Introduction

Hadoop framework distributes code execution automatically in a multi node cluster. This code is also distributed against the dataset. Code development in Hadoop can be done in Java and one has to implement a map function and a reduce function; both manipulate keys and values as inputs and outputs. At a higher level, there are two scripting languages that simplify the code: PIG is a specific scripting language, HIVE looks like SQL. So using HIVE is quite easy. It has a bunch of extension functions (called user defined functions) to transform data like regular expression tools and so on. A developer can add user defined functions, by developing them in Java. Another way to have a procedural logic that complements SQL Set-based language is to use a language like Python:

image 

The goal of that post is to show an example of such a combination.

Here is how that could look on a small cluster. The work load is distributed on the different worker nodes:
image

At a worker node level, a Python process is created by core. Each process receives its part of the whole dataset:
image

Windows Azure comes with its Hadoop as a service called HDInsight. This allows to execute HIVE, PIG, and other Map/reduce jobs a few minutes after requesting the creation of a cluster. For HIVE, HDInsight comes with a sample table. Let’s run a HIVE + Python job against that hivesampletable table.

Hive and Python Script

In this example, we use a Python module to calculate the hash of a label in the sample table.

Hive is used to get the data, partition it and send the rows to the Python processes which are created on the different cluster nodes. Here is the code:

add file simple_sample.py;

SELECT TRANSFORM (clientid, devicemake, devicemodel)
    USING 'D:\Python27\python.exe simple_sample.py' AS 
    (clientid string, phoneLabel string, phoneHash string)
FROM hivesampletable
ORDER BY clientid LIMIT 50;

This can be read has: in the first 50 rows of hivesampletable table, select clientid, devicemake, devicemodel , pass them to the simple_sample.py python script that can be run with D:\Python27\python.exe. The script will send back columns clientid (a string), phoneLabel (a string) and phoneHash (a string).

Hive sends data to the simple_sample.py scripts. Here is the code of that script:

import sys
import string
import hashlib

while True:
    line = sys.stdin.readline()
    if not line:
        break

    line = string.strip(line, "\n ")
    clientid, devicemake, devicemodel = string.split(line, "\t")
    phone_label = devicemake + ' ' + devicemodel
    print "\t".join([clientid, phone_label, hashlib.md5(phone_label).hexdigest()])

This script expects stdin lines. It parses them, and obtains the columned passed by Hive: clientid, devicemake, devicemodel. From that columns, it deduces the resulting columns: clientid, phoneLabel, phoneHash. In order to calculate phoneHash, it uses an imported module (hashlib). In order to output the result, the python script writes it to stdout, separated by TAB.

Let’s run it with PowerShell

Here is a sample PowerShell script that

  • creates an HDInsight cluster
  • Runs the job
  • Gets the result
  • Removes the cluster

Before running the script, the HIVE and the Python script must have been copied to the the Windows Azure storage:

image

Here is the PowerShell script:

Import-Module azure
Add-AzureAccount

$Subscription = 'Azdem169A44055X'
$defaultStorageAccount = 'monstockageazure'
$clusterName = 'monclusterhadoop'
$clusterVersion='2.1'
$clusterAdmin = 'cornac'
$clusterPassword = 'LElzgqy#n87'

$passwd = ConvertTo-SecureString $clusterPassword -AsPlainText -Force
$clusterCredentials = New-Object System.Management.Automation.PSCredential ($clusterAdmin, $passwd)

Set-AzureSubscription -SubscriptionName $Subscription -CurrentStorageAccount $defaultStorageAccount
Select-AzureSubscription -Current $Subscription

$storageAccount1 = (Get-AzureSubscription $Subscription).CurrentStorageAccountName
$key1 = Get-AzureStorageKey -StorageAccountName $storageAccount1 | %{ $_.Primary }

New-AzureHDInsightClusterConfig -ClusterSizeInNodes 3 |
    Set-AzureHDInsightDefaultStorage -StorageAccountName "${storageAccount1}.blob.core.windows.net" -StorageAccountKey $key1 `
        -StorageContainerName $clusterName |
    New-AzureHDInsightCluster -Name $clusterName -Version $clusterVersion -Location "North Europe" -Credential $clusterCredentials

Use-AzureHDInsightCluster "monclusterhadoop"

$hiveJobVT = New-AzureHDInsightHiveJobDefinition -File "wasb://messcripts@monstockageazure.blob.core.windows.net/simple_sample.hql"
$hiveJobVT.Files.Add("wasb://messcripts@monstockageazure.blob.core.windows.net/simple_sample.py")
$startedHiveJobVT = $hiveJobVT | Start-AzureHDInsightJob -Credential $clusterCredentials -Cluster "monclusterhadoop"

$startedHiveJobVT | Wait-AzureHDInsightJob -Credential $clusterCredentials

Get-AzureHDInsightJobOutput -StandardError -JobId $startedHiveJobVT.JobId -Cluster "monclusterhadoop"
Get-AzureHDInsightJobOutput -StandardOutput -JobId $startedHiveJobVT.JobId -Cluster "monclusterhadoop"

Remove-AzureHDInsightCluster -Name $clusterName

Here is a sample execution result:

PS C:\benjguin\BigData_Hadoop\demos\simple> Import-Module azure
Add-AzureAccount


PS C:\benjguin\BigData_Hadoop\demos\simple> Import-Module azure
Add-AzureAccount

$Subscription = 'Azdem169A44055X'
$defaultStorageAccount = 'monstockageazure'
$clusterName = 'monclusterhadoop'
$clusterVersion='2.1'
$clusterAdmin = 'cornac'
$clusterPassword = 'LElzgqy#n87'

$passwd = ConvertTo-SecureString $clusterPassword -AsPlainText -Force
$clusterCredentials = New-Object System.Management.Automation.PSCredential ($clusterAdmin, $passwd)

Set-AzureSubscription -SubscriptionName $Subscription -CurrentStorageAccount $defaultStorageAccount
Select-AzureSubscription -Current $Subscription

$storageAccount1 = (Get-AzureSubscription $Subscription).CurrentStorageAccountName
$key1 = Get-AzureStorageKey -StorageAccountName $storageAccount1 | %{ $_.Primary }

New-AzureHDInsightClusterConfig -ClusterSizeInNodes 3 |
    Set-AzureHDInsightDefaultStorage -StorageAccountName "${storageAccount1}.blob.core.windows.net" -StorageAccountKey $key1 `
        -StorageContainerName $clusterName |
    New-AzureHDInsightCluster -Name $clusterName -Version $clusterVersion -Location "North Europe" -Credential $clusterCredentials



ClusterSizeInNodes    : 3
ConnectionUrl         : https://monclusterhadoop.azurehdinsight.net
CreateDate            : 03/03/2014 14:15:50
DefaultStorageAccount : monstockageazure.blob.core.windows.net
HttpUserName          : cornac
Location              : North Europe
Name                  : monclusterhadoop
State                 : Running
StorageAccounts       : {}
SubscriptionId        : 0fa85b4c-aa27-44ba-84e5-fa51aac32734
UserName              : cornac
Version               : 2.1.4.0.526800
VersionStatus         : Compatible

PS C:\benjguin\BigData_Hadoop\demos\simple> Use-AzureHDInsightCluster "monclusterhadoop"

$hiveJobVT = New-AzureHDInsightHiveJobDefinition -File "wasb://messcripts@monstockageazure.blob.core.windows.net/simple_sample.hql"
$hiveJobVT.Files.Add("wasb://messcripts@monstockageazure.blob.core.windows.net/simple_sample.py")
$startedHiveJobVT = $hiveJobVT | Start-AzureHDInsightJob -Credential $clusterCredentials -Cluster "monclusterhadoop"

$startedHiveJobVT | Wait-AzureHDInsightJob -Credential $clusterCredentials

Get-AzureHDInsightJobOutput -StandardError -JobId $startedHiveJobVT.JobId -Cluster "monclusterhadoop"
Get-AzureHDInsightJobOutput -StandardOutput -JobId $startedHiveJobVT.JobId -Cluster "monclusterhadoop"
Successfully connected to cluster monclusterhadoop


Cluster         : monclusterhadoop
ExitCode        : 0
Name            : Hive: simple_sample.hql
PercentComplete : map = 100%,  reduce = 100%
Query           : 
State           : Completed
StatusDirectory : b4328d2f-589c-412e-83e5-f8a544cb321c
SubmissionTime  : 03/03/2014 14:36:48
JobId           : job_201403031426_0003


Logging initialized using configuration in file:/C:/apps/dist/hive-0.11.0.1.3.5.0-03/conf/hive-log4j.properties
Added resource: simple_sample.py
Total MapReduce jobs = 1
Launching Job 1 out of 1
Number of reduce tasks determined at compile time: 1
In order to change the average load for a reducer (in bytes):
  set hive.exec.reducers.bytes.per.reducer=<number>
In order to limit the maximum number of reducers:
  set hive.exec.reducers.max=<number>
In order to set a constant number of reducers:
  set mapred.reduce.tasks=<number>
Starting Job = job_201403031426_0004, Tracking URL = http://jobtrackerhost:50030/jobdetails.jsp?jobid=job_201403031426_0004
Kill Command = "C:\apps\dist\hadoop-1.2.0.1.3.5.0-03\bin\hadoop.cmd" job  -kill job_201403031426_0004
Hadoop job information for Stage-1: number of mappers: 1; number of reducers: 1
2014-03-03 14:37:20,821 Stage-1 map = 0%,  reduce = 0%
2014-03-03 14:37:25,883 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:26,915 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:27,946 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:28,962 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:29,977 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:30,993 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:32,008 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:33,024 Stage-1 map = 100%,  reduce = 0%, Cumulative CPU 5.469 sec
2014-03-03 14:37:34,024 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 5.469 sec
2014-03-03 14:37:35,040 Stage-1 map = 100%,  reduce = 33%, Cumulative CPU 5.469 sec
2014-03-03 14:37:36,055 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 9.265 sec
2014-03-03 14:37:37,055 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 9.265 sec
2014-03-03 14:37:38,055 Stage-1 map = 100%,  reduce = 100%, Cumulative CPU 9.265 sec
MapReduce Total cumulative CPU time: 9 seconds 265 msec
Ended Job = job_201403031426_0004
MapReduce Jobs Launched: 
Job 0: Map: 1  Reduce: 1   Cumulative CPU: 9.265 sec   HDFS Read: 266 HDFS Write: 2684 SUCCESS
Total MapReduce CPU Time Spent: 9 seconds 265 msec
OK
Time taken: 36.86 seconds, Fetched: 50 row(s)

100004    Motorola Droid X    02a4198bedd37119dabcbb2e8fb4ec92
100015    Apple iPod Touch 4.3.x    d9bc8c98d6a6556656e774a64f7b8bb2
100015    Apple iPod Touch 4.3.x    d9bc8c98d6a6556656e774a64f7b8bb2
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100035    LG VS910    b4bfdffa3e288ed0283ae8c8a37c455e
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100036    Samsung SCH-i400    6b314786cda6123fc06eeb855825aea7
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100041    RIM 9650    d476f3687700442549a83fac4560c51c
100042    Apple iPhone 4.2.x    375ad9a0ddc4351536804f1d5d0ea9b9
100042    Apple iPhone 4.2.x    375ad9a0ddc4351536804f1d5d0ea9b9
100042    Apple iPhone 4.2.x    375ad9a0ddc4351536804f1d5d0ea9b9

Remove-AzureHDInsightCluster -Name $clusterName

Smile

Benjamin (@benjguin)