🐍Python: Installing PySpark and GraphFrames on a Windows 10 Machine

One of the To-Dos for this week was to set up the environment so I can start learning PySpark and GraphFrames based on the examples from Needham & Hodler’s free book on Graph Algorithms. Therefore, I downloaded and installed the Java SDK 8 from the Oracle website (requires an Oracle account) and the latest stable version of Python (Python 3.9.2), downloaded and unzipped the Apache Spark package locally on a Windows 10 machine, respectively the Winutils tool as described here.

The setup requires several environment variables that need to be created, respectively the Path variable needs to be extended with further values (delimited by ";"). In the end I added the following values:

Variable	Value
HADOOP_HOME	D:\Programs\spark-3.0.2-bin-hadoop2.7
SPARK_HOME	D:\Programs\spark-3.0.2-bin-hadoop2.7
JAVA_HOME	D:\Programs\Java\jdk1.8.0_281
PYTHONPATH	D:\Programs\Python\Python39\
PYTHONPATH	;%SPARK_HOME%\python
PYTHONPATH	%SPARK_HOME%\python\lib\py4j-0.10.9-src.zip
PATH	%HADOOP_HOME%\bin
PATH	%SPARK_HOME%\bin
PATH	%PYTHONPATH%
PATH	%PYTHONPATH%\DLLs
PATH	%PYTHONPATH%\Lib
PATH	%JAVA_HOME%\bin

I tried then running the first example from Chapter 3 using the Spyder IDE, though the environment didn’t seem to recognize the 'graphframes' library. As long it's not already available, the graphframes .jar file (e.g. graphframes-0.8.1-spark3.0-s_2.12.jar) corresponding to the installed Spark version must be downloaded and copied in the Spark folder where the other .jar files are available (e.g. .\spark-3.0.2-bin-hadoop2.7\jars). With this change I could finally run my example, though it took me several tries to get this right. 

During Python's installation I had to change the value for the LongPathsEnabled setting from 0 to 1 via regedit to allow path lengths longer than 260 characters, as mentioned in the documentation. The setting is available via the following path:
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem

In the process I also tried installing ‘pyspark’ and ‘graphframes’ via the Anaconda tool with the following commands:

pip3 install --user pyspark
pip3 install --user graphframes

From Anaconda’s point of view the installation was correct, fact which pointed me to the missing 'graphframe' library.

It took me 4-5 hours of troubleshooting and searching until I got my environment setup. I still have two more warnings to solve, though I will look into this later:
WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
WARN ProcfsMetricsGetter: Exception when trying to compute pagesize, as a result reporting of ProcessTree metrics is stopped

Notes:
Spaces in the folder's names might creates issues. Therefore, I used 'Programs' instead of 'Program Files' as main folder. 
There seem to be some confusion what environment variables are needed and how they need to be configured.
Unfortunately, the troubleshooting involved in setting up an environment and getting a simple example to work seems to be a recurring story over the years. Same situation was with the programming languages from 15-20 years ago.