Why data format affects performance in Apache Spark - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why does Parquet format improve query speed compared to CSV?

In Apache Spark, you have two datasets: one stored as CSV files and another as Parquet files. Both contain the same data. Why does reading Parquet files usually result in faster query performance?

A Parquet files are compressed and store data in a columnar format, allowing Spark to read only needed columns, reducing disk I/O.

B CSV files are binary and harder to parse, so Spark spends more time decoding them.

C Parquet files are stored in memory by default, so reading them is instant.

D CSV files contain metadata that slows down Spark's query optimizer.

💡 Hint

Think about how data is stored and how Spark reads only parts of the data it needs.

Parquet is a columnar storage format that compresses data and stores metadata, enabling Spark to skip reading unnecessary columns. CSV is row-based and uncompressed, so Spark reads all data even if only some columns are needed.

❓ data_output

intermediate

Comparing file sizes of CSV and Parquet

You save the same DataFrame in Spark as CSV and Parquet formats. Which option correctly shows the typical size difference between the two files?

Apache Spark

df.write.csv('data_csv')
df.write.parquet('data_parquet')

# After saving, check file sizes of 'data_csv' and 'data_parquet' folders

A The Parquet folder size is larger because it stores extra metadata.

B The CSV folder size is smaller because CSV is a plain text format.

C Both folders have the same size because they store the same data.

D The Parquet folder size is usually smaller than the CSV folder size due to compression.

💡 Hint

Consider how compression affects file size.

Parquet files use compression and efficient encoding, so they usually take less disk space than CSV files, which are plain text and uncompressed.

❓ Predict Output

advanced

What is the output of this Spark code reading JSON vs Parquet?

Consider this Spark code snippet:

df_json = spark.read.json('data.json')
df_parquet = spark.read.parquet('data.parquet')

print(df_json.count())
print(df_parquet.count())

Assuming both files contain the same data, what will be the output?

Apache Spark

df_json = spark.read.json('data.json')
df_parquet = spark.read.parquet('data.parquet')

print(df_json.count())
print(df_parquet.count())

A df_parquet.count() prints zero because Parquet files need special schema.

B Both print the same number of rows because the data is identical.

C df_json.count() is slower and prints fewer rows due to parsing errors.

D df_json.count() raises an error because JSON is not supported.

💡 Hint

Think about how Spark reads data formats and counts rows.

Both JSON and Parquet readers load the data and count rows. If the data is the same and valid, counts will match. Performance differs but output count is the same.

❓ visualization

advanced

Which plot best shows performance difference by data format?

You run the same query on datasets stored as CSV, JSON, and Parquet. You want to visualize query execution time for each format. Which plot type best shows this comparison clearly?

A A line chart showing execution time over multiple runs for one format only.

B A pie chart showing percentage of total execution time per format.

C A bar chart with data formats on the x-axis and execution time on the y-axis.

D A scatter plot with random points representing execution times.

💡 Hint

Think about comparing categories with numeric values.

A bar chart clearly compares execution times across different data formats side by side. Pie charts and scatter plots are less clear for this purpose.

🔧 Debug

expert

Why does this Spark job run slowly with CSV but fast with Parquet?

You have this Spark code:

df = spark.read.csv('data.csv', header=True)
df_filtered = df.filter(df['age'] > 30)
df_filtered.show()

The same filter on a Parquet file runs much faster. What is the main reason for this difference?

A CSV format lacks column statistics and schema, so Spark cannot push down filters and must scan all data.

B Parquet files are stored in memory, so filtering is instant.

C CSV files are encrypted by default, slowing down reading.

D Spark applies filter only on Parquet files, ignoring filters on CSV.

💡 Hint

Consider how Spark optimizes queries differently for formats.

Parquet stores metadata and column statistics enabling filter pushdown, so Spark reads only relevant data. CSV lacks this, so Spark scans entire files, making it slower.