Example #1
0
def main():
  parser = OptionParser()
  parser.add_option("-v", "--verbose", dest="verbose", help="verbose log output", default=False)
  parser.add_option("-l", "--label", dest="label", help="label of the agent", default=None)
  parser.add_option("--zk-quorum", dest=Constants.ZK_QUORUM, help="Zookeeper Quorum", default=None)
  parser.add_option("--zk-reg-path", dest=Constants.ZK_REG_PATH, help="Zookeeper Registry Path", default=None)
  parser.add_option("--debug", dest="debug", help="Agent debug hint", default="")
  (options, args) = parser.parse_args()

  if not Constants.AGENT_WORK_ROOT in os.environ and not 'PWD' in os.environ:
    parser.error("AGENT_WORK_ROOT environment variable or PWD must be set.")
  if Constants.AGENT_WORK_ROOT in os.environ:
    options.root_folder = os.environ[Constants.AGENT_WORK_ROOT]
  else:
    # some launch environments do not end up setting all environment variables
    options.root_folder = os.environ['PWD']

  if not 'AGENT_LOG_ROOT' in os.environ:
    parser.error("AGENT_LOG_ROOT environment variable must be set.")
  options.log_folder = os.environ['AGENT_LOG_ROOT']
  all_log_folders = [x.strip() for x in options.log_folder.split(',')]
  if len(all_log_folders) > 1:
    options.log_folder = all_log_folders[0]

  # If there are multiple log folder, separate by comma, pick one

  if not options.label:
    parser.error("label is required.");

  if not IS_WINDOWS:
    bind_signal_handlers()

  # Check for configuration file.
  agentConfig = AgentConfig(options.root_folder, options.log_folder, options.label)
  update_config_from_file(agentConfig)

  # update configurations if needed
  if options.zk_quorum:
      agentConfig.set(AgentConfig.SERVER_SECTION, Constants.ZK_QUORUM, options.zk_quorum)

  if options.zk_reg_path:
      agentConfig.set(AgentConfig.SERVER_SECTION, Constants.ZK_REG_PATH, options.zk_reg_path)

  if options.debug:
    agentConfig.set(AgentConfig.AGENT_SECTION, AgentConfig.APP_DBG_CMD, options.debug)

  logFile = posixpath.join(agentConfig.getResolvedPath(AgentConfig.LOG_DIR), logFileName)
  setup_logging(options.verbose, logFile)
  update_log_level(agentConfig, logFile)

  secDir = posixpath.join(agentConfig.getResolvedPath(AgentConfig.RUN_DIR), "security")
  logger.info("Security/Keys directory: " + secDir)
  agentConfig.set(AgentConfig.SECURITY_SECTION, "keysdir", secDir)

  perform_prestart_checks(agentConfig)
  ensure_folder_layout(agentConfig)
  # create security dir if necessary
  ensure_path_exists(secDir)

  write_pid()

  logger.info("Using AGENT_WORK_ROOT = " + options.root_folder)
  logger.info("Using AGENT_LOG_ROOT = " + options.log_folder)

  if len(all_log_folders) > 1:
    logger.info("Selected log folder from available: " + ",".join(all_log_folders))

  # Extract the AM hostname and secured port from ZK registry
  zk_lookup_tries = 0
  while zk_lookup_tries < Constants.MAX_AM_CONNECT_RETRIES:
    registry = Registry(options.zk_quorum, options.zk_reg_path)
    amHost, amUnsecuredPort, amSecuredPort = registry.readAMHostPort()

    tryConnect = True
    if not amHost or not amSecuredPort or not amUnsecuredPort:
      logger.info("Unable to extract AM host details from ZK, retrying ...")
      tryConnect = False
      time.sleep(NetUtil.CONNECT_SERVER_RETRY_INTERVAL_SEC)

    if tryConnect:
      if amHost:
        agentConfig.set(AgentConfig.SERVER_SECTION, "hostname", amHost)

      if amSecuredPort:
        agentConfig.set(AgentConfig.SERVER_SECTION, "secured_port", amSecuredPort)

      if amUnsecuredPort:
        agentConfig.set(AgentConfig.SERVER_SECTION, "port", amUnsecuredPort)

      server_url = SERVER_STATUS_URL.format(
        agentConfig.get(AgentConfig.SERVER_SECTION, 'hostname'),
        agentConfig.get(AgentConfig.SERVER_SECTION, 'port'),
        agentConfig.get(AgentConfig.SERVER_SECTION, 'check_path'))
      print("Connecting to the server at " + server_url + "...")
      logger.info('Connecting to the server at: ' + server_url)

      # Wait until server is reachable and continue to query ZK
      netutil = NetUtil()
      retries = netutil.try_to_connect(server_url, 3, logger)
      if retries < 3:
        break;
      pass
    pass
    zk_lookup_tries += 1
  pass

  # Launch Controller communication
  global controller
  controller = Controller(agentConfig)
  controller.start()
  
  try:
    while controller.is_alive():
      controller.join(timeout=1.0)
  except (KeyboardInterrupt, SystemExit):
    logger.info("... agent interrupted")
    pass