How to create a hadoop runner?

Question

I have the following simple mrjob script, which reads a large file line by line, performs an operation on each line and prints the output:

#!/usr/bin/env python                                                                                                           

from mrjob.job import MRJob

class LineProcessor(MRJob):
    def mapper(self, _, line):
        yield (line.upper(), None) # toy example: mapper just uppercase the line

if __name__ == '__main__':
    # mr_job = LineProcessor(args=['-r', 'hadoop', '/path/to/input']) # error!
    mr_job = LineProcessor(args=['/path/to/input'])
    with mr_job.make_runner() as runner:
        runner.run() 
        for line in runner.stream_output():
            key, value = mr_job.parse_output_line(line)
            print key.encode('utf-8')  # don't care about value in my case

(This is just a toy example; processing each line is expensive in my real case, which is why I want to run distributed.)

It works as a local process only. If I try to use '-r', 'hadoop' (see commented out above) I get the following strange error:

  File "mrjob/runner.py", line 727, in _get_steps
    'error getting step information: %s', stderr)
Exception: ('error getting step information: %s', 'Traceback (most recent call last):\n  File "script.py", line 11, in <module>\n    with mr_job.make_runner() as runner:\n  File "mrjob/job.py", line 515, in make_runner\n    " __main__, which doesn\'t work." % w)\nmrjob.job.UsageError: make_runner() was called with --steps. This probably means you tried to use it from __main__, which doesn\'t work.\n')

How can I actually run it on hadoop, i.e., create a HadoopJobRunner?

Answer 1

Are you missing

def steps(self):
        return [self.mr(
                          mapper_init = ...
                          mapper = self.mapper,
                          combiner = ...,
                          reducer = ...,
                  )]

in your LineProcessor ?