Self joins in Apache Spark

Introduction

A self join helps you compare rows within the same table or dataset. It lets you find relationships between data in one table by joining it to itself.

Finding pairs of employees who work in the same department.

Comparing sales records from the same dataset to find duplicates or related entries.

Finding customers who referred other customers in the same customer list.

Matching products that have similar attributes within the same product list.

Syntax

Apache Spark

df1 = df.alias('a')
df2 = df.alias('b')
joined_df = df1.join(df2, condition, 'join_type')

You create two aliases of the same DataFrame to join it with itself.

The condition defines how rows from the two aliases match.

Examples

This example finds pairs of employees in the same department but with different IDs.

Apache Spark

from pyspark.sql import SparkSession

spark = SparkSession.builder.getOrCreate()
data = [(1, 'Alice', 10), (2, 'Bob', 10), (3, 'Charlie', 20)]
df = spark.createDataFrame(data, ['id', 'name', 'dept'])

# Create aliases
df1 = df.alias('a')
df2 = df.alias('b')

# Join on same department but different employees
joined = df1.join(df2, (df1.dept == df2.dept) & (df1.id != df2.id))
joined.show()

This example joins the table to itself to find all employees in the same department, including themselves.

Apache Spark

df1 = df.alias('left')
df2 = df.alias('right')

# Find employees who share the same department
result = df1.join(df2, df1.dept == df2.dept, 'inner')
result.select('left.name', 'right.name', 'left.dept').show()

Sample Program

This program finds pairs of employees who work in the same department but are different people. It shows their IDs, names, and department.

Apache Spark

from pyspark.sql import SparkSession

spark = SparkSession.builder.appName('SelfJoinExample').getOrCreate()

# Sample data: employee id, name, department
employees = [
    (1, 'Alice', 10),
    (2, 'Bob', 10),
    (3, 'Charlie', 20),
    (4, 'David', 10),
    (5, 'Eve', 30)
]

# Create DataFrame
emp_df = spark.createDataFrame(employees, ['id', 'name', 'dept'])

# Create two aliases of the same DataFrame
emp1 = emp_df.alias('emp1')
emp2 = emp_df.alias('emp2')

# Join on same department but different employee ids
joined_df = emp1.join(emp2, (emp1.dept == emp2.dept) & (emp1.id != emp2.id), 'inner')

# Select columns to show pairs of employees in the same department
result = joined_df.select(emp1.id.alias('emp1_id'), emp1.name.alias('emp1_name'), emp2.id.alias('emp2_id'), emp2.name.alias('emp2_name'), emp1.dept)

result.show()

Important Notes

Always use aliases to avoid confusion when joining a DataFrame to itself.

Make sure to exclude matching the same row with itself if that is not desired, by adding a condition like id != id.

Self joins can create many rows if the matching condition is broad, so be careful with large datasets.

Summary

Self joins let you compare rows within the same dataset.

Use aliases to join a DataFrame to itself.

Filter carefully to avoid matching rows to themselves unless needed.

Practice

1.

What is the main purpose of a self join in Apache Spark?

easy

A. To join a DataFrame with itself to compare rows within the same dataset

B. To join two different DataFrames based on a common key

C. To merge two DataFrames vertically by stacking rows

D. To filter rows based on a condition

Solution

1. Step 1: Understand the concept of self join

   A self join means joining a DataFrame to itself to compare or relate rows within the same dataset.

2. Step 2: Differentiate from other joins

   Other joins combine different DataFrames or filter rows, but self join specifically uses the same DataFrame twice.

3. Final Answer:

   To join a DataFrame with itself to compare rows within the same dataset -> Option A

4. Quick Check:

   Self join = join DataFrame to itself [OK]

Hint: Self join means joining the same table twice [OK]

Common Mistakes:

• Confusing self join with joining different DataFrames
• Thinking self join stacks rows vertically
• Assuming self join is just filtering rows

2.

Which of the following is the correct way to create aliases for a self join in Spark using DataFrame API?

df1 = df.alias('a')
df2 = df.alias('b')
joined = df1.join(df2, ...)

easy

A. df1 = df.alias('a'); df2 = df.alias('b'); joined = df1.join(df2, 'id')

B. df1 = df; df2 = df; joined = df1.join(df2, 'id')

C. df1 = df.as('a'); df2 = df.as('b'); joined = df1.join(df2, 'id')

D. df1 = df.copy(); df2 = df.copy(); joined = df1.join(df2, 'id')

Solution

1. Step 1: Use alias() method for self join

   In Spark, to join a DataFrame to itself, you create aliases using alias() to distinguish the two references.

2. Step 2: Check the syntax correctness

   df1 = df.alias('a'); df2 = df.alias('b'); joined = df1.join(df2, 'id') uses alias() correctly and joins on a column 'id'. df1 = df.as('a'); df2 = df.as('b'); joined = df1.join(df2, 'id') uses as() which is not a valid DataFrame method in Spark.

3. Final Answer:

   df1 = df.alias('a'); df2 = df.alias('b'); joined = df1.join(df2, 'id') -> Option A

4. Quick Check:

   Use alias() for self join [OK]

Hint: Use alias() to rename DataFrame before self join [OK]

Common Mistakes:

• Using as() instead of alias()
• Not creating aliases and causing ambiguous columns
• Copying DataFrame instead of aliasing

3.

Given the following Spark code, what will be the output count?

from pyspark.sql import SparkSession
spark = SparkSession.builder.getOrCreate()
data = [(1, 'A'), (2, 'B'), (3, 'A')]
df = spark.createDataFrame(data, ['id', 'group'])
df1 = df.alias('df1')
df2 = df.alias('df2')
joined = df1.join(df2, (df1.group == df2.group) & (df1.id != df2.id))
print(joined.count())

medium

A. 2

B. 4

C. 3

D. 6

Solution

1. Step 1: Understand the data and join condition

   Data has 3 rows: (1, 'A'), (2, 'B'), (3, 'A'). The join matches rows with the same group but different ids.

2. Step 2: Count matching pairs

   Group 'A' has ids 1 and 3, so pairs are (1,3) and (3,1) = 2 matches. Group 'B' has only one id, so no match. Total matches = 2 rows. However, since the join is a cross join on the condition, each pair appears twice (once as (1,3) and once as (3,1)), so total count is 4.

3. Final Answer:

   4 -> Option B

4. Quick Check:

   Count matching pairs with condition = 4 [OK]

Hint: Count pairs with same group but different ids [OK]

Common Mistakes:

• Counting only one direction of pairs
• Including rows where id equals
• Ignoring symmetric pairs

4.

Identify the error in this Spark self join code snippet:

df1 = df.alias('a')
df2 = df.alias('b')
joined = df1.join(df2, df1.id == df2.id)
joined.show()

What is the problem?

medium

A. Syntax error in join condition

B. Missing alias() method for DataFrame

C. The join condition causes each row to join with itself, no filtering

D. DataFrame df is not defined

Solution

1. Step 1: Analyze the join condition

   The join condition is df1.id == df2.id, which matches rows with the same id, including the same row joined to itself.

2. Step 2: Identify missing filter to exclude self matches

   There is no condition to exclude rows where the two aliases refer to the same row, so the join includes self matches.

3. Final Answer:

   The join condition causes each row to join with itself, no filtering -> Option C

4. Quick Check:

   Join condition includes self matches [OK]

Hint: Add filter to exclude self matches in self join [OK]

Common Mistakes:

• Not excluding rows where both aliases are same
• Assuming alias() is missing
• Confusing syntax error with logic error

5.

You have a DataFrame df with columns employee_id, manager_id, and department. You want to find pairs of employees who work in the same department and share the same manager but are different employees. Which self join condition is correct?

hard

A. df1.employee_id == df2.employee_id AND df1.manager_id == df2.manager_id

B. df1.employee_id == df2.manager_id AND df1.department == df2.department

C. df1.manager_id != df2.manager_id AND df1.department == df2.department

D. df1.department == df2.department AND df1.manager_id == df2.manager_id AND df1.employee_id != df2.employee_id

Solution

1. Step 1: Understand the requirement

   We want pairs of employees in the same department, with the same manager, but different employee IDs.

2. Step 2: Translate to join condition

   The join condition must check equality of department and manager_id, and inequality of employee_id to avoid self pairs.

3. Final Answer:

   df1.department == df2.department AND df1.manager_id == df2.manager_id AND df1.employee_id != df2.employee_id -> Option D

4. Quick Check:

   Same department & manager, different employees [OK]

Hint: Match same manager and department, exclude same employee [OK]

Common Mistakes:

• Matching employee_id to manager_id incorrectly
• Not excluding same employee pairs
• Checking for different managers instead of same