Python scripts for avro conversion using Hadoop Streaming

I

have a 10 GB input file and I’m trying to convert it to avro using a python hadoop stream, it works successfully but I can’t read the output using the avro reader.

It gives the “utf8” codec unable to decode byte 0xb4 in position 13924: Invalid starting byte.

The problem here is that I use the standard output for the mapper output of the hadoop stream, and if I use the filename and use the script locally, the avro output is readable.

Any ideas, how to solve this problem? I think the problem is dealing with keys/values in streaming….

hadoop jar /opt/cloudera/parcels/CDH/lib/hadoop-0.20-mapreduce/contrib/streaming/hadoop-streaming.jar \
                      -input "xxx.txt" \
                      -mapper "/opt/anaconda/anaconda21/bin/python mapper.py x.avsc"  \
                      -reducer NONE \
                      -output "xxxxx" -file "mapper.py" \
                      -lazyOutput \
                      -file "x.avsc"

The mapper script is

import sys
import re
import os
from avro import schema, datafile
import avro.io as io
import StringIO

schema_str = open("xxxxx.avsc", 'r').read()
SCHEMA = schema.parse(schema_str)
rec_writer = io. DatumWriter(SCHEMA)
df_writer  = datafile. DataFileWriter(sys.stdout, rec_writer, SCHEMA,)
header = []
for field in SCHEMA.fields:
        header.append(field.name)

for line in sys.stdin:
    fields = line.rstrip().split("\x01")
    data   = dict(zip(header, fields))
    try:
        df_writer.append(data)
    except Exception, e:
        print "failed with data: %s" % str(data)
        print str(e)
df_writer.close()

This problem can finally be solved. Use the output format class and leave the avro binary transformation to it. In the streaming mapper, you only need to emit a JSON record.

hadoop jar /opt/cloudera/parcels/CDH/lib/hadoop-0.20-mapreduce/contrib/streaming/hadoop-streaming.jar \
              -libjars avro-json-1.2.jar \
              -jobconf output.schema.url=hdfs:///x.avsc \
              -input "xxxxx" \
              -mapper "/opt/anaconda/anaconda21/bin/python mapper.py x.avsc"  \
              -reducer NONE \
              -output "/xxxxx"  \
              -outputformat com.cloudera.science.avro.streaming.AvroAsJSONOutputFormat \
              -lazyOutput \
              -file "mapper.py" \
              -file "x.avsc"

Here is mapper.py

import sys
from avro import schema
import json

schema_str = open("xxxxx.avsc", 'r').read()
SCHEMA = schema.parse(schema_str)

header = []
for field in SCHEMA.fields:
    header.append(field.name)

for line in sys.stdin:
    fields = line.rstrip().split("\x01")
    data   = dict(zip(header, fields))
    try:
       print >> sys.stdout, json.dumps(data, encoding='ISO-8859-1')
    except Exception, e:
       print "failed with data: %s" % str(data)
       print str(e)