Partition tuning (repartition vs coalesce) in Apache Spark

Introduction

Partition tuning helps your data run faster by changing how data is split across computers. Repartition and coalesce are two ways to do this.

When you want to increase the number of partitions to spread data more evenly.

When you want to reduce the number of partitions to save resources.

When your data is unevenly spread and causes slow processing.

When you want to prepare data for a shuffle operation like join or aggregation.

When you want to avoid a full shuffle to save time and resources.

Syntax

Apache Spark

DataFrame.repartition(numPartitions)
DataFrame.coalesce(numPartitions)

repartition reshuffles all data and can increase or decrease partitions.

coalesce reduces partitions without full shuffle, faster but only for decreasing partitions.

Examples

This creates 10 partitions by fully reshuffling the data.

Apache Spark

df_repart = df.repartition(10)

This reduces partitions to 2 without a full shuffle, faster but only reduces partitions.

Apache Spark

df_coal = df.coalesce(2)

This reshuffles data into 5 partitions, even if the original had 5 partitions.

Apache Spark

df_same = df.repartition(5)

Sample Program

This code creates a simple DataFrame with 100 numbers. It shows how to check the number of partitions, then repartitions to 10 partitions with a full shuffle, and finally coalesces to 5 partitions without a full shuffle.

Apache Spark

from pyspark.sql import SparkSession

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

# Create a DataFrame with 100 numbers
numbers = spark.range(0, 100)

# Check original number of partitions
original_partitions = numbers.rdd.getNumPartitions()

# Repartition to 10 partitions (full shuffle)
numbers_repart = numbers.repartition(10)
repart_partitions = numbers_repart.rdd.getNumPartitions()

# Coalesce to 5 partitions (no full shuffle)
numbers_coal = numbers_repart.coalesce(5)
coal_partitions = numbers_coal.rdd.getNumPartitions()

print(f'Original partitions: {original_partitions}')
print(f'After repartition to 10: {repart_partitions}')
print(f'After coalesce to 5: {coal_partitions}')

spark.stop()

Important Notes

Use repartition when you want to increase or evenly distribute partitions but expect a full shuffle.

Use coalesce to reduce partitions quickly without a full shuffle, but only for decreasing partitions.

Too many partitions can slow down processing; too few can cause data to be unevenly processed.

Summary

Repartition reshuffles data and can increase or decrease partitions.

Coalesce reduces partitions without full shuffle, faster but only decreases partitions.

Choose the right method based on whether you want to increase or decrease partitions and how much shuffle you can afford.

Practice

1. Which statement best describes the difference between repartition and coalesce in Apache Spark?

easy

A. coalesce only increases partitions with shuffle; repartition only decreases partitions without shuffle.

B. coalesce can increase or decrease partitions with shuffle; repartition only decreases partitions without shuffle.

C. repartition only increases partitions without shuffle; coalesce only decreases partitions with shuffle.

D. repartition can increase or decrease partitions with shuffle; coalesce only decreases partitions without shuffle.

Solution

1. Step 1: Understand repartition behavior

   repartition reshuffles data and can increase or decrease the number of partitions, causing a full shuffle.

2. Step 2: Understand coalesce behavior

   coalesce reduces the number of partitions without a full shuffle, making it faster but it cannot increase partitions.

3. Final Answer:

   repartition can increase or decrease partitions with shuffle; coalesce only decreases partitions without shuffle. -> Option D

4. Quick Check:

   repartition = shuffle & flexible, coalesce = no shuffle & reduce [OK]

Hint: Repartition reshuffles; coalesce only reduces partitions [OK]

Common Mistakes:

• Thinking coalesce can increase partitions
• Assuming repartition does not shuffle
• Confusing which method is faster

2. Which of the following is the correct syntax to reduce the number of partitions of a DataFrame df to 5 using coalesce?

easy

A. df.repartition(5)

B. df.coalesce(5)

C. df.coalesce().numPartitions(5)

D. df.reducePartitions(5)

Solution

1. Step 1: Recall coalesce syntax

   The correct method to reduce partitions is df.coalesce(numPartitions).

2. Step 2: Identify correct usage

   df.coalesce(5) uses df.coalesce(5), which correctly reduces partitions to 5.

3. Final Answer:

   df.coalesce(5) -> Option B

4. Quick Check:

   coalesce syntax = df.coalesce(number) [OK]

Hint: Use df.coalesce(number) to reduce partitions [OK]

Common Mistakes:

• Using repartition instead of coalesce to reduce partitions
• Calling non-existent methods like reducePartitions
• Incorrect method chaining syntax

3. Given the following Spark code, what will be the number of partitions after execution?

val df = spark.range(1000).repartition(10)
val df2 = df.coalesce(4)

medium

A. 4 partitions

B. 10 partitions

C. 14 partitions

D. 1000 partitions

Solution

1. Step 1: Analyze initial repartition

   df is repartitioned to 10 partitions explicitly.

2. Step 2: Apply coalesce

   df2 uses coalesce(4) which reduces partitions from 10 to 4 without shuffle.

3. Final Answer:

   4 partitions -> Option A

4. Quick Check:

   repartition to 10, then coalesce to 4 = 4 partitions [OK]

Hint: Coalesce reduces partitions to given number [OK]

Common Mistakes:

• Assuming partitions add up (10 + 4 = 14)
• Thinking coalesce increases partitions
• Confusing initial partition count with final

4. You want to increase the number of partitions of a DataFrame df from 4 to 8. You try df.coalesce(8) but it does not increase partitions. What is the issue?

medium

A. You need to call df.coalesce(8).shuffle() to increase partitions.

B. coalesce requires shuffle to increase partitions.

C. coalesce cannot increase partitions; use repartition instead.

D. The DataFrame must be cached before increasing partitions.

Solution

1. Step 1: Understand coalesce limitations

   coalesce only reduces partitions and does not increase them.

2. Step 2: Choose correct method to increase partitions

   To increase partitions, repartition must be used as it performs a shuffle.

3. Final Answer:

   coalesce cannot increase partitions; use repartition instead. -> Option C

4. Quick Check:

   coalesce only reduces, repartition reshuffles [OK]

Hint: Use repartition to increase partitions, not coalesce [OK]

Common Mistakes:

• Trying to increase partitions with coalesce
• Assuming coalesce triggers shuffle
• Believing caching affects partition count

5. You have a large DataFrame with 100 partitions. You want to reduce partitions to 10 for faster processing but want to avoid a full shuffle to save time. Which method should you use and why?

hard

A. Use df.coalesce(10) because it reduces partitions without a full shuffle.

B. Use df.repartition(10) because it reduces partitions without shuffle.

C. Use df.repartition(10) because it avoids shuffle and reduces partitions.

D. Use df.coalesce(10) because it increases partitions with shuffle.

Solution

1. Step 1: Identify partition reduction without shuffle

   coalesce reduces partitions without a full shuffle, making it faster.

2. Step 2: Compare with repartition

   repartition always causes a full shuffle, which is slower but can increase or decrease partitions.

3. Final Answer:

   Use df.coalesce(10) because it reduces partitions without a full shuffle. -> Option A

4. Quick Check:

   Reduce partitions fast = coalesce without shuffle [OK]

Hint: Coalesce reduces partitions fast without shuffle [OK]

Common Mistakes:

• Using repartition and expecting no shuffle
• Thinking coalesce increases partitions
• Confusing shuffle behavior of methods