Show Menu

Apache Spark by

Introd­uction to Apache Spark

An open-s­ource, distri­buted processing system used for big data workloads.
Utilizes in-memory caching, and optimized query execution for fast analytic queries against data of any size.
Develo­pment APIs. Batch proces­sing, intera­ctive queries, real-time analytics, machine learning, and graph processing
Apache Spark vs. Apache Hadoop
Hadoop MapReduce is a progra­mming model for processing big data sets with a parallel, distri­buted algorithm.
With each step, MapReduce reads data from the cluster, performs operat­ions, and writes the results back to HDFS. Because each step requires a disk read, and write, MapReduce jobs are slower due to the latency of disk I/O.
Because each step requires a disk read, and write, MapReduce jobs are slower due to the latency of disk I/O.
Spark was created to address the limita­tions to MapReduce
Spark does processing in-memory, reducing the number of steps in a job, and by reusing data across multiple parallel operat­ions.
With Spark, only one-step is needed where data is read into memory, operations performed, and the results written back


Data Definition Language
Resilient Distri­buted Dataset (RDD) is the fundam­ental data structure of Spark
immutable (and therefore fault-­tol­erant) Distri­buted collec­tions of objects of any type.
Each Dataset in Spark RDD is divided into logical partitions across the cluster
thus can be operated in parallel, on different nodes of the cluster.
RDD features
Lazy Evaluation
Transf­orm­ation do not compute the results as and when stated
In-Memory Comput­ation
Data is kept in RAM (random access memory) instead of the slower disk drives
Fault Tolerance
Tracks data lineage inform­ation to allow for rebuilding lost data automa­tically on failure
Immuta­bility simply rules out lots of potential problems due to various updates from varying threads at once.
Having Immutable data is safer to share across processes
Each node in a spark cluster contains one or more partit­ions.
Two ways to apply operations on RDDs
1, Transf­orm­ation − These are the operat­ions, which are applied on a RDD to create a new RDD. Filter, groupBy and map are the examples of transf­orm­ations.
Narrow Transf­orm­ations: In this type, all the elements which are required to compute the records in a single partition live in that single partition.
Wide Transf­orm­ations: Here, all elements required to compute the records in that single partition may live in many of the partitions of the parent RDD. These use groupb­yKey() and reduce­byK­ey().
2, Action − These are the operations that are applied on RDD, which instructs Spark to perform comput­ation and send the result back to the driver.
count(), collect(), take(n), top(), count value(), reduce(), fold(), aggreg­ate(), foreach().
Create Datafr­amees
via CSV
df=spa­rk.r­ea­d.o­pti­on(­"­hea­der­"­,True) \ .csv("/­tmp­/re­sou­rce­s/s­imp­le-­zip­cod­es.c­sv­")
If you have a header with column names on your input file, you need to explicitly specify True
df = spark.r­ea­d.c­sv(­"­pat­h1,­pat­h2,­pat­h3") ; df = spark.r­ea­d.c­sv(­"­Folder path")
Using the read.csv() method you can also read multiple csv files, just pass all file names by separating comma as a path
Using nullValues option you can specify the string in a CSV to consider as null. For example, if you want to consider a date column with a value "­190­0-0­1-0­1" set null on DataFrame.
Used to partition the large dataset (DataF­rame) into smaller files based on one or multiple columns while writing to disk
df.wri­te.o­pt­ion­("he­ade­r",True) \ .parti­tio­nBy­("st­ate­") \ .mode(­"­ove­rwr­ite­") \ .csv("/­tmp­/zi­pco­des­-st­ate­")
PySpark splits the records based on the partition column and stores each partition data into a sub-di­rec­tory., If we have a total of 6 different states hence, it creates 6 direct­ories
df.wri­te.o­pt­ion­("he­ade­r",True) \ .parti­tio­nBy­("st­ate­"­,"ci­ty") \ .mode(­"­ove­rwr­ite­") \ .csv("/­tmp­/zi­pco­des­-st­ate­")
t creates a folder hierarchy for each partition; we have mentioned the first partition as state followed by city hence, it creates a city folder inside the state folder (one folder for each city in a state).


from pyspark.sql import functions as F
# Select Columns"firstName").show()"firstName","lastName") \

#  split multiple array column data into rows 
df2 =,explode(df.subjectandID))

#  Show all entries where age >24['age'] > 24).show() 

#  Show name and 0 or 1 depending on age >  or < than 30"Name", 
 F.when(df.age > 30, 1)
 .otherwise(0)) \

# Show firstName if in the given options

# Show firstName, and lastName if lastName is Smith."firstName","Smith"))  

# Like also excepts wildcard matches."firstName","%Sm"))  

# Show firstName, and TRUE if
 df.lastName \ lastName starts with Sm

Startswith - Endswith"firstName
 .startswith("Sm")) \

# Show last names ending in th"th"))\

# Return substrings of firstName
 Substring, 3) \ 
 .alias("name")) \

# Show  values where age is between 22 and 24, 24)) \ 

# Show all entries in firstName and age + 1["firstName"],df["age"]+ 1), .show()


Dealing with nulls
df=­ro­p(how = 'any', thresh = 2)
To drop null values we use the na function with the drop() attribute.
how: ‘any’ or ‘all’. If ‘any’, drop a row if it contains any nulls. If ‘all’, drop a row only if all its values are null
thresh: default None If specified, drop rows that have less than thresh non-null values. This overwrites the how parameter.
subset: optional optional list of column names to consider.
To fill nulls­il­l(50)
union() method of the DataFrame is used to merge two DataFr­ame’s of the same struct­ure­/sc­hema.
unionDF = df.uni­on(df2)
returns the new DataFrame with all rows from two Dataframes regardless of duplicate data.
use the use the distinct() function to return just one record when duplicate exists.() function to return just one record when duplicate exists.
use the distinct() function to return just one record when duplicate exists.

Creating a Session

import pyspark # importing the module
 # importing the SparkSession module
from pyspark.sql import SparkSession

 # creating a session
session = SparkSession.builder.appName('First App')

 # calling the session variable

Creating delta tables

# Define the input and output formats and paths and the table name.
read_format = 'delta'
write_format = 'delta'
load_path = '/databricks-datasets/learning-spark-v2/people/'
save_path = '/tmp/delta/people-10m'
table_name = 'default.people10m'

# Load the data from its source.
people = spark \
  .read \
  .format(read_format) \

# Write the data to its target.
people.write \
  .format(write_format) \

# Create the table.
spark.sql("CREATE TABLE " + table_name + " USING DELTA LOCATION '" + save_path + "'")


Data prepro­cessing

To select one or multiple columns the select() function works # selecting one column, column_2, .., column_N) 
# selecting many columns

To add a column the dataframe.withColumn() function 
takes two parameters

New column name to add
Existing column name to use for (not necessary if the 
new column has nothing to do with the existing column)

# adding columns in dataframe
data = data.withColumn('Age_after_3_y', data['Age']+3)

to change data type
You would also need cast() along with withColumn().
 The below statement changes the datatype from String
 to Integer for the salary column.


Change a value
Pass an existing column name as a first argument 
and a column as the value to be assigned as a second argument 


df.drop("salary") \

rename an existing column 

df.withColumnRenamed("gender","sex") \
Adding columns - df.wit­hCo­lum­n('­new­Col', newVal)
Changing data types - df.wit­hCo­lum­n("n­ewC­ol",­col­("Ol­dCo­l").c­as­t("N­ewD­T")).show()
Changing Values - df.wit­hCo­lum­n('­old­col', col("ol­dco­l") operation)

Dropping = withCo­lum­nRe­named

Renaming = withCo­lum­nRe­named

Sorting and Grouping

df.sor­t("c­ol", ascending = false)
Default sorting technique used by order by is ASC
df.gro­upb­y("c­ol").agg() / df.gro­upb­y("a­ge").co­unr()

Spark SQL

spark.s­ql­(select * from tablename)


No comments yet. Add yours below!

Add a Comment

Your Comment

Please enter your name.

    Please enter your email address

      Please enter your Comment.

          More Cheat Sheets by datamansam