This session covers how to work with PySpark interface to develop Spark applications. From loading, ingesting, and applying transformation on the data. The session covers how to work with different data sources of data, apply transformation, python best practices in developing Spark Apps. The demo covers integrating Apache Spark apps, In memory processing capabilities, working with notebooks, and integrating analytics tools into Spark Applications.

1. Programming in Spark using PySpark
Mostafa Elzoghbi
Sr. Technical Evangelist – Microsoft
@MostafaElzoghbi
http://mostafa.rocks

2. Session Objectives & Takeaways
• Programming Spark
• Spark Program Structure
• Working with RDDs
• Transformations versus Actions
• Lambda, Shared Variables (Broadcast vs accumulators)
• Visualizing big data in Spark
• Spark in the cloud (Azure)
• Working with cluster types, notebooks, scaling.

3. Python Spark (pySpark)
• We are using the Python programming interface to Spark (pySpark)
• pySpark provides an easy-to-use programming abstraction and parallel
runtime:
“Here’s an operation, run it on all of the data”
• RDDs are the key concept

4. Apache Spark Driver and
Workers
• A Spark program is two programs:
• A driver program and a workers program
• Worker programs run on cluster nodes or in
local threads
• RDDs (Resilient Distributed Datasets) are
distributed

5. Spark Essentials: Master
• The master parameter for a SparkContext determines which type and size of
cluster to use

6. Spark Context
• A Spark program first creates a SparkContext object
» Tells Spark how and where to access a cluster
» pySpark shell and Databricks cloud automatically create the sc variable
» iPython and programs must use a constructor to create a new
SparkContext
• Use SparkContext to create RDDs

7. Resilient Distributed Datasets
• The primary abstraction in Spark
» Immutable once constructed
» Track lineage information to efficiently recompute lost data
» Enable operations on collection of elements in parallel
• You construct RDDs
» by parallelizing existing Python collections (lists)
» by transforming an existing RDDs
» from files in HDFS or any other storage system

8. RDDs
• Spark revolves around the concept of a resilient distributed dataset (RDD),
which is a fault-tolerant collection of elements that can be operated on in
parallel.
• Two types of operations: transformations and actions
• Transformations are lazy (not computed immediately)
• Transformed RDD is executed when action runs on it
• Persist (cache) RDDs in memory or disk

10. Creating an RDD
• Create RDDs from Python collections (lists)
• From HDFS, text files, Hypertable, Amazon S3, Apache Hbase, SequenceFiles,
any other Hadoop InputFormat, and directory or glob wildcard: /data/201404*

11. Working with RDDs
• Create an RDD from a data source: <list>
• Apply transformations to an RDD: map filter
• Apply actions to an RDD: collect count

12. Spark Transformations
• Create new datasets from an existing one
• Use lazy evaluation: results not computed right away –
• instead Spark remembers set of transformations applied to base dataset
» Spark optimizes the required calculations
» Spark recovers from failures and slow workers
• Think of this as a recipe for creating result

13. Python lambda Functions
• Small anonymous functions (not bound to a name)
lambda a, b: a+b
» returns the sum of its two arguments
• Can use lambda functions wherever function objects are required
• Restricted to a single expression

14. Spark Actions
• Cause Spark to execute recipe to transform source
• Mechanism for getting results out of Spark

15. Spark Program Lifecycle
1. Create RDDs from external data or parallelize a collection in your driver
program
2. Lazily transform them into new RDDs
3. cache() some RDDs for reuse -- IMPORTANT
4. Perform actions to execute parallel
5. Computation and produce results

16. pySpark Shared Variables
• Broadcast Variables
» Efficiently send large, read-only value to all workers
» Saved at workers for use in one or more Spark operations
» Like sending a large, read-only lookup table to all the nodes
At the driver: broadcastVar = sc.broadcast([1, 2, 3])
At a worker: broadcastVar.value

17. • Accumulators
» Aggregate values from workers back to driver
» Only driver can access value of accumulator
» For tasks, accumulators are write-only
» Use to count errors seen in RDD across workers
>>> accum = sc.accumulator(0)
>>> rdd = sc.parallelize([1, 2, 3, 4])
>>> def f(x):
>>> global accum
>>> accum += x
>>> rdd.foreach(f)
>>> accum.value
Value: 10

18. Visualizing Big Data in the browser
• Challenges:
• Manipulating large data can take long time
Memory: caching -> Scale clusters
CPU: Parallelism -> Scale clusters
• We have more data points than possible pixels
> Summarize: Aggregation, Pivoting (more data than pixels)
> Model (Clustering, Classification, D. Reduction, …etc)
> Sample: approximate (faster) and exact sampling
• Internal Tools: Matplotlib, GGPlot, D3, SVC, and more.

19. Spark Kernels and MAGIC keywords
• PySpark kernel supports set of %%MAGIC keywords
• It supports built-in IPython built-in magics, including %%sh.
• Auto visualization
• Magic keywords:
• %%SQL % Spark SQL
• %%lsmagic % List all supported magic keywords (Important)
• %env % Set environment variable
• %run % Execute python code
• %who % List all variables of global scope
• Run code from a different kernel in a notebook.

20. Spark in Azure
Hadoop clusters in Azure are packaged under “HDInsight” service

21. Spark in Azure
• Create clusters in few clicks
• Apache Spark comes only in Linux OS.
• Multiple HDP versions
• Comes with preloaded: SSH, Hive, Oozie, DLS, Vnets.
• Multiple Storage options:
• Azure Storage
• ADL store
• External metadata store in SQL server database for
Hive and Oozie.
• All notebooks are stored in the storage account
associated with Spark cluster
• Zeppelin notebook is available on certain Spark
versions but not all.

22. Programming Spark Apps in HDInsight
• Supports four kernels in Jupyter in HDInsight Spark clusters in Azure

23. DEMO
Spark Apps using Jupyter

24. References
• Spark Programming Guide
http://spark.apache.org/docs/latest/programming-guide.html
• edx.org: Free Apache Spark courses
• Visualizations for Databricks
https://docs.cloud.databricks.com/docs/latest/databricks_guide/01%20
Databricks%20Overview/15%20Visualizations/0%20Visualizations%20Ov
erview.html
• SPARKHub by Databricks
https://sparkhub.databricks.com/resources/

25. Thank you
• Check out my blog big data articles: http://mostafa.rocks
• Follow me on Twitter: @MostafaElzoghbi
• Want some help in building cloud solutions? Contact me to know more.