Programmatic Deployment to Elastic Mapreduce with Boto and Bootstrap Action

A while back I wrote about How to combine Elastic Mapreduce/Hadoop with other Amazon Web Services. This posting is a small update to that, showing how to deploy extra packages with Boto for Python. Note that Boto can deploy mappers and reducers in written any language supported by Elastic Mapreduce. In the example below (it can also be found on github –, i.e. check out with git clone

Imports and connection to elastic mapreduce on AWS

#!/usr/bin/env python
import boto
import boto.emr
from boto.emr.step import StreamingStep
from boto.emr.bootstrap_action import BootstrapAction
import time

# set your aws keys and S3 bucket, e.g. from environment or .boto

conn = boto.connect_emr(AWSKEY,SECRETKEY)

Bootstrap step being created
In this case a shell script from s3, note that this could contain sudo commands in order to do apt-get installs, e.g to install classic programming language packages like gfortran or open-cobol, or more modern languages like ghc6 (haskell), or any code, e.g. checking out latest version of a programming language (e.g. Google Go with hg clone -r release $GOROOT) interpreter/compiler and compile it before using it in your mappers or reducers

bootstrap_step = BootstrapAction("download.tst", "s3://elasticmapreduce/bootstrap-actions/",None)

Create map and reduce processing step
Using cache_files also adds a python library available for import (the other way could be to do sudo easy_install boto in the bootstrap step, which would be easier since the boto module wouldn’t have to be unpacked manually in the python code, see my previous posting for details about unpacking). Note that the mapper and reducer could be any language as long as you either have compiled in or have installed an interpreter for it with the bootstrap step.

step = StreamingStep(
  cache_files = ["s3n://" + S3_BUCKET + "/boto.mod#boto.mod"],
  output='s3n://' + S3_BUCKET + '/output/wordcount_output')

jobid = conn.run_jobflow(
    log_uri="s3://" + S3_BUCKET + "/logs", 
    steps = [step],

Wait for job to start
This waits for the Elastic Mapreduce Job to start and prints out status, one of the statuses between starting and running being bootstrapping.

state = conn.describe_jobflow(jobid).state
print "job state = ", state
print "job id = ", jobid
while state != u'COMPLETED':
    print time.localtime()
    state = conn.describe_jobflow(jobid).state
    print "job state = ", state
    print "job id = ", jobid

print "final output can be found in s3://" + S3_BUCKET + "/output" + TIMESTAMP
print "try: $ s3cmd sync s3://" + S3_BUCKET + "/output" + TIMESTAMP + " ."

Validation of what really happened
One way to validate is to check that your mappers and reducers written in any language (i.e. for which compiler that you installed with bootstrap action), e.g. the classic mapreduce word count written in classic languages like Cobol or Fortran 95? The other way is to check the s3 logs, the log directory for an elastic mapreduce job has the following subdirectories:

daemons  jobs  node  steps  task-attempts

In the node directory, each EC2 instance used in the job has a directory, and underneath each of them there is a bootstrap_actions directory with the master.log and stderr, stdout and controller logs. In the case presented above bootstrap output is shown underneath.
stderr output

--2010-10-01 17:38:38--
Connecting to||:80... connected.
HTTP request sent, awaiting response... 
  HTTP/1.1 200 OK
  x-amz-id-2: NezTUU9MIzPwo72lJWPYIMo2wwlbDGi1IpDbV/mO07Nca4VarSV8l7j/2ArmclCB
  x-amz-request-id: 3E71CC3323EC1189
  Date: Fri, 01 Oct 2010 17:38:39 GMT
  Last-Modified: Thu, 03 Jun 2010 01:57:13 GMT
  ETag: "47a007dae0ff192c166764259246388c"
  Content-Type: application/octet-stream
  Content-Length: 153
  Connection: keep-alive
  Server: AmazonS3
Length: 153 [application/octet-stream]
Saving to: `file.tar.gz'

     0K                                                       100% 24.3M=0s

2010-10-01 17:38:38 (24.3 MB/s) - `file.tar.gz' saved [153/153]


2010-10-01T17:38:35.141Z INFO Fetching file 's3://elasticmapreduce/bootstrap-actions/'
2010-10-01T17:38:38.411Z INFO Working dir /mnt/var/lib/bootstrap-actions/1
2010-10-01T17:38:38.411Z INFO Executing /mnt/var/lib/bootstrap-actions/1/
2010-10-01T17:38:38.936Z INFO Execution ended with ret val 0
2010-10-01T17:38:38.938Z INFO Execution succeeded

The posting has shown how to programmatically install packages (e.g. programming languages) to EC2 nodes running elastic mapreduce. Since elastic mapreduce in streaming mode supports any programming language this can make it easier to deploy and test mappers and reducers written in your favorite language, and even automate it. (Opens a few doors for parallelization of legacy code)

Atbrox on LinkedIn

Best regards,
Amund Tveit, co-founder of Atbrox

This entry was posted in cloud computing, Hadoop and Mapreduce and tagged , , , , , , . Bookmark the permalink.

4 Responses to Programmatic Deployment to Elastic Mapreduce with Boto and Bootstrap Action

  1. Pingback: Weekly Record « Vanessa's Imiloa

  2. Pingback: Running my first EMR – lessons learned | Tom Ron