK_ENV': self.benchmarker.config.benchmark_env
  39. })
  40. )
  41. buffer = ""
  42. for token in output:
  43. if token.startswith('{"stream":'):
  44. token = json.loads(token)
  45. token = token[token.keys()[0]].encode('utf-8')
  46. buffer += token
  47. elif token.startswith('{"errorDetail":'):
  48. token = json.loads(token)
  49. raise Exception(token['errorDetail']['message'])
  50. while "\n" in buffer:
  51. index = buffer.index("\n")
  52. line = buffer[:index]
  53. buffer = buffer[index + 1:]
  54. log(line,
  55. prefix=log_prefix,
  56. file=build_log,
  57. color=Fore.WHITE + Style.BRIGHT \
  58. if re.match(r'^Step \d+\/\d+', line) else '')
  59. if buffer:
  60. log(buffer,
  61. prefix=log_prefix,
  62. file=build_log,
  63. color=Fore.WHITE + Style.BRIGHT \
  64. if re.match(r'^Step \d+\/\d+', buffer) else '')
  65. except Exception:
  66. tb = traceback.format_exc()
  67. log("Docker build failed; terminating",
  68. prefix=log_prefix,
  69. file=build_log,
  70. color=Fore.RED)
  71. log(tb, prefix=log_prefix, file=build_log)
  72. self.benchmarker.time_logger.log_build_end(
  73. log_prefix=log_prefix, file=build_log)
  74. raise
  75. self.benchmarker.time_logger.log_build_end(
  76. log_prefix=log_prefix, file=build_log)
  77. def clean(self):
  78. '''
  79. Cleans all the docker images from the system
  80. '''
  81. self.server.images.prune()
  82. for image in self.server.images.list():
  83. if len(image.tags) > 0:
  84. # 'techempower/tfb.test.gemini:0.1' -> 'techempower/tfb.test.gemini'
  85. image_tag = image.tags[0].split(':')[0]
  86. if image_tag != 'techempower/tfb' and 'techempower' in image_tag:
  87. self.server.images.remove(image.id, force=True)
  88. self.server.images.prune()
  89. self.database.images.prune()
  90. for image in self.database.images.list():
  91. if len(image.tags) > 0:
  92. # 'techempower/tfb.test.gemini:0.1' -> 'techempower/tfb.test.gemini'
  93. image_tag = image.tags[0].split(':')[0]
  94. if image_tag != 'techempower/tfb' and 'techempower' in image_tag:
  95. self.database.images.remove(image.id, force=True)
  96. self.database.images.prune()
  97. def build(self, test, build_log_dir=os.devnull):
  98. '''
  99. Builds the test docker containers
  100. '''
  101. log_prefix = "%s: " % test.name
  102. # Build the test image
  103. test_docker_file = "%s.dockerfile" % test.name
  104. build_log_file = build_log_dir
  105. if build_log_dir is not os.devnull:
  106. build_log_file = os.path.join(
  107. build_log_dir,
  108. "%s.log" % test_docker_file.replace(".dockerfile", "").lower())
  109. try:
  110. self.__build(
  111. base_url=self.benchmarker.config.server_docker_host,
  112. build_log_file=build_log_file,
  113. log_prefix=log_prefix,
  114. path=test.directory,
  115. dockerfile=test_docker_file,
  116. tag="techempower/tfb.test.%s" % test_docker_file.replace(
  117. ".dockerfile", ""))
  118. except Exception:
  119. return 1
  120. return 0
  121. def run(self, test, run_log_dir):
  122. '''
  123. Run the given Docker container(s)
  124. '''
  125. log_prefix = "%s: " % test.name
  126. container = None
  127. try:
  128. def watch_container(docker_container, docker_file):
  129. with open(
  130. os.path.join(
  131. run_log_dir, "%s.log" % docker_file.replace(
  132. ".dockerfile", "").lower()), 'w') as run_log:
  133. for line in docker_container.logs(stream=True):
  134. log(line, prefix=log_prefix, file=run_log)
  135. extra_hosts = None
  136. name = "tfb-server"
  137. if self.benchmarker.config.network is None:
  138. extra_hosts = {
  139. socket.gethostname():
  140. str(self.benchmarker.config.server_host),
  141. 'tfb-server':
  142. str(self.benchmarker.config.server_host),
  143. 'tfb-database':
  144. str(self.benchmarker.config.database_host)
  145. }
  146. name = None
  147. sysctl = {'net.core.somaxconn': 65535}
  148. ulimit = [{
  149. 'name': 'nofile',
  150. 'hard': 200000,
  151. 'soft': 200000
  152. }, {
  153. 'name': 'rtprio',
  154. 'hard': 99,
  155. 'soft': 99
  156. }]
  157. container = self.server.containers.run(
  158. "techempower/tfb.test.%s" % test.name,
  159. name=name,
  160. network=self.benchmarker.config.network,
  161. network_mode=self.benchmarker.config.network_mode,
  162. stderr=True,
  163. detach=True,
  164. init=True,
  165. extra_hosts=extra_hosts,
  166. privileged=True,
  167. ulimits=ulimit,
  168. sysctls=sysctl,
  169. remove=True,
  170. log_config={'type': None})
  171. watch_thread = Thread(
  172. target=watch_container,
  173. args=(
  174. container,
  175. "%s.dockerfile" % test.name,
  176. ))
  177. watch_thread.daemon = True
  178. watch_thread.start()
  179. except Exception:
  180. with open(
  181. os.path.join(run_log_dir, "%s.log" % test.name.lower()),
  182. 'w') as run_log:
  183. tb = traceback.format_exc()
  184. log("Running docker container: %s.dockerfile failed" %
  185. test.name,
  186. prefix=log_prefix,
  187. file=run_log)
  188. log(tb, prefix=log_prefix, file=run_log)
  189. return container
  190. @staticmethod
  191. def __stop_container(container):
  192. try:
  193. container.kill()
  194. time.sleep(2)
  195. except:
  196. # container has already been killed
  197. pass
  198. @staticmethod
  199. def __stop_all(docker_client):
  200. for container in docker_client.containers.list():
  201. if len(container.image.tags) > 0 \
  202. and 'techempower' in container.image.tags[0] \
  203. and 'tfb:latest' not in container.image.tags[0]:
  204. DockerHelper.__stop_container(container)
  205. def stop(self, containers=None):
  206. '''
  207. Attempts to stop a container or list of containers.
  208. If no containers are passed, stops all running containers.
  209. '''
  210. is_multi_setup = self.benchmarker.config.server_docker_host != \
  211. self.benchmarker.config.database_docker_host
  212. if containers:
  213. if not isinstance(containers, list):
  214. containers = [containers]
  215. for container in containers:
  216. DockerHelper.__stop_container(container)
  217. else:
  218. self.__stop_all(self.server)
  219. if is_multi_setup:
  220. self.__stop_all(self.database)
  221. self.database.containers.prune()
  222. if is_multi_setup:
  223. # Then we're on a 3 machine set up
  224. self.server.containers.prune()
  225. self.client.containers.prune()
  226. def build_databases(self):
  227. '''
  228. Builds all the databases necessary to run the list of benchmarker tests
  229. '''
  230. built = []
  231. for test in self.benchmarker.tests:
  232. db = test.database.lower()
  233. if db not in built and db != "none":
  234. image_name = "techempower/%s:latest" % db
  235. log_prefix = image_name + ": "
  236. database_dir = os.path.join(self.benchmarker.config.db_root,
  237. db)
  238. docker_file = "%s.dockerfile" % db
  239. self.__build(
  240. base_url=self.benchmarker.config.database_docker_host,
  241. path=database_dir,
  242. dockerfile=docker_file,
  243. log_prefix=log_prefix,
  244. build_log_file=os.devnull,
  245. tag="techempower/%s" % db)
  246. built.append(db)
  247. def start_database(self, database):
  248. '''
  249. Sets up a container for the given database and port, and starts said docker
  250. container.
  251. '''
  252. image_name = "techempower/%s:latest" % database
  253. log_prefix = image_name + ": "
  254. sysctl = {
  255. 'net.core.somaxconn': 65535,
  256. 'kernel.sem': "250 32000 256 512"
  257. }
  258. ulimit = [{'name': 'nofile', 'hard': 65535, 'soft': 65535}]
  259. container = self.database.containers.run(
  260. "techempower/%s" % database,
  261. name="tfb-database",
  262. network=self.benchmarker.config.network,
  263. network_mode=self.benchmarker.config.network_mode,
  264. detach=True,
  265. ulimits=ulimit,
  266. sysctls=sysctl,
  267. remove=True,
  268. log_config={'type': None})
  269. # Sleep until the database accepts connections
  270. slept = 0
  271. max_sleep = 60
  272. database_ready = False
  273. while not database_ready and slept < max_sleep:
  274. time.sleep(1)
  275. slept += 1
  276. database_ready = test_database(self.benchmarker.config, database)
  277. if not database_ready:
  278. log("Database was not ready after startup", prefix=log_prefix)
  279. return container
  280. def build_wrk(self):
  281. '''
  282. Builds the techempower/tfb.wrk container
  283. '''
  284. self.__build(
  285. base_url=self.benchmarker.config.client_docker_host,
  286. path=self.benchmarker.config.wrk_root,
  287. dockerfile="wrk.dockerfile",
  288. log_prefix="wrk: ",
  289. build_log_file=os.devnull,
  290. tag="techempower/tfb.wrk")
  291. def test_client_connection(self, url):
  292. '''
  293. Tests that the app server at the given url responds successfully to a
  294. request.
  295. '''
  296. try:
  297. self.client.containers.run(
  298. 'techempower/tfb.wrk',
  299. 'curl --fail --max-time 5 %s' % url,
  300. remove=True,
  301. log_config={'type': None},
  302. network=self.benchmarker.config.network,
  303. network_mode=self.benchmarker.config.network_mode)
  304. except Exception:
  305. return False
  306. return True
  307. def server_container_exists(self, container_id_or_name):
  308. '''
  309. Returns True if the container still exists on the server.
  310. '''
  311. try:
  312. self.server.containers.get(container_id_or_name)
  313. return True
  314. except:
  315. return False
  316. def benchmark(self, script, variables, raw_file):
  317. '''
  318. Runs the given remote_script on the wrk container on the client machine.
  319. '''
  320. def watch_container(container):
  321. with open(raw_file, 'w') as benchmark_file:
  322. for line in container.logs(stream=True):
  323. log(line, file=benchmark_file)
  324. sysctl = {'net.core.somaxconn': 65535}
  325. ulimit = [{'name': 'nofile', 'hard': 65535, 'soft': 65535}]
  326. watch_container(
  327. self.client.containers.run(
  328. "techempower/tfb.wrk",
  329. "/bin/bash /%s" % script,
  330. environment=variables,
  331. network=self.benchmarker.config.network,
  332. network_mode=self.benchmarker.config.network_mode,
  333. detach=True,
  334. stderr=True,
  335. ulimits=ulimit,
  336. sysctls=sysctl,
  337. remove=True,
  338. log_config={'type': None}))