FlashDB

FlashDB is my attempt to build a relational SQL database from scratch in C++.

I’m building it to understand how databases actually work internally instead of treating a database as a black box.

The architecture is inspired by Edward Sciore’s Database Design and Implementation.

Architecture

SQL
 ↓
Lexer
 ↓
Parser
 ↓
AST
 ↓
Planner
 ↓
Executor
 ↓
Record Manager
 ↓
Buffer Manager
 ↓
File Manager
 ↓
Disk

Development

Storage

• File Manager — manages database files and fixed-size pages.
• Buffer Manager — caches pages in memory and tracks dirty and pinned pages.
• Record Manager — stores, reads, updates, and deletes records.

Metadata

• Schema — defines table columns and data types.
• Table Catalog — manages table definitions.
• Record Layout — manages field positions and record sizes.

SQL

The SQL frontend converts SQL text into tokens, parses them, and builds an AST representing each query.

Supported Operations

-- Create a table
CREATE TABLE student (
    id INT,
    name VARCHAR
);

-- Insert a row
INSERT INTO student VALUES (1, 'Alice');

-- Read rows
SELECT name
FROM student;

-- Filter rows
SELECT name
FROM student
WHERE id = 1;

-- Update rows
UPDATE student
SET name = 'Bob'
WHERE id = 1;

-- Delete rows
DELETE FROM student
WHERE id = 1;

Keyword Groups

DDL — Data Definition

CREATE
TABLE

DML — Data Manipulation

INSERT
INTO
VALUES
UPDATE
SET
DELETE

Query

SELECT
FROM
WHERE

Data Types

INT
VARCHAR

Supported Operators

=

Current Limitations

• No joins
• No ORDER BY
• No GROUP BY
• No aggregate functions
• No subqueries
• No index-based query execution
• Limited data types
• Limited expression operators

Query Engine

• Planner — converts SQL ASTs into query plans.
• Executor — executes query plans.
• Operators — provides scans, filters, and projections.

Transactions

• Transactions — groups database operations into units of work.
• Logging — records changes for durability.
• Recovery — restores database state after failures.

Indexing

• B+ Tree — planned index structure for efficient lookups.
• Index Manager — manages indexes associated with tables.
• Index Scans — enables indexed query execution.

Optimization

• Query Optimizer — chooses execution plans.
• Cost Estimation — estimates plan costs.
• Plan Selection — selects an execution strategy.

Project Structure

FlashDB/
├── src/
│   ├── file/
│   ├── log/
│   ├── buffer/
│   ├── tx/
│   ├── record/
│   ├── metadata/
│   ├── parser/
│   ├── planner/
│   ├── query/
│   └── index/
├── tests/
├── docs/
├── CMakeLists.txt
└── README.md

Build

cmake -S . -B build
cmake --build build

Test

ctest --test-dir build --output-on-failure

Project Goals

FlashDB was initially built as an educational project to understand how SQL and database systems work internally. It has since grown into a working database engine that closely follows the structure and behavior of a real database system.