RandomWriter example writes 10 gig (by default) of random data/host to DFS using Map/Reduce.

Each map takes a single file name as input and writes random BytesWritable keys and values to the DFS sequence file. The maps do not emit any output and the reduce phase is not used.

The specifics of the generated data are configurable. The configuration variables are:

Name

Default Value

Description

test.randomwriter.maps_per_host

10

Number of maps/host

test.randomwrite.bytes_per_map

1073741824

Number of bytes written/map

test.randomwrite.min_key

10

minimum size of the key in bytes

test.randomwrite.max_key

1000

maximum size of the key in bytes

test.randomwrite.min_value

0

minimum size of the value

test.randomwrite.max_value

20000

maximum size of the value

This example uses a useful pattern for dealing with Hadoop's constraints on InputSplits. Since each input split can only consist of a file and byte range and we want to control how many maps there are (and we don't really have any inputs), we create a directory with a set of artificial files, each of which contains the filename that we want a given map to write to. Then, using the text line reader and this "fake" input directory, we generate exactly the right number of maps. Each map gets a single record that is the filename, to which it is supposed to write its output.

To run the example, the command syntax is
bin/hadoop jar hadoop-*-examples.jar randomwriter <out-dir> [<configuration file>]

RandomWriter supports generic options : see DevelopmentCommandLineOptions

RandomWriter (last edited 2009-09-20 23:55:03 by localhost)