Step 1: Download winutils.exe for the version of Hadoop your Spark uses. In our case, Hadoop is 2.7, so download the winutils.exe for this version only. Thank you for your patience, i downloaded the winutils.exe and i created a HADOOPHOME and i put it in Path. When i executed my job, they load my files and full this. 问题在windows 环境使用Java下调试远程虚拟机中的Hadoop集群报错，问题很奇怪，说是少了 winutils.exe 文件，而且少了HADOOPHOME 的环境变量；我是部署在虚拟机CentOS 7 上的集群，难道Windows 上使用 它的Hadoop还需要自己安装环境，事实上，是真的。. This is a known issue to many Hadoop related projects and has been asked about in many different online communities such as Stack Overflow or Cloudera's issue tracking system. Java.io.IOException: Could not locate executable null bin winutils.exe in the Hadoop binaries.

1. Winutils.exe Hadoop
2. Winutils.exe

I decided to teach myself how to work with big data and came across Apache Spark. While I had heard of Apache Hadoop, to use Hadoop for working with big data, I had to write code in Java which I was not really looking forward to as I love to write code in Python. Spark supports a Python programming API called PySpark that is actively maintained and was enough to convince me to start learning PySpark for working with big data.

Winutils.exe Hadoop

In this post, I describe how I got started with PySpark on Windows. My laptop is running Windows 10. So the screenshots are specific to Windows 10. I am also assuming that you are comfortable working with the Command Prompt on Windows. You do not have to be an expert, but you need to know how to start a Command Prompt and run commands such as those that help you move around your computer’s file system. In case you need a refresher, a quick introduction might be handy.

Often times, many open source projects do not have good Windows support. So I had to first figure out if Spark and PySpark would work well on Windows. The official Spark documentation does mention about supporting Windows.

Installing Prerequisites

PySpark requires Java version 7 or later and Python version 2.6 or later. Let’s first check if they are already installed or install them and make sure that PySpark can work with these two components.

Java

Java is used by many other software. So it is quite possible that a required version (in our case version 7 or later) is already available on your computer. To check if Java is available and find it’s version, open a Command Prompt and type the following command.

If Java is installed and configured to work from a Command Prompt, running the above command should print the information about the Java version to the console. For example, I got the following output on my laptop.

Instead if you get a message like

It means you need to install Java. To do so,

1. Go to the Java download page. In case the download link has changed, search for Java SE Runtime Environment on the internet and you should be able to find the download page.

2. Click the Download button beneath JRE

3. Accept the license agreement and download the latest version of Java SE Runtime Environment installer. I suggest getting the exe for Windows x64 (such as jre-8u92-windows-x64.exe) unless you are using a 32 bit version of Windows in which case you need to get the Windows x86 Offline version.

4. Run the installer.

After the installation is complete, close the Command Prompt if it was already open, open it and check if you can successfully run java -version command.

Python

Python is used by many other software. So it is quite possible that a required version (in our case version 2.6 or later) is already available on your computer. To check if Python is available and find it’s version, open a Command Prompt and type the following command.

If Python is installed and configured to work from a Command Prompt, running the above command should print the information about the Python version to the console. For example, I got the following output on my laptop.

Instead if you get a message like

It means you need to install Python. To do so,

1. Go to the Python download page.

2. Click the Latest Python 2 Release link.

3. Download the Windows x86-64 MSI installer file. If you are using a 32 bit version of Windows download the Windows x86 MSI installer file.

4. When you run the installer, on the Customize Python section, make sure that the option Add python.exe to Path is selected. If this option is not selected, some of the PySpark utilities such as pyspark and spark-submit might not work.

After the installation is complete, close the Command Prompt if it was already open, open it and check if you can successfully run python --version command.

Installing Apache Spark

1. Go to the Spark download page.

2. For Choose a Spark release, select the latest stable release of Spark.

3. For Choose a package type, select a version that is pre-built for the latest version of Hadoop such as Pre-built for Hadoop 2.6.

4. For Choose a download type, select Direct Download.

5. Click the link next to Download Spark to download a zipped tarball file ending in .tgz extension such as spark-1.6.2-bin-hadoop2.6.tgz.

6. In order to install Apache Spark, there is no need to run any installer. You can extract the files from the downloaded tarball in any folder of your choice using the 7Zip tool.

   Make sure that the folder path and the folder name containing Spark files do not contain any spaces.

In my case, I created a folder called spark on my C drive and extracted the zipped tarball in a folder called spark-1.6.2-bin-hadoop2.6. So all Spark files are in a folder called C:sparkspark-1.6.2-bin-hadoop2.6. From now on, I will refer to this folder as SPARK_HOME in this post.

To test if your installation was successful, open a Command Prompt, change to SPARK_HOME directory and type binpyspark. This should start the PySpark shell which can be used to interactively work with Spark. I got the following messages in the console after running binpyspark command.

The last message provides a hint on how to work with Spark in the PySpark shell using the sc or sqlContext names. For example, typing sc.version in the shell should print the version of Spark. You can exit from the PySpark shell in the same way you exit from any Python shell by typing exit().

The PySpark shell outputs a few messages on exit. So you need to hit enter to get back to the Command Prompt.

Configuring the Spark Installation

Starting the PySpark shell produces a lot of messages of type INFO, ERROR and WARN. In this section we will see how to remove these messages.

By default, the Spark installation on Windows does not include the winutils.exe utility that is used by Spark. If you do not tell your Spark installation where to look for winutils.exe, you will see error messages when running the PySpark shell such as

This error message does not prevent the PySpark shell from starting. However if you try to run a standalone Python script using the binspark-submit utility, you will get an error. For example, try running the wordcount.py script from the examples folder in the Command Prompt when you are in the SPARK_HOME directory.

which produces the following error that also points to missing winutils.exe

Installing winutils

Let’s download the winutils.exe and configure our Spark installation to find winutils.exe.

1. Create a hadoopbin folder inside the SPARK_HOME folder.

2. Download the winutils.exe for the version of hadoop against which your Spark installation was built for. In my case the hadoop version was 2.6.0. So I downloaded the winutils.exe for hadoop 2.6.0 and copied it to the hadoopbin folder in the SPARK_HOME folder.

3. Create a system environment variable in Windows called SPARK_HOME that points to the SPARK_HOME folder path. Search the internet in case you need a refresher on how to create environment variables in your version of Windows such as articles like these.

4. Create another system environment variable in Windows called HADOOP_HOME that points to the hadoop folder inside the SPARK_HOME folder.

Since the hadoop folder is inside the SPARK_HOME folder, it is better to create HADOOP_HOME environment variable using a value of %SPARK_HOME%hadoop. That way you don’t have to change HADOOP_HOME if SPARK_HOME is updated.

If you now run the binpyspark script from a Windows Command Prompt, the error messages related to winutils.exe should be gone. For example, I got the following messages after running the binpyspark utility after configuring winutils

The binspark-submit utility can also be successfully used to run wordcount.py script.

Winutils.exe

Configuring the log level for Spark

There are still a lot of extra INFO messages in the console everytime you start or exit from a PySpark shell or run the spark-submit utility. So let’s make one more change to our Spark installation so that only warning and error messages are written to the console. In order to do this

1. Copy the log4j.properties.template file in the SPARK_HOMEconf folder as log4j.properties file in the SPARK_HOMEconf folder.

2. Set the log4j.rootCategory property value to WARN, console

3. Save the log4j.properties file.

Now any informative messages will not be logged to the console. For example, I got the following messages after running the binpyspark utility once I configured the log level to WARN.

Summary

In order to work with PySpark, start a Windows Command Prompt and change into your SPARK_HOME directory.

• To start a PySpark shell, run the binpyspark utility. Once your are in the PySpark shell use the sc and sqlContext names and type exit() to return back to the Command Prompt.

• To run a standalone Python script, run the binspark-submit utility and specify the path of your Python script as well as any arguments your Python script needs in the Command Prompt. For example, to run the wordcount.py script from examples directory in your SPARK_HOME folder, you can run the following command

  binspark-submit examplessrcmainpythonwordcount.py README.md

References

I used the following references to gather information about this post.

• Downloading Spark and Getting Started (chapter 2) from O’Reilly’s Learning Spark book.

Share on: Twitter ❄ Facebook ❄ Google+ ❄ Email

Any suggestions or feedback? Leave your comments below.

Whether you want to unit test your Spark Scala application using Scala Tests or want to run some Spark application on Windows, you need to perform a few basics settings and configurations before you do so. In this post, I will explain the configurations that will help you start your journey to run your spark application seamlessly on your windows machines. Let’s get started –

First, note that you don’t need Hadoop installation in your windows machine to run Spark. You need a way to use POSIX like file access operations in windows which is implemented using winutils.exe using some Windows APIs.

Step 1. Download winutils.exe binary from this link – https://github.com/steveloughran/winutils, and place it on a folder like this – – C:/hadoop/bin, make sure you are downloading the same version as on which your Spark version is compiled against. You can check the version of Hadoop your spark version was compiled with using pom of spark binary you are using – https://search.maven.org/artifact/org.apache.spark/spark-parent_2.11/2.4.4/pom

Step 2. set HADOOP_HOME and PATH – In your environment variables either using Control Panel ( available to all apps – recommended option) or on command prompt ( for the current session) – set HADOOP_HOME as C:/hadoop or the path inside which you created bin directory where winutils.exe is present.

Next is to add %HADOOP_HOME%/bin to the PATH.

That’s all !!

Now you can run any spark app on your local windows machine in IntelliJ, Eclipse, or in spark-shell. Please comment below for any issues!

More Spark Posts –