python - 使用 --pool-emr-job-flows 时，MRJob 无法在 EMR 上启动新作业

Question

我正在使用 MRJob 在 Amazon 的 EMR 上运行一个迭代的 hadoop 程序。

当我不使用“--pool-emr-job-flows”选项时，一切正常(但速度很慢)。当我使用这个选项时，

Traceback (most recent call last):
  File "ic_bfs_eval.py", line 297, in <module>
    res = main()
  File "ic_bfs_eval.py", line 262, in main
    frac, mr_rounds = bfs(db_name, T, samples, total_steps_cap)
  File "ic_bfs_eval.py", line 183, in bfs
    runner.run()
  File "/Library/Python/2.7/site-packages/mrjob-0.4.3_dev-py2.7.egg/mrjob/runner.py", line 620, in __exit__
    self.cleanup()
  File "/Library/Python/2.7/site-packages/mrjob-0.4.3_dev-py2.7.egg/mrjob/emr.py", line 987, in cleanup
    super(EMRJobRunner, self).cleanup(mode=mode)
  File "/Library/Python/2.7/site-packages/mrjob-0.4.3_dev-py2.7.egg/mrjob/runner.py", line 566, in cleanup
    self._cleanup_job()
  File "/Library/Python/2.7/site-packages/mrjob-0.4.3_dev-py2.7.egg/mrjob/emr.py", line 1061, in _cleanup_job
    self._opts['ec2_key_pair_file'])
  File "/Library/Python/2.7/site-packages/mrjob-0.4.3_dev-py2.7.egg/mrjob/ssh.py", line 209, in ssh_terminate_single_job
    num_jobs_match = HADOOP_JOB_LIST_NUM_RE.match(job_list_lines[0])
IndexError: list index out of range

我正在像这样初始化一个 MRJob:

mrJob2 = MRBFSSampleIter(args=["-c", "~/mrjob.conf",
                                       "-r", "emr",
                                       "--no-output",
                                       "--output-dir", tmp_dir_out,
                                       "--pool-emr-job-flows", tmp_dir_in])

关于为什么会发生这种情况有什么想法吗？

Answer 1

当我设置 ssh key 对时，这对我来说就消失了。我认为它仍然是一个错误，因为 ssh 应该是可选的。但最简单的解决方法是按照 http://mrjob.readthedocs.org/en/latest/guides/emr-quickstart.html#configuring-ssh-credentials 中所述设置 key 对。