mrjob returned non-zero exit status 256

I'm new to map reduce and I'm trying to run a map reduce job using mrjob package of python. However, I encountered this error:

ERROR:mrjob.launch:Step 1 of 1 failed: Command '['/usr/bin/hadoop', 'jar', '/usr/lib/hadoop-mapreduce/hadoop-streaming.jar', '-files', 
'hdfs:///user/hadoop/tmp/mrjob/word_count.hadoop.20180831.035452.437014/files/mrjob.zip#mrjob.zip,
hdfs:///user/hadoop/tmp/mrjob/word_count.hadoop.20180831.035452.437014/files/setup-wrapper.sh#setup-wrapper.sh,
hdfs:///user/hadoop/tmp/mrjob/word_count.hadoop.20180831.035452.437014/files/word_count.py#word_count.py', '-archives', 
'hdfs:///user/hadoop/tmp/mrjob/word_count.hadoop.20180831.035452.437014/files/word_count_ccmr.tar.gz#word_count_ccmr.tar.gz', '-D', 
'mapreduce.job.maps=4', '-D', 'mapreduce.job.reduces=4', '-D', 'mapreduce.map.java.opts=-Xmx1024m', '-D', 'mapreduce.map.memory.mb=1200', '-D', 
'mapreduce.output.fileoutputformat.compress=true', '-D', 'mapreduce.output.fileoutputformat.compress.codec=org.apache.hadoop.io.compress.BZip2Codec', '-D', 
'mapreduce.reduce.java.opts=-Xmx1024m', '-D', 'mapreduce.reduce.memory.mb=1200', '-input', 'hdfs:///user/hadoop/test-1.warc', '-output', 
'hdfs:///user/hadoop/gg', '-mapper', 'sh -ex setup-wrapper.sh python word_count.py --step-num=0 --mapper', '-combiner', 
'sh -ex setup-wrapper.sh python word_count.py --step-num=0 --combiner', '-reducer', 'sh -ex setup-wrapper.sh python word_count.py --step-num=0 --reducer']' 
returned non-zero exit status 256

I've tried running it locally with python ./word_count.py input/test-1.warc > output and it's successful.

I'm using

1. python 2.7.14
2. Hadoop 2.8.3-amzn-1
3. pip 18.0
4. mrjob 0.6.4

Any ideas? Thanks!

This is my command in running the mapreduce job. I got it from cc-mrjob repository. The file is called run_hadoop.sh and I use chmod +x run_hadoop.sh

#!/bin/sh

JOB="$1"
INPUT="$2"
OUTPUT="$3"

sudo chmod +x $JOB.py

if [ -z "$JOB" ] || [ -z "$INPUT" ] || [ -z "$OUTPUT" ]; then
    echo "Usage: $0 <job> <input> <outputdir>"
    echo "  Run a CommonCrawl mrjob on Hadoop"
    echo
    echo "Arguments:"
    echo "  <job>     CCJob implementation"
    echo "  <input>   input path"
    echo "  <output>  output path (must not exist)"
    echo
    echo "Example:"
    echo "  $0 word_count input/test-1.warc  hdfs:///.../output/"
    echo
    echo "Note: don't forget to adapt the number of maps/reduces and the memory requirements"
    exit 1
fi

# strip .py from job name
JOB=${JOB%.py}

# wrap Python files for deployment, cf. below option --setup,
# see for details
# http://pythonhosted.org/mrjob/guides/setup-cookbook.html#putting-your-source-tree-in-pythonpath
tar cvfz ${JOB}_ccmr.tar.gz *.py

# number of maps resp. reduces 
NUM_MAPS=4
NUM_REDUCES=4

if [ -n "$S3_LOCAL_TEMP_DIR" ]; then
    S3_LOCAL_TEMP_DIR="--s3_local_temp_dir=$S3_LOCAL_TEMP_DIR"
else
    S3_LOCAL_TEMP_DIR=""
fi
python $JOB.py \
       -r hadoop \
       --jobconf "mapreduce.map.memory.mb=1200" \
       --jobconf "mapreduce.map.java.opts=-Xmx1024m" \
       --jobconf "mapreduce.reduce.memory.mb=1200" \
       --jobconf "mapreduce.reduce.java.opts=-Xmx1024m" \
       --jobconf "mapreduce.output.fileoutputformat.compress=true" \
       --jobconf "mapreduce.output.fileoutputformat.compress.codec=org.apache.hadoop.io.compress.BZip2Codec" \
       --jobconf "mapreduce.job.reduces=$NUM_REDUCES" \
       --jobconf "mapreduce.job.maps=$NUM_MAPS" \
       --setup 'export PYTHONPATH=$PYTHONPATH:'${JOB}'_ccmr.tar.gz#/' \
       --no-output \
       --cleanup NONE \
       $S3_LOCAL_TEMP_DIR \
       --output-dir "$OUTPUT" \
       "hdfs:///user/hadoop/$INPUT"

and I run it with ./run_hadoop.sh word_count test-1.warc output

where

• word_count is the job (file called word_count.py)
• test-1.warc is the input (located in hdfs:///user/hadoop/test-1.warc)
• and output is the output dir (located in hdfs:///user/hadoop/output) And I also make sure I always use different output for different job to prevent duplicate folder)

* Update *

I took a look at the syslog in HUE interface. And there's this error

org.apache.hadoop.mapreduce.v2.app.rm.RMContainerAllocator: Could not deallocate container for task attemptId attempt_1536113332062_0001_r_000003_0

is this related to the error I'm getting?

I also got this in one of the stderr of map attempt

/bin/sh: run_prestart: line 1: syntax error: unexpected end of file

and

No module named boto3

However, I installed boto3 using pip install boto3 in my emr. Is the module not available in hadoop?

I got it working by following this blog

http://benjamincongdon.me/blog/2018/02/02/MapReduce-on-Python-is-better-with-MRJob-and-EMR/

Essentially,

you have to include a .conf file for runner in hadoop. e.g. mrjob.conf

inside that file, use this

runners:
  hadoop:
    setup:
      - 'set -e'
      - VENV=/tmp/$mapreduce_job_id
      - if [ ! -e $VENV ]; then virtualenv $VENV; fi
      - . $VENV/bin/activate
      - 'pip install boto3'
      - 'pip install warc'
      - 'pip install https://github.com/commoncrawl/gzipstream/archive/master.zip'
    sh_bin: '/bin/bash -x'

and use the conf file by refering it to the run_hadoop.sh

python $JOB.py \
        --conf-path mrjob.conf \ <---- OUR CONFIG FILE
        -r hadoop \
        --jobconf "mapreduce.map.memory.mb=1200" \
        --jobconf "mapreduce.map.java.opts=-Xmx1024m" \
        --jobconf "mapreduce.reduce.memory.mb=1200" \
        --jobconf "mapreduce.reduce.java.opts=-Xmx1024m" \
        --jobconf "mapreduce.output.fileoutputformat.compress=true" \
        --jobconf "mapreduce.output.fileoutputformat.compress.codec=org.apache.hadoop.io.compress.BZip2Codec" \
        --jobconf "mapreduce.job.reduces=$NUM_REDUCES" \
        --jobconf "mapreduce.job.maps=$NUM_MAPS" \
        --setup 'export PYTHONPATH=$PYTHONPATH:'${JOB}'_ccmr.tar.gz#/' \
        --cleanup NONE \
        $S3_LOCAL_TEMP_DIR \
        --output-dir "hdfs:///user/hadoop/$OUTPUT" \
        "hdfs:///user/hadoop/$INPUT"

now if you call ./run_hadoop.sh word_count input/test-1.warc output, it should work!