docker_helper.py 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394
  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. from psutil import virtual_memory
  13. # total memory limit allocated for the test container
  14. mem_limit = int(round(virtual_memory().total * .95))
  15. class DockerHelper:
  16. def __init__(self, benchmarker=None):
  17. self.benchmarker = benchmarker
  18. self.client = docker.DockerClient(
  19. base_url=self.benchmarker.config.client_docker_host)
  20. self.server = docker.DockerClient(
  21. base_url=self.benchmarker.config.server_docker_host)
  22. self.database = docker.DockerClient(
  23. base_url=self.benchmarker.config.database_docker_host)
  24. def __build(self, base_url, path, build_log_file, log_prefix, dockerfile,
  25. tag):
  26. '''
  27. Builds docker containers using docker-py low-level api
  28. '''
  29. self.benchmarker.time_logger.mark_build_start()
  30. with open(build_log_file, 'w') as build_log:
  31. try:
  32. client = docker.APIClient(base_url=base_url)
  33. output = client.build(
  34. path=path,
  35. dockerfile=dockerfile,
  36. tag=tag,
  37. forcerm=True,
  38. timeout=3600,
  39. pull=True,
  40. buildargs=({
  41. 'BENCHMARK_ENV': self.benchmarker.config.results_environment
  42. })
  43. )
  44. buffer = ""
  45. for token in output:
  46. if token.startswith('{"stream":'):
  47. token = json.loads(token)
  48. token = token[token.keys()[0]].encode('utf-8')
  49. buffer += token
  50. elif token.startswith('{"errorDetail":'):
  51. token = json.loads(token)
  52. raise Exception(token['errorDetail']['message'])
  53. while "\n" in buffer:
  54. index = buffer.index("\n")
  55. line = buffer[:index]
  56. buffer = buffer[index + 1:]
  57. log(line,
  58. prefix=log_prefix,
  59. file=build_log,
  60. color=Fore.WHITE + Style.BRIGHT \
  61. if re.match(r'^Step \d+\/\d+', line) else '')
  62. if buffer:
  63. log(buffer,
  64. prefix=log_prefix,
  65. file=build_log,
  66. color=Fore.WHITE + Style.BRIGHT \
  67. if re.match(r'^Step \d+\/\d+', buffer) else '')
  68. except Exception:
  69. tb = traceback.format_exc()
  70. log("Docker build failed; terminating",
  71. prefix=log_prefix,
  72. file=build_log,
  73. color=Fore.RED)
  74. log(tb, prefix=log_prefix, file=build_log)
  75. self.benchmarker.time_logger.log_build_end(
  76. log_prefix=log_prefix, file=build_log)
  77. raise
  78. self.benchmarker.time_logger.log_build_end(
  79. log_prefix=log_prefix, file=build_log)
  80. def clean(self):
  81. '''
  82. Cleans all the docker images from the system
  83. '''
  84. self.server.images.prune()
  85. for image in self.server.images.list():
  86. if len(image.tags) > 0:
  87. # 'techempower/tfb.test.gemini:0.1' -> 'techempower/tfb.test.gemini'
  88. image_tag = image.tags[0].split(':')[0]
  89. if image_tag != 'techempower/tfb' and 'techempower' in image_tag:
  90. self.server.images.remove(image.id, force=True)
  91. self.server.images.prune()
  92. self.database.images.prune()
  93. for image in self.database.images.list():
  94. if len(image.tags) > 0:
  95. # 'techempower/tfb.test.gemini:0.1' -> 'techempower/tfb.test.gemini'
  96. image_tag = image.tags[0].split(':')[0]
  97. if image_tag != 'techempower/tfb' and 'techempower' in image_tag:
  98. self.database.images.remove(image.id, force=True)
  99. self.database.images.prune()
  100. def build(self, test, build_log_dir=os.devnull):
  101. '''
  102. Builds the test docker containers
  103. '''
  104. log_prefix = "%s: " % test.name
  105. # Build the test image
  106. test_docker_file = "%s.dockerfile" % test.name
  107. build_log_file = build_log_dir
  108. if build_log_dir is not os.devnull:
  109. build_log_file = os.path.join(
  110. build_log_dir,
  111. "%s.log" % test_docker_file.replace(".dockerfile", "").lower())
  112. try:
  113. self.__build(
  114. base_url=self.benchmarker.config.server_docker_host,
  115. build_log_file=build_log_file,
  116. log_prefix=log_prefix,
  117. path=test.directory,
  118. dockerfile=test_docker_file,
  119. tag="techempower/tfb.test.%s" % test_docker_file.replace(
  120. ".dockerfile", ""))
  121. except Exception:
  122. return 1
  123. return 0
  124. def run(self, test, run_log_dir):
  125. '''
  126. Run the given Docker container(s)
  127. '''
  128. log_prefix = "%s: " % test.name
  129. container = None
  130. try:
  131. def watch_container(docker_container, docker_file):
  132. with open(
  133. os.path.join(
  134. run_log_dir, "%s.log" % docker_file.replace(
  135. ".dockerfile", "").lower()), 'w') as run_log:
  136. for line in docker_container.logs(stream=True):
  137. log(line, prefix=log_prefix, file=run_log)
  138. extra_hosts = None
  139. name = "tfb-server"
  140. if self.benchmarker.config.network is None:
  141. extra_hosts = {
  142. socket.gethostname():
  143. str(self.benchmarker.config.server_host),
  144. 'tfb-server':
  145. str(self.benchmarker.config.server_host),
  146. 'tfb-database':
  147. str(self.benchmarker.config.database_host)
  148. }
  149. name = None
  150. sysctl = {'net.core.somaxconn': 65535}
  151. ulimit = [{
  152. 'name': 'nofile',
  153. 'hard': 200000,
  154. 'soft': 200000
  155. }, {
  156. 'name': 'rtprio',
  157. 'hard': 99,
  158. 'soft': 99
  159. }]
  160. container = self.server.containers.run(
  161. "techempower/tfb.test.%s" % test.name,
  162. name=name,
  163. network=self.benchmarker.config.network,
  164. network_mode=self.benchmarker.config.network_mode,
  165. stderr=True,
  166. detach=True,
  167. init=True,
  168. extra_hosts=extra_hosts,
  169. privileged=True,
  170. ulimits=ulimit,
  171. mem_limit=mem_limit,
  172. sysctls=sysctl,
  173. remove=True,
  174. log_config={'type': None})
  175. watch_thread = Thread(
  176. target=watch_container,
  177. args=(
  178. container,
  179. "%s.dockerfile" % test.name,
  180. ))
  181. watch_thread.daemon = True
  182. watch_thread.start()
  183. except Exception:
  184. with open(
  185. os.path.join(run_log_dir, "%s.log" % test.name.lower()),
  186. 'w') as run_log:
  187. tb = traceback.format_exc()
  188. log("Running docker container: %s.dockerfile failed" %
  189. test.name,
  190. prefix=log_prefix,
  191. file=run_log)
  192. log(tb, prefix=log_prefix, file=run_log)
  193. return container
  194. @staticmethod
  195. def __stop_container(container):
  196. try:
  197. container.kill()
  198. time.sleep(2)
  199. except:
  200. # container has already been killed
  201. pass
  202. @staticmethod
  203. def __stop_all(docker_client):
  204. for container in docker_client.containers.list():
  205. if len(container.image.tags) > 0 \
  206. and 'techempower' in container.image.tags[0] \
  207. and 'tfb:latest' not in container.image.tags[0]:
  208. DockerHelper.__stop_container(container)
  209. def stop(self, containers=None):
  210. '''
  211. Attempts to stop a container or list of containers.
  212. If no containers are passed, stops all running containers.
  213. '''
  214. is_multi_setup = self.benchmarker.config.server_docker_host != \
  215. self.benchmarker.config.database_docker_host
  216. if containers:
  217. if not isinstance(containers, list):
  218. containers = [containers]
  219. for container in containers:
  220. DockerHelper.__stop_container(container)
  221. else:
  222. self.__stop_all(self.server)
  223. if is_multi_setup:
  224. self.__stop_all(self.database)
  225. self.database.containers.prune()
  226. if is_multi_setup:
  227. # Then we're on a 3 machine set up
  228. self.server.containers.prune()
  229. self.client.containers.prune()
  230. def build_databases(self):
  231. '''
  232. Builds all the databases necessary to run the list of benchmarker tests
  233. '''
  234. built = []
  235. for test in self.benchmarker.tests:
  236. db = test.database.lower()
  237. if db not in built and db != "none":
  238. image_name = "techempower/%s:latest" % db
  239. log_prefix = image_name + ": "
  240. database_dir = os.path.join(self.benchmarker.config.db_root,
  241. db)
  242. docker_file = "%s.dockerfile" % db
  243. self.__build(
  244. base_url=self.benchmarker.config.database_docker_host,
  245. path=database_dir,
  246. dockerfile=docker_file,
  247. log_prefix=log_prefix,
  248. build_log_file=os.devnull,
  249. tag="techempower/%s" % db)
  250. built.append(db)
  251. def start_database(self, database):
  252. '''
  253. Sets up a container for the given database and port, and starts said docker
  254. container.
  255. '''
  256. image_name = "techempower/%s:latest" % database
  257. log_prefix = image_name + ": "
  258. sysctl = {
  259. 'net.core.somaxconn': 65535,
  260. 'kernel.sem': "250 32000 256 512"
  261. }
  262. ulimit = [{'name': 'nofile', 'hard': 65535, 'soft': 65535}]
  263. container = self.database.containers.run(
  264. "techempower/%s" % database,
  265. name="tfb-database",
  266. network=self.benchmarker.config.network,
  267. network_mode=self.benchmarker.config.network_mode,
  268. detach=True,
  269. ulimits=ulimit,
  270. sysctls=sysctl,
  271. remove=True,
  272. log_config={'type': None})
  273. # Sleep until the database accepts connections
  274. slept = 0
  275. max_sleep = 60
  276. database_ready = False
  277. while not database_ready and slept < max_sleep:
  278. time.sleep(1)
  279. slept += 1
  280. database_ready = test_database(self.benchmarker.config, database)
  281. if not database_ready:
  282. log("Database was not ready after startup", prefix=log_prefix)
  283. return container
  284. def build_wrk(self):
  285. '''
  286. Builds the techempower/tfb.wrk container
  287. '''
  288. self.__build(
  289. base_url=self.benchmarker.config.client_docker_host,
  290. path=self.benchmarker.config.wrk_root,
  291. dockerfile="wrk.dockerfile",
  292. log_prefix="wrk: ",
  293. build_log_file=os.devnull,
  294. tag="techempower/tfb.wrk")
  295. def test_client_connection(self, url):
  296. '''
  297. Tests that the app server at the given url responds successfully to a
  298. request.
  299. '''
  300. try:
  301. self.client.containers.run(
  302. 'techempower/tfb.wrk',
  303. 'curl --fail --max-time 5 %s' % url,
  304. remove=True,
  305. log_config={'type': None},
  306. network=self.benchmarker.config.network,
  307. network_mode=self.benchmarker.config.network_mode)
  308. except Exception:
  309. return False
  310. return True
  311. def server_container_exists(self, container_id_or_name):
  312. '''
  313. Returns True if the container still exists on the server.
  314. '''
  315. try:
  316. self.server.containers.get(container_id_or_name)
  317. return True
  318. except:
  319. return False
  320. def benchmark(self, script, variables, raw_file):
  321. '''
  322. Runs the given remote_script on the wrk container on the client machine.
  323. '''
  324. def watch_container(container):
  325. with open(raw_file, 'w') as benchmark_file:
  326. for line in container.logs(stream=True):
  327. log(line, file=benchmark_file)
  328. sysctl = {'net.core.somaxconn': 65535}
  329. ulimit = [{'name': 'nofile', 'hard': 65535, 'soft': 65535}]
  330. watch_container(
  331. self.client.containers.run(
  332. "techempower/tfb.wrk",
  333. "/bin/bash /%s" % script,
  334. environment=variables,
  335. network=self.benchmarker.config.network,
  336. network_mode=self.benchmarker.config.network_mode,
  337. detach=True,
  338. stderr=True,
  339. ulimits=ulimit,
  340. sysctls=sysctl,
  341. remove=True,
  342. log_config={'type': None}))