What is the difference between a transformation and an action in Apache Spark?
Apache Spark

What is the difference between a transformation and an action in Apache Spark?

What is an action in Apache Spark? What do you understand as transformations in Apache Spark?
Bartosz Mikulski
Bartosz Mikulski – 2 min read

The concepts of actions and transformations in Apache Spark are not something we think about every day while using Spark, but occasionally it is good to know the distinction.

Table of Contents

  1. What is a transformation?
  2. What is an action?
  3. How to tell the difference

First of all, we may be asked that question during a job interview. It is a good question because it helps spot people who did not bother reading the documentation of the tool they are using. Fortunately, after reading this article, you will know the difference.

In addition to that, we may need to know the difference to explain why the code we are reviewing is too slow. For example, some data engineers have strange ideas like calculating counts in the middle of a Spark job to log the number of rows or store them as a metric. While a well-placed count may help to debug and tremendously speed up problem-solving, counting the number of rows in every other line of code is massive overkill. The difference between a transformation and an action helps us explain why doing it is a significant bottleneck.

What is a transformation?

A transformation is every Spark operation that returns a DataFrame, Dataset, or an RDD. When we build a chain of transformations, we add building blocks to the Spark job, but no data gets processed. That is possible because transformations are lazy executed. Spark will calculate the value when it is necessary.

Of course, this also means that Spark needs to recompute the values when we re-use the same transformations. We can avoid that by using the persist or cache functions.

What is an action?

Actions, on the other hand, are not lazily executed. When we put an action in the code and Spark reaches that line of code when running the job, it will have to perform all of the transformations that lead to that action to produce a value.

Producing value is the key concept here. While transformations return one of the Spark data types, actions return a count of elements (for example, the count function), a list of them (collect, take, etc.), or store the data in external storage (write, saveAsTextFile, and others).

How to tell the difference

When we look at the Spark API, we can easily spot the difference between transformations and actions. If a function returns a DataFrame, Dataset, or RDD, it is a transformation. If it returns anything else or does not return a value at all (or returns Unit in the case of Scala API), it is an action.

Older post

Use LatestOnlyOperator to skip some tasks while running a backfill in Airflow

How to skip some tasks when backfilling a DAG in the past

Newer post

How to postpone Airflow DAG until files get uploaded into an S3 bucket

How to use Airflow sensors to detect that files have been uploaded into an S3 bucket

Related Posts

Data Engineering Apache Spark

What is shuffling in Apache Spark, and when does it happen?

When does an Apache Spark cluster perform the shuffle operation?
Published on: 20 May 2022

Apache Spark

How to measure Spark performance and gather metrics about written data

How to track Spark metrics in AWS CloudWatch
Published on: 05 Mar 2021

Apache Spark

How to combine two DataFrames with no common columns in Apache Spark

Use full outer join to combine two Apache Spark DataFrames with no common columns
Published on: 06 Dec 2020

Are you looking for an experienced AI consultant? Do you need assistance with your RAG or Agentic Workflow?
Schedule a call, send me a message on LinkedIn. Schedule a call or send me a message on LinkedIn

>