I have a map reduce program running to read the HDFS file as below:
hadoop jar /opt/mapr/hadoop/hadoop-0.20.2/contrib/streaming/hadoop-0.20.2-dev-streaming.jar -Dmapred.reduce.tasks=1000 -file $homedir/mapper.py -mapper $homedir/mapper.py -file $homedir/reducer.py -reducer $homedir/reducer.py -input /user/data/* -output /output/ 2> output.text
Anything to be confirm, the path /user/data/* has folders including files, /user/data/* will iterate all files under all subfolders right ?
The hdfs text file contains a JSON string for each line so the mapper read the file as below:
for line in sys.stdin:
try:
object = json.loads(line)
But the owner of the HDFS changed the file from text into sequencefile. and I found the map reduce program output a lot of zero sized files, which probably means it did not successfully read the file from HDFS.
What should I change to code so that I can read from the sequencefile ? I also have a HIVE external table to perform the aggregation and sorting based on that output of mapreduce, and the HIVE was STORED AS TEXTFILE before, should I change to STORED AS SEQUENCEFILE ?
Thanks,
Have a look at this
Run below python file before your mapreduce job
input : your sequence file
output : your input to mapreduce
import sys
from hadoop.io import SequenceFile
if __name__ == '__main__':
if len(sys.argv) < 3:
print 'usage: SequenceFileReader <filename> <output>'
else:
reader = SequenceFile.Reader(sys.argv[1])
key_class = reader.getKeyClass()
value_class = reader.getValueClass()
key = key_class()
value = value_class()
#reader.sync(4042)
position = reader.getPosition()
f = open(sys.argv[2],'w')
while reader.next(key, value):
f.write(value.toString()+'\n')
reader.close()
f.close()
You wont have to change you original python file now.