def test_failed_job(self):
        mr_job = MRTwoStepJob(['-r', 'dataproc', '-v'])
        mr_job.sandbox()

        with no_handlers_for_logger('mrjob.dataproc'):
            stderr = StringIO()
            log_to_stream('mrjob.dataproc', stderr)

            self._dataproc_client.job_get_advances_states = (
                collections.deque(['SETUP_DONE', 'RUNNING', 'ERROR']))

            with mr_job.make_runner() as runner:
                self.assertIsInstance(runner, DataprocJobRunner)

                self.assertRaises(StepFailedException, runner.run)

                self.assertIn(' => ERROR\n', stderr.getvalue())

                cluster_id = runner.get_cluster_id()

        # job should get terminated
        cluster = (
            self._dataproc_client._cache_clusters[_TEST_PROJECT][cluster_id])
        cluster_state = self._dataproc_client.get_state(cluster)
        self.assertEqual(cluster_state, 'DELETING')

    def test_non_log_lines(self):
        lines = StringIO('foo\n'
                         'bar\n'
                         '15/12/11 13:26:08 ERROR streaming.StreamJob:'
                         ' Error Launching job :'
                         ' Output directory already exists\n'
                         'Streaming Command Failed!')

        with no_handlers_for_logger('mrjob.logs.parse'):
            stderr = StringIO()
            log_to_stream('mrjob.logs.parse', stderr)

            self.assertEqual(
            list(_parse_hadoop_log_lines(lines)), [
                # ignore leading non-log lines
                dict(
                    timestamp='15/12/11 13:26:08',
                    level='ERROR',
                    logger='streaming.StreamJob',
                    thread=None,
                    # no way to know that Streaming Command Failed! wasn't part
                    # of a multi-line message
                    message=('Error Launching job :'
                             ' Output directory already exists\n'
                             'Streaming Command Failed!'))
            ])

            # should be one warning for each leading non-log line
            log_lines = stderr.getvalue().splitlines()
            self.assertEqual(len(log_lines), 2)

    def test_cleanup_options(self):
        stderr = StringIO()
        with no_handlers_for_logger('mrjob.runner'):
            log_to_stream('mrjob.runner', stderr)
            opts = RunnerOptionStore(
                'inline',
                dict(cleanup=['LOCAL_SCRATCH', 'REMOTE_SCRATCH'],
                     cleanup_on_failure=['JOB_FLOW', 'SCRATCH']),
                [])

            self.assertEqual(opts['cleanup'], ['LOCAL_TMP', 'CLOUD_TMP'])
            self.assertIn(
                'Deprecated cleanup option LOCAL_SCRATCH has been renamed'
                ' to LOCAL_TMP', stderr.getvalue())
            self.assertIn(
                'Deprecated cleanup option REMOTE_SCRATCH has been renamed'
                ' to CLOUD_TMP', stderr.getvalue())

            self.assertEqual(opts['cleanup_on_failure'], ['CLUSTER', 'TMP'])
            self.assertIn(
                'Deprecated cleanup_on_failure option JOB_FLOW has been'
                ' renamed to CLUSTER', stderr.getvalue())
            self.assertIn(
                'Deprecated cleanup_on_failure option SCRATCH has been renamed'
                ' to TMP', stderr.getvalue())

    def test_non_log_lines(self):
        lines = StringIO(
            "foo\n"
            "bar\n"
            "15/12/11 13:26:08 ERROR streaming.StreamJob:"
            " Error Launching job :"
            " Output directory already exists\n"
            "Streaming Command Failed!"
        )

        with no_handlers_for_logger("mrjob.logs.parse"):
            stderr = StringIO()
            log_to_stream("mrjob.logs.parse", stderr)

            self.assertEqual(
                list(_parse_hadoop_log_lines(lines)),
                [
                    # ignore leading non-log lines
                    dict(
                        timestamp="15/12/11 13:26:08",
                        level="ERROR",
                        logger="streaming.StreamJob",
                        thread=None,
                        # no way to know that Streaming Command Failed! wasn't part
                        # of a multi-line message
                        message=(
                            "Error Launching job :" " Output directory already exists\n" "Streaming Command Failed!"
                        ),
                    )
                ],
            )

            # should be one warning for each leading non-log line
            log_lines = stderr.getvalue().splitlines()
            self.assertEqual(len(log_lines), 2)

 def assert_hadoop_version(self, JobClass, version_string):
     mr_job = JobClass()
     mock_log = StringIO()
     with no_handlers_for_logger("mrjob.job"):
         log_to_stream("mrjob.job", mock_log)
         self.assertEqual(mr_job.jobconf()["hadoop_version"], version_string)
         self.assertIn("should be a string", mock_log.getvalue())

    def updated_and_warnings(self, jobconf, hadoop_version):
        jobconf = jobconf.copy()
        with no_handlers_for_logger("mrjob.runner"):
            stderr = StringIO()
            log_to_stream("mrjob.runner", stderr)
            self.runner._update_jobconf_for_hadoop_version(jobconf, hadoop_version)

        return jobconf, stderr.getvalue()

 def test_messy_error(self):
     counter_string = b'Job JOBID="_001" FAILED_REDUCES="0" COUNTERS="THIS IS NOT ACTUALLY A COUNTER"'
     with no_handlers_for_logger(''):
         stderr = StringIO()
         log_to_stream('mrjob.parse', stderr, level=logging.WARN)
         self.assertEqual(({}, 1),
                          parse_hadoop_counters_from_line(counter_string))
         self.assertIn('Cannot parse Hadoop counter string',
                       stderr.getvalue())

    def test_option_debug_printout(self):
        stderr = StringIO()

        with no_handlers_for_logger():
            log_to_stream('mrjob.runner', stderr, debug=True)

            InlineMRJobRunner(owner='dave')

        self.assertIn("'owner'", stderr.getvalue())
        self.assertIn("'dave'", stderr.getvalue())

    def get_debug_printout(self, opt_store_class, alias, opts):
        stderr = StringIO()

        with no_handlers_for_logger():
            log_to_stream('mrjob.runner', stderr, debug=True)

            # debug printout happens in constructor
            opt_store_class(alias, opts, [])

        return stderr.getvalue()

    def test_empty_runner_error(self):
        conf = dict(runner=dict(local=dict(local_tmp_dir='/tmp')))
        path = self.save_conf('basic', conf)

        stderr = StringIO()
        with no_handlers_for_logger():
            log_to_stream('mrjob.runner', stderr)
            RunnerOptionStore('inline', {}, [path])
            self.assertEqual(
                "No configs specified for inline runner\n",
                stderr.getvalue())

    def test_runner_option_store(self):
        stderr = StringIO()
        with no_handlers_for_logger('mrjob.conf'):
            log_to_stream('mrjob.conf', stderr)
            opts = RunnerOptionStore(
                'inline', dict(base_tmp_dir='/scratch'), [])

            self.assertEqual(opts['local_tmp_dir'], '/scratch')
            self.assertNotIn('base_tmp_dir', opts)
            self.assertIn('Deprecated option base_tmp_dir has been renamed'
                          ' to local_tmp_dir', stderr.getvalue())

    def test_indentation_is_required(self):
        lines = ["File System Counters", "   FILE: Number of bytes read=8"]

        with no_handlers_for_logger("mrjob.logs.parse"):
            stderr = StringIO()
            log_to_stream("mrjob.logs.parse", stderr)

            # counter line is interpreted as group
            self.assertEqual(_parse_indented_counters(lines), {})

            # should complain
            self.assertNotEqual(stderr.getvalue(), "")

    def test_recurse(self):
        path = os.path.join(self.tmp_dir, 'LOL.conf')
        recurse_conf = dict(include=path)
        with open(path, 'w') as f:
            dump_mrjob_conf(recurse_conf, f)

        stderr = StringIO()
        with no_handlers_for_logger():
            log_to_stream('mrjob.conf', stderr)
            RunnerOptionStore('inline', {}, [path])
            self.assertIn('%s tries to recursively include %s!' % (path, path),
                          stderr.getvalue())

 def test_attrs_should_be_classes(self):
     with no_handlers_for_logger('mrjob.job'):
         stderr = StringIO()
         log_to_stream('mrjob.job', stderr)
         job = self.StrangeJob()
         self.assertIsInstance(job.input_protocol(), JSONProtocol)
         self.assertIsInstance(job.internal_protocol(), JSONProtocol)
         self.assertIsInstance(job.output_protocol(), JSONProtocol)
         logs = stderr.getvalue()
         self.assertIn('INPUT_PROTOCOL should be a class', logs)
         self.assertIn('INTERNAL_PROTOCOL should be a class', logs)
         self.assertIn('OUTPUT_PROTOCOL should be a class', logs)

    def test_with_header(self):
        lines = ["Counters: 1", "  File System Counters", "    FILE: Number of bytes read=86"]

        with no_handlers_for_logger("mrjob.logs.parse"):
            stderr = StringIO()
            log_to_stream("mrjob.logs.parse", stderr)

            self.assertEqual(
                _parse_indented_counters(lines), {"File System Counters": {"FILE: Number of bytes read": 86}}
            )

            # header shouldn't freak it out
            self.assertEqual(stderr.getvalue(), "")

    def test_passthrough(self):
        runner = InlineMRJobRunner()

        with no_handlers_for_logger("mrjob.runner"):
            stderr = StringIO()
            log_to_stream("mrjob.runner", stderr)

            self.assertEqual(runner.ls, runner.fs.ls)
            # no special rules for underscore methods
            self.assertEqual(runner._cat_file, runner.fs._cat_file)

            self.assertIn("deprecated: call InlineMRJobRunner.fs.ls() directly", stderr.getvalue())
            self.assertIn("deprecated: call InlineMRJobRunner.fs._cat_file() directly", stderr.getvalue())

    def test_deprecated_alias(self):
        with no_handlers_for_logger('mrjob.util'):
            stderr = StringIO()
            log_to_stream('mrjob.util', stderr)

            self.assertEqual(
                list(buffer_iterator_to_line_iterator(
                    chunk for chunk in
                    [b'The quick\nbrown fox\njumped over\nthe lazy\ndogs.\n'])
                ),
                [b'The quick\n', b'brown fox\n', b'jumped over\n',
                 b'the lazy\n', b'dogs.\n'])

            self.assertIn('has been renamed', stderr.getvalue())

    def test_io_error(self):
        self.mock_paths = [
            IOError(),
        ]

        with no_handlers_for_logger('mrjob.logs.ls'):
            stderr = StringIO()
            log_to_stream('mrjob.logs.ls', stderr)

            self.assertEqual(list(_ls_logs(self.mock_fs, '/path/to/logs')), [])

            self.mock_fs.ls.assert_called_once_with('/path/to/logs')

            self.assertIn("couldn't ls() /path/to/logs", stderr.getvalue())

    def test_dry_run(self):
        stdout = StringIO()
        self.maybe_terminate_quietly(
            stdout=stdout, max_hours_idle=0.01, dry_run=True)

        # dry_run doesn't actually try to lock
        expected_stdout_lines = self.EXPECTED_STDOUT_LINES + [
            'Terminated job flow j-IDLE_AND_LOCKED (IDLE_AND_LOCKED);'
            ' was idle for 2:00:00, 1:00:00 to end of hour']

        self.assertEqual(set(stdout.getvalue().splitlines()),
                         set(expected_stdout_lines))

        # shouldn't *actually* terminate clusters
        self.assertEqual(self.ids_of_terminated_clusters(), [])

    def test_prefer_own_methods(self):
        # TODO: currently can't initialize HadoopRunner without setting these
        runner = HadoopJobRunner(hadoop_bin="hadoop", hadoop_home="kansas", hadoop_streaming_jar="streaming.jar")

        with no_handlers_for_logger("mrjob.runner"):
            stderr = StringIO()
            log_to_stream("mrjob.runner", stderr)

            self.assertEqual(runner.ls, runner.fs.ls)

            # Hadoop Runner has its own version
            self.assertNotEqual(runner.get_hadoop_version, runner.fs.get_hadoop_version)

            self.assertIn("deprecated: call HadoopJobRunner.fs.ls() directly", stderr.getvalue())
            self.assertNotIn("get_hadoop_version", stderr.getvalue())