Delta Lake introduction in Apache Spark - Step-by-Step Execution

Concept Flow - Delta Lake introduction

Start with Raw Data

↓

Write Data as Delta Table

↓

Perform Transactions: Insert, Update, Delete

↓

Delta Lake Manages ACID & Versioning

↓

Read Data with Time Travel or Latest Snapshot

↓

Use Data for Analytics or ML

Delta Lake stores data with ACID transactions and version control, allowing safe updates and easy data reads.

Execution Sample

Apache Spark

from delta.tables import DeltaTable

# Write data as Delta
spark.range(5).write.format("delta").save("/tmp/delta-table")

# Read Delta Table
deltaTable = DeltaTable.forPath(spark, "/tmp/delta-table")

This code writes a simple range of numbers as a Delta table and then reads it back.

Execution Table

Step	Action	Input/Condition	Result/Output
1	Create DataFrame	spark.range(5)	DataFrame with numbers 0 to 4
2	Write DataFrame as Delta	format='delta', path='/tmp/delta-table'	Delta table created at path
3	Read Delta Table	DeltaTable.forPath(spark, '/tmp/delta-table')	DeltaTable object referencing saved data
4	Perform Update	deltaTable.update(condition="id == 3", set={"id": 30})	Row with id=3 updated to id=30
5	Read Latest Data	deltaTable.toDF().show()	DataFrame shows updated data with id=30
6	Time Travel Read	spark.read.format('delta').option('versionAsOf', 0).load('/tmp/delta-table')	DataFrame shows original data before update
7	Exit	No more actions	End of demonstration

💡 All steps executed to show Delta Lake write, update, read, and time travel

Variable Tracker

Variable	Start	After Step 1	After Step 2	After Step 3	After Step 4	After Step 5	After Step 6	Final
DataFrame	None	Range(0-4)	Range(0-4)	Range(0-4)	Range with id=30 updated	Range with id=30 updated	Range(0-4)	Range(0-4) or updated depending on read
deltaTable	None	None	None	DeltaTable object created	DeltaTable object	DeltaTable object	DeltaTable object	DeltaTable object

Key Moments - 2 Insights

Why does the data show old values when using time travel after an update?

Because Delta Lake keeps versions of data, time travel reads a previous version (see execution_table step 6) showing data before the update.

What happens if you try to update data without Delta Lake?

Without Delta Lake, updates can cause data corruption or partial writes. Delta Lake ensures ACID transactions (see concept_flow step 'Delta Lake Manages ACID & Versioning').

Visual Quiz - 3 Questions

Look at the execution table, what is the result after step 4 (Perform Update)?

A All rows are deleted

B New row with id=30 is added

C Row with id=3 is changed to id=30

D No change to data

💡 Hint

Check the 'Result/Output' column in execution_table row for step 4

At which step does the DeltaTable object get created?

A Step 2

B Step 3

C Step 1

D Step 5

💡 Hint

Look at the 'Action' and 'Result/Output' columns in execution_table for when DeltaTable.forPath is called

If you skip step 2 (writing data as Delta), what happens when you try step 3?

A Error because no Delta table exists at path

B DeltaTable object is created with empty data

C Data is read from a CSV file instead

D Data is automatically created

💡 Hint

DeltaTable.forPath requires existing Delta data, see concept_flow step 'Write Data as Delta Table'

Concept Snapshot

Delta Lake stores data as tables with ACID transactions.
You write data using format 'delta' and a path.
You can update, delete, and read data safely.
Delta Lake keeps versions for time travel reads.
Use DeltaTable API to manage data programmatically.

Full Transcript

Delta Lake is a storage layer that adds reliability to data lakes. It allows you to write data as Delta tables, which support safe updates and deletes with ACID transactions. You start by writing data in Delta format to a path. Then you can read it back using the DeltaTable API. Delta Lake keeps versions of the data, so you can read previous snapshots using time travel. This example showed writing a range of numbers, updating one row, reading the latest data, and reading an older version. Delta Lake helps keep data consistent and easy to manage for analytics and machine learning.

Practice

1. What is the main purpose of Delta Lake in Apache Spark?

easy

A. To provide reliable data storage with ACID transactions

B. To speed up Spark job execution by caching data

C. To visualize data using charts and graphs

D. To replace Spark SQL with a new query language

Solution

1. Step 1: Understand Delta Lake's role

   Delta Lake adds reliability by supporting ACID transactions on big data.

2. Step 2: Compare options

   Speeding up jobs, visualization, or replacing Spark SQL are not Delta Lake's main goals.

3. Final Answer:

   To provide reliable data storage with ACID transactions -> Option A

4. Quick Check:

   Delta Lake = ACID transactions [OK]

Hint: Delta Lake ensures data correctness with ACID [OK]

Common Mistakes:

• Confusing Delta Lake with caching or visualization tools
• Thinking Delta Lake replaces Spark SQL
• Assuming Delta Lake only speeds up queries

2. Which of the following is the correct way to create a Delta table in Spark using PySpark?

easy

A. df.write.format('delta').save('/path/to/delta-table')

B. df.saveAsTable('delta', '/path/to/delta-table')

C. df.write.save('/path/to/delta-table', format='delta')

D. df.createDeltaTable('/path/to/delta-table')

Solution

1. Step 1: Recall Delta table creation syntax

   In PySpark, to save a DataFrame as a Delta table, use df.write.format('delta').save(path).

2. Step 2: Check each option

   df.write.format('delta').save('/path/to/delta-table') matches the correct syntax; others are invalid or incorrect method calls.

3. Final Answer:

   df.write.format('delta').save('/path/to/delta-table') -> Option A

4. Quick Check:

   Use write.format('delta').save() [OK]

Hint: Use write.format('delta').save() to create Delta tables [OK]

Common Mistakes:

• Using saveAsTable with wrong parameters
• Passing format as a keyword argument to save()
• Assuming createDeltaTable is a valid method

3. Given the following PySpark code, what will be the output count after the delete operation?

from delta.tables import DeltaTable

spark.sql("""CREATE TABLE IF NOT EXISTS events (id INT, event STRING) USING DELTA""")
spark.sql("INSERT INTO events VALUES (1, 'start'), (2, 'stop'), (3, 'pause')")
deltaTable = DeltaTable.forName(spark, 'events')
deltaTable.delete("id = 2")
count = spark.sql("SELECT COUNT(*) FROM events").collect()[0][0]
print(count)

medium

A. 3

B. 2

C. 1

D. 0

Solution

1. Step 1: Understand initial data

   The table 'events' has 3 rows with ids 1, 2, and 3.

2. Step 2: Analyze delete operation

   Deleting rows where id = 2 removes one row, leaving 2 rows.

3. Final Answer:

   2 -> Option B

4. Quick Check:

   3 rows - 1 deleted = 2 rows [OK]

Hint: Delete removes matching rows; count reflects remaining [OK]

Common Mistakes:

• Assuming delete removes all rows
• Forgetting to filter by condition in delete
• Confusing insert and delete effects

4. Identify the error in the following PySpark code snippet that tries to update a Delta table:

from delta.tables import DeltaTable

deltaTable = DeltaTable.forPath(spark, '/delta/events')
deltaTable.update(condition="id == 1", set={"event": 'resume'})

medium

A. The update method requires a SQL expression string, not a dictionary

B. The condition should be passed as a SQL expression, not a string

C. The set parameter must use column objects, not strings

D. The update method requires the condition parameter to be a string expression, and set to a dictionary with column names as keys and expressions as values

Solution

1. Step 1: Review update method signature

   DeltaTable.update expects condition as a string expression and set as a dictionary mapping column names to SQL expressions.

2. Step 2: Check given code

   Condition is correct string expression. set uses dict with column name key, but value 'resume' must be SQL expression like "'resume'".

3. Step 3: Identify missing quotes in code snippet

   The value in set lacks quotes for string literal, violating the requirement for expressions as values described in The update method requires the condition parameter to be a string expression, and set to a dictionary with column names as keys and expressions as values.

4. Final Answer:

   The update method requires the condition parameter to be a string expression, and set to a dictionary with column names as keys and expressions as values -> Option D

5. Quick Check:

   Update needs condition string and set dict [OK]

Hint: Update needs condition string and set dict with column keys [OK]

Common Mistakes:

• Passing condition as non-string
• Using incorrect types in set dictionary
• Not quoting string literals in set expressions

5. You have a Delta table with user transactions. You want to keep only the latest transaction per user and remove older duplicates. Which Delta Lake feature helps you do this efficiently?

hard

A. Delta Lake's VACUUM to clean old files

B. Delta Lake's Time Travel to query past versions

C. Delta Lake's MERGE INTO to upsert and deduplicate data

D. Delta Lake's OPTIMIZE to compact small files

Solution

1. Step 1: Understand the goal

   We want to keep only the latest transaction per user, removing duplicates.

2. Step 2: Identify Delta Lake features

   MERGE INTO allows conditional updates and inserts, perfect for upserting and deduplication.

3. Step 3: Exclude other options

   Time Travel queries past data, VACUUM cleans files, OPTIMIZE compacts files, none directly remove duplicates.

4. Final Answer:

   Delta Lake's MERGE INTO to upsert and deduplicate data -> Option C

5. Quick Check:

   Use MERGE INTO for upsert and deduplication [OK]

Hint: Use MERGE INTO to update and remove duplicates efficiently [OK]

Common Mistakes:

• Confusing Time Travel with deduplication
• Thinking VACUUM removes duplicates
• Assuming OPTIMIZE deletes old data