Narrow vs Wide Transformations

Overview

Learn to distinguish between Narrow Transformations (which run locally on partitions) and Wide Transformations (which force network shuffles).

What You Will Learn

Detailed Concept Explanation

Spark transformations are categorized into narrow or wide based on their partition dependencies.

Narrow Transformations

• Behavior: Each input partition maps to at most one output partition. The data does not leave the executor node.
• Examples: map(), filter(), flatMap(), union(), select().
• Performance: Very fast, as operations run in parallel across partitions with zero network I/O.

Wide Transformations

• Behavior: Multiple input partitions map to multiple output partitions. Records with matching keys are routed across nodes. This network reorganization is called a Shuffle.
• Examples: groupBy(), join(), repartition(), distinct().
• Performance: Expensive due to network latency, disk serialization, and synchronization barriers.

Code Examples

Python (PySpark) Implementation

from pyspark.sql import SparkSession

spark = SparkSession.builder.appName("NarrowWide").getOrCreate()

df = spark.createDataFrame([("A", 1), ("B", 2)], ["key", "val"])

# Narrow: filter keeps data on same partition
narrow_df = df.filter(df.val > 1)

# Wide: groupByKey shuffles data across network
wide_df = df.groupBy("key").count()

Execution Plan Diagram (Python & Scala)

Scala Implementation

import org.apache.spark.sql.SparkSession

val spark = SparkSession.builder().appName("NarrowWideScala").getOrCreate()
import spark.implicits._

val df = Seq(("A", 1), ("B", 2)).toDF("key", "val")

val narrow = df.filter($"val" > 1)
val wide = df.groupBy("key").count()

Interview Perspective

Related Topics

• Shuffle
• Stages
• Repartition vs Coalesce