Apache Spark


Big data processing is a hot topic in the technical world. There are a large number of libraries and frameworks that are being used for regular big data processing. We will discuss one framework, the Apache Spark, which is a data processing engine for massively distributed data. We will also discuss one of its popular methods, the Spark Coalesce, which is used to reduce the partitions of the dataset, along with its example.

What is Apache Spark?

Apache Spark is a free and open-source data processing tool that uses basic programming structures to persist and analyze data in real-time across several clusters of computers.

Spark began as an AMPLab experiment at UC Berkeley in 2009. It is part of the BSD license which is released in 2010. Spark became an Apache project in 2013. In 2014, Databricks achieved a new world record by sorting large-scale datasets with Spark. Spark supports various programming languages like R, Python, Java, and Scala and is 100 times quicker than MapReduce in processing the data since it is done in the memory. It contains fewer lines of code, and for authentication, it uses a shared code. It can also operate on YARN, taking advantage of Kerberos’ capabilities. It’s based on Hadoop MapReduce and extends the MapReduce concept to effectively employ the new types of computations.

Spark’s primary advantage over Hadoop is utilizing an in-memory processing architecture. To use the distributed replication storage, Spark may operate on top of HDFS. Spark may be utilized in the same Hadoop cluster as MapReduce or as a standalone processing framework. YARN can also run Spark applications. Instead of using a local memory space for calculation, Spark employs an in-memory computing, allowing users to process the data in RAM and retrieve it quickly. Spark is not intended to replace Hadoop; it might be considered a compliment.

MapReduce and Spark are used in tandem: MapReduce handles batch processing while Spark handles real-time processing. The Spark code may be reused for batch processing, joining streams against historical data, and doing ad-hoc queries on the stream state. Spark includes streaming data tools, interactive/declarative searches, machine learning, as well as map and reduce.

What is the Spark COALESCE Method?

The COALESCE method is used to lower the number of partitions of the data set. Coalesce avoids full shuffle by shuffling the data using the Hash Partitioner (Default) and adjusts to the existing partitions rather than generating new ones. This means that it can only reduce the number of partitions. In Spark SQL, the coalesce regular method is a non-aggregate method. To reduce the quantity of the data moved, coalesce uses existing divisions. If all columns are null, the coalesce returns the first non-null value. Coalesce is quicker than Repartition because it avoids the entire shuffling, whereas Repartition does a full shuffling, which is time-consuming and expensive. At least one column is required for the coalescence, and all columns must be of the same or compatible kinds.

Example of Using the COALESCE Method

To test the Spark COALESCE function, use the following command:

testDF = sqlContext.createDataFrame([(1,1), (2,2), (None,3), (4,None)], ["id", "value"])


| id|value|


| 1| 1|

| 2| 2|

|null| 3|

| 4| null|


Import the required PySpark functions using the following command:

from pyspark.sql.types import FloatType

from pyspark.sql.functions import *

To create a new column with non-null values, apply the following command:

tmp = testDF.withColumn('col', coalesce(testDF['id'], testDF['value']))



| id|value|col|


| 1| 1| 1|

| 2| 2| 2|

|null| 3| 3|

| 4| null| 4|



We discussed about the Apache Spark, Spark SQL, and Spark SQL Coalesce method. We learned that it can be used to reduce the partitions of the data frame, along with one example. We also realized that the Coalesce method can only be used to reduce the number of partitions while partitioning can be used to decrease or increase the partitions.

About the author

Simran Kaur

Simran works as a technical writer. The graduate in MS Computer Science from the well known CS hub, aka Silicon Valley, is also an editor of the website. She enjoys writing about any tech topic, including programming, algorithms, cloud, data science, and AI. Travelling, sketching, and gardening are the hobbies that interest her.