DAG

Overview

Learn how Spark compiles your transformations into a Directed Acyclic Graph (DAG) to model execution pipelines and recover from node failures.

What You Will Learn

In this lesson, you will learn:

DAG Definition: Directed (ordered steps), Acyclic (no loops), Graph (vertices/edges).
Lineage: How DAG tracks dependencies.
Visualizing Plans: Inspecting DAGs using the .explain() method.

Detailed Concept Explanation

A DAG is the logical representation of your Spark query execution plan.

Directed: Each step in the pipeline points to the next step.
Acyclic: There are no loops or circular loops in the graph.
Graph: Made of vertices (transformations) and edges (dependencies/shuffles).

When an action is called, the DAG Scheduler divides the graph into execution stages based on shuffle boundaries, which are then split into parallel tasks sent to executors.

Code Examples

Python (PySpark) Implementation

python

from pyspark.sql import SparkSession

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

data = [("IT", 5000), ("HR", 4000), ("IT", 6000)]
df = spark.createDataFrame(data, ["dept", "salary"])

# DAG contains: Filter -> GroupBy -> Sum
result = df.filter(df.salary > 4000).groupBy("dept").sum("salary")

# Print the physical execution plan compiled from the DAG
result.explain()

Expected Output

text

== Physical Plan ==
AdaptiveSparkPlan isAdaptive=true
+- HashAggregate(keys=[dept#0], functions=[sum(salary#1L)])
   +- Exchange hashpartitioning(dept#0, 200)
      +- HashAggregate(keys=[dept#0], functions=[partial_sum(salary#1L)])
         +- Filter (isnotnull(salary#1L) AND (salary#1L > 4000))
            +- Scan ExistingRDD[dept#0,salary#1L]

Execution Plan Diagram (Python & Scala)

Execution Plan Diagram

SparkSession.builder

createDataFrame

filter(salary > 4000)

groupBy(dept).sum()

explain()

Scala Implementation

scala

import org.apache.spark.sql.SparkSession

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

val df = Seq(("IT", 5000), ("HR", 4000), ("IT", 6000)).toDF("dept", "salary")
val result = df.filter($"salary" > 4000).groupBy("dept").sum("salary")

result.explain()

Interview Perspective

What is a DAG in Apache Spark?

A Directed Acyclic Graph (DAG) is the logical execution plan constructed by Spark's DAG Scheduler when an action is triggered. It maps out the sequence of transformations (vertices) and the dependencies between them (edges), allowing Spark to optimize stages and recompute lost data partitions if worker nodes fail.