I. Introduction

Database performance optimization often hinges on effective indexing strategies, but even indexed LIKE queries can bottleneck under high data volumes. This case study demonstrates how SQLFlash transforms inefficient LIKE prefix queries into high-performance range scans, achieving 100x speed improvements through index-aware SQL rewriting. We dive into B+ tree mechanics, execution plan analysis, and practical optimization patterns for AI-driven database scaling.

II. Preparation

-- Create test table with 5M records  
CREATE TABLE users (  
    id INT PRIMARY KEY AUTO_INCREMENT,  
    username VARCHAR(255) NOT NULL COMMENT 'Format: user_8digits',  
    email VARCHAR(255) NOT NULL  
) ENGINE=InnoDB;

-- Bulk data generation procedure
DELIMITER $$
CREATE PROCEDURE GenerateTestData()  
BEGIN
    DECLARE i INT DEFAULT 0;  
    WHILE i < 5000000 DO  
        INSERT INTO users (username, email)  
        SELECT CONCAT('user_', LPAD(io + ii, 8, '0')),  
               CONCAT(LPAD(io + ii, 8, '0'), '@test.com')  
        FROM (  
            SELECT (i + thousands.id * 1000) AS io, units.id AS ii  
            FROM (SELECT 0 AS id UNION SELECT 1 UNION SELECT 2 UNION SELECT 3 UNION SELECT 4 UNION SELECT 5 UNION SELECT 6 UNION SELECT 7 UNION SELECT 8 UNION SELECT 9) AS units,  
                 (SELECT 0 AS id UNION SELECT 1 UNION SELECT 2 UNION SELECT 3 UNION SELECT 4 UNION SELECT 5 UNION SELECT 6 UNION SELECT 7 UNION SELECT 8 UNION SELECT 9) AS hundreds,  
                 (SELECT 0 AS id UNION SELECT 1 UNION SELECT 2 UNION SELECT 3 UNION SELECT 4 UNION SELECT 5 UNION SELECT 6 UNION SELECT 7 UNION SELECT 8 UNION SELECT 9) AS thousands  
        ) AS seq  
        LIMIT 1000;  
        SET i = i + 1000;  
    END WHILE;  
END$$  
DELIMITER;

CALL GenerateTestData();  
ALTER TABLE users ADD INDEX idx_username (username);

Original Query

The baseline LIKE query hits indexes but suffers from row-by-row validation:

SELECT * FROM users
WHERE username LIKE 'user_0123%'; - Full table scan despite index

SQLFlash Optimization

We tried to rewrite the SQL by SQLFlash.

Rewrite using range conditions to leverage index ordering:

SELECT * FROM users
WHERE username >= 'user_0123'
AND username < 'user_0124'; - Precision index range scan

Performance Metrics

Test Environment: MySQL 8.0.18, 8GB InnoDB Buffer Pool

View Detailed Report

Based on the explanation provided by SQLFlash, in this case, the LIKE prefix query is optimized by rewriting username LIKE 'user_0123%' as username >= 'user_0123' AND username < 'user_0124'. This optimization can effectively leverage the ordered characteristics of the character index and reduce full table scans.

III. Technical Deep Dive

A. Index Utilization Modes

• LIKE: type: range + Row-by-row verification of LIKE conditions

• ​​Range: Directly targets the precise interval with >=x AND <y

B. Execution Plan Comparison

-- Original Query Plan  
| id | type | key          | rows   | Extra       |  
|----|------|--------------|--------|-------------|  
| 1  | ALL  | idx_username | 4.98M  | Using where |  

-- Optimized Query Plan  
| id | type  | key          | rows  | Extra               |  
|----|-------|--------------|-------|---------------------|  
| 1  | RANGE | idx_username | 18.8K | Using index condition |

C. Performance Drivers

1. Reduced I/O: 16 data pages → 1 page read

2. CPU Efficiency: Eliminates 200M string validations

3. Index Selectivity: 200x precision gain

IV. Equivalence Proof

The equivalence between the two query patterns is established on the dual guarantees of index orderliness and string dictionary order rules:

A. Index Sorting Determinacy

The B+ tree index of the username field strictly adheres to lexicographical order. For fixed-format data with the user_ prefix followed by 8 digits, its index sorting is mathematically equivalent to numerical sorting of the numeric portion. For example:

user_00000001 < user_00000002 < ... < user_01239999 < user_01240000

B. Range Boundary Closure

When using LIKE 'user_0123%', it is effectively equivalent to:

username >= 'user_0123' AND username < 'user_0124'

This is because the maximum possible value for 'user_0123%' is 'user_01239999...' (assuming fixed-length padding), while the next lexicographical starting point is 'user_0124'. This rewrite achieves precise interval closure through string comparison operators, maintaining mathematical equivalence to the LIKE pattern.

C. Character Set Collation Guarantee

Under common collations like utf8_general_ci, the lexicographical order of digits 0-9 is identical to their numerical order. This ensures that string comparisons on the username field are isomorphic to numerical comparisons of its numeric segments. As a result, range queries avoid any positional misalignment or missing records.

V. The Essence of the Difference in Execution Plan Deviation

This difference ultimately leads to order — of — magnitude differences in the I/O overhead, CPU verification cost, and execution path length between the two. The characteristic that the result sets are completely identical makes the optimization reliable.

VI. Conclusion

This case highlights how SQL semantics-aware rewriting unlocks hidden performance potential:
• Same index → 100x faster results
• Eliminates full scans through range optimization
• Maintains result accuracy through mathematical equivalence

Adopt this pattern to transform LIKE queries into high-performance operations.

Original quote:https://sqlflash.ai/blog/mysql-like-optimization/