How to sort by key and value in mapreduce?

Question

I have a text file:

10  1   15
10  12  30
10  9   45
10  8   40
10  15  55
12  9   0
12  7   18
12  10  1
9   1   1
9   2   1
9   0   1
14  5   5

And I would like to get this file as an output of my MapReduce job:

9   0   1   
9   1   1   
9   2   1   
10  1   15  
10  9   40  
10  9   45  
10  12  30  
10  15  55  
12  7   18  
12  9   0   
12  10  1   
14  5   5

It means it has to be sorted by 1st, 2nd and 3rd columns.

I use this command:

#!/bin/bash

IN_DIR="/user/cloudera/temp"
OUT_DIR="/user/cloudera/temp_out"
NUM_REDUCERS=1

hdfs dfs -rmr ${OUT_DIR} > /dev/null

hadoop jar /usr/lib/hadoop-mapreduce/hadoop-streaming.jar \
-D mapred.jab.name="Parsing mista pages job 1 (parsing)" \
-D stream.num.map.output.key.fields=3 \
-D mapreduce.job.output.key.comparator.class=org.apache.hadoop.mapreduce.lib.partition.KeyFieldBasedComparator \
-D mapreduce.partition.keycomparator.options='-k1,1n -k2,2n -k3,3n' \
-D mapreduce.job.reduces=${NUM_REDUCERS} \
-mapper 'cat' \
-reducer 'cat' \
-input ${IN_DIR} \
-output ${OUT_DIR}

hdfs dfs -cat ${OUT_DIR}/* | head -100

And get exactly what I want. BUT. When I do NUM_REDUCERS=2 I get this output:

[cloudera@quickstart ~]$ hdfs dfs -cat /user/cloudera/temp_out/part-00000 | head -100
9   1   1   
10  9   45  
10  12  30  
10  15  55  
12  7   18  
12  10  1   
14  5   5

[cloudera@quickstart ~]$ hdfs dfs -cat /user/cloudera/temp_out/part-00001 | head -100
9   0   1   
9   2   1   
10  1   15  
10  9   40  
12  9   0

Why partitioner splits my data with same keys (for example '9') to different reducers?

How can I force partitioner to split Mapper output by the key and sort it by value. For example, if I have 4 reducers the reducers input should be:

reducer 1
9   0   1   
9   1   1   
9   2   1   

reducer 2
10  1   15  
10  9   40  
10  9   45  
10  12  30  
10  15  55  

reducer 3
12  7   18  
12  9   0   
12  10  1

reducer 4:   
14  5   5

Answer 1

you can overwrite the default Partioner to put each key into diferent reduce .Set the same Nums of reduce . let each reduce to deal with only one key .

for example()

groupMap.put("9", 0);
groupMap.put("10", 1);
groupMap.put("12", 2);
groupMap.put("14", 3);

Add -partitioner argument to use your own partition in your job. I think it might works for you