Read phenomena (dirty reads, phantom reads) in SQL

Introduction

Read phenomena happen when database transactions see data changes in unexpected ways. Understanding them helps keep data correct and reliable.

When multiple people or programs change data at the same time.

When you want to know why your query shows different results if run twice quickly.

When you want to avoid errors caused by reading unfinished changes.

When designing how your database handles many users safely.

When debugging strange data results in your application.

Syntax

SQL

-- No specific syntax, but these phenomena happen during transactions
BEGIN TRANSACTION;
-- Read or write queries here
COMMIT;

Read phenomena occur during transactions when isolation levels are low.

You control them by setting transaction isolation levels like READ UNCOMMITTED, READ COMMITTED, REPEATABLE READ, or SERIALIZABLE.

Examples

Transaction 2 reads data that Transaction 1 changed but did not save permanently yet.

SQL

-- Dirty Read example
-- Transaction 1 updates a row but does not commit
BEGIN TRANSACTION;
UPDATE accounts SET balance = 500 WHERE id = 1;
-- Transaction 2 reads the updated balance before commit
SELECT balance FROM accounts WHERE id = 1;
-- If Transaction 1 rolls back, Transaction 2 read wrong data (dirty read)

Transaction 1 sees new rows appear between reads because another transaction added them.

SQL

-- Phantom Read example
-- Transaction 1 reads rows matching a condition
BEGIN TRANSACTION;
SELECT * FROM orders WHERE status = 'pending';
-- Transaction 2 inserts a new 'pending' order and commits
BEGIN TRANSACTION;
INSERT INTO orders (id, status) VALUES (101, 'pending');
COMMIT;
-- Transaction 1 reads again and sees new row (phantom)

Sample Program

This shows a dirty read if Transaction 2 reads the updated price before Transaction 1 commits or rolls back.

SQL

-- Setup sample table
CREATE TABLE products (id INT PRIMARY KEY, name VARCHAR(20), price INT);
INSERT INTO products VALUES (1, 'Pen', 10), (2, 'Book', 20);

-- Transaction 1 starts
BEGIN TRANSACTION;
UPDATE products SET price = 15 WHERE id = 1;
-- Transaction 2 reads before Transaction 1 commits
SELECT price FROM products WHERE id = 1;
-- Transaction 1 rolls back
ROLLBACK;

Important Notes

Dirty reads happen when you read data that another transaction changed but not yet saved.

Phantom reads happen when new rows appear or disappear between reads in the same transaction.

Using higher isolation levels can prevent these phenomena but may slow down the database.

Summary

Read phenomena cause unexpected data views during concurrent transactions.

Dirty reads read uncommitted changes; phantom reads see new or missing rows between reads.

Control these by choosing the right transaction isolation level.

Practice

1. Which of the following best describes a dirty read in database transactions?

easy

A. Reading data that has been modified but not yet committed by another transaction.

B. Seeing new rows appear in a query result after re-running the same query.

C. Reading only committed data and never uncommitted changes.

D. Locking rows to prevent other transactions from reading them.

Solution

1. Step 1: Understand dirty read definition

   A dirty read happens when a transaction reads data that another transaction has changed but not committed yet.

2. Step 2: Compare options to definition

   Reading data that has been modified but not yet committed by another transaction. matches this exactly. Seeing new rows appear in a query result after re-running the same query. describes phantom reads, C describes a safe read, and D is about locking, not dirty reads.

3. Final Answer:

   Reading data that has been modified but not yet committed by another transaction. -> Option A

4. Quick Check:

   Dirty read = reading uncommitted data [OK]

Hint: Dirty read = reading uncommitted changes [OK]

Common Mistakes:

• Confusing dirty read with phantom read
• Thinking dirty read only reads committed data
• Mixing locking with read phenomena

2. Which SQL transaction isolation level allows dirty reads?

easy

A. SERIALIZABLE

B. REPEATABLE READ

C. READ COMMITTED

D. READ UNCOMMITTED

Solution

1. Step 1: Recall isolation levels and dirty reads

   Only the READ UNCOMMITTED isolation level allows dirty reads, meaning it can read uncommitted changes.

2. Step 2: Match options to isolation levels

   READ COMMITTED and higher levels prevent dirty reads. SERIALIZABLE is the strictest. So, READ UNCOMMITTED is correct.

3. Final Answer:

   READ UNCOMMITTED -> Option D

4. Quick Check:

   Dirty reads allowed only in READ UNCOMMITTED [OK]

Hint: Dirty reads only happen at READ UNCOMMITTED level [OK]

Common Mistakes:

• Choosing READ COMMITTED as allowing dirty reads
• Confusing phantom reads with dirty reads
• Not knowing isolation level hierarchy

3. Consider two transactions running concurrently on a table Orders. Transaction 1 reads all rows where status = 'pending'. Transaction 2 inserts a new row with status = 'pending' and commits. If Transaction 1 repeats the read, what phenomenon might it observe?

medium

A. Dirty read

B. Non-repeatable read

C. Phantom read

D. Lost update

Solution

1. Step 1: Analyze the scenario with inserted rows

   Transaction 1 reads rows with status 'pending'. Transaction 2 inserts a new 'pending' row and commits.

2. Step 2: Identify the phenomenon on repeated read

   When Transaction 1 reads again, it sees the new row that wasn't there before. This is a phantom read, where new rows appear between reads.

3. Final Answer:

   Phantom read -> Option C

4. Quick Check:

   New rows appearing = Phantom read [OK]

Hint: New rows between reads = phantom read [OK]

Common Mistakes:

• Confusing phantom read with dirty read
• Thinking non-repeatable read applies to new rows
• Mixing lost update with read phenomena

4. A developer wrote this SQL transaction code but observes inconsistent reads. Identify the error related to read phenomena:

BEGIN TRANSACTION;
SELECT * FROM Products WHERE price > 100;
-- Another transaction inserts a new product with price 150 and commits
SELECT * FROM Products WHERE price > 100;
COMMIT;

What is the likely cause?

medium

A. Using READ UNCOMMITTED isolation level causing dirty reads

B. Using READ COMMITTED isolation level causing phantom reads

C. Using SERIALIZABLE isolation level causing lost updates

D. Using REPEATABLE READ isolation level causing non-repeatable reads

Solution

1. Step 1: Understand the scenario with repeated SELECTs

   The transaction reads products with price > 100 twice. Between reads, another transaction inserts a new matching row and commits.

2. Step 2: Identify the read phenomenon

   Because the isolation level is likely READ COMMITTED, the second read sees the new row, causing a phantom read.

3. Final Answer:

   Using READ COMMITTED isolation level causing phantom reads -> Option B

4. Quick Check:

   Phantom reads occur at READ COMMITTED [OK]

Hint: Phantom reads happen at READ COMMITTED level [OK]

Common Mistakes:

• Assuming dirty reads cause this behavior
• Confusing lost updates with read phenomena
• Thinking REPEATABLE READ allows phantom reads

5. You want to prevent both dirty reads and phantom reads in your application. Which transaction isolation level should you choose?

hard

A. SERIALIZABLE

B. READ COMMITTED

C. REPEATABLE READ

D. READ UNCOMMITTED

Solution

1. Step 1: Review isolation levels and phenomena prevented

   READ UNCOMMITTED allows dirty reads. READ COMMITTED prevents dirty reads but allows phantom reads. REPEATABLE READ prevents dirty and non-repeatable reads but may allow phantom reads.

2. Step 2: Identify level preventing both dirty and phantom reads

   SERIALIZABLE is the strictest level and prevents dirty reads, non-repeatable reads, and phantom reads by fully isolating transactions.

3. Final Answer:

   SERIALIZABLE -> Option A

4. Quick Check:

   SERIALIZABLE prevents dirty and phantom reads [OK]

Hint: Use SERIALIZABLE to prevent dirty and phantom reads [OK]

Common Mistakes:

• Choosing REPEATABLE READ thinking it blocks phantom reads
• Confusing READ COMMITTED with preventing all read anomalies
• Ignoring performance impact of SERIALIZABLE