CDB

CDB is a SQL database built for learning about the inner workings of databases. CDB was heavily inspired by SQLite, CockroachDB, and the CMU Database Group. CDB implements a subset of SQL described below.

Language reference

System tables

cdb_schema which holds the database schema.

SELECT

graph LR
begin(( ))
explain([EXPLAIN])
queryPlan([QUERY PLAN])
select([SELECT])
expression([expression])
as([AS identifier])
colSep[","]
from([FROM])

begin --> explain
explain --> queryPlan
queryPlan --> select
begin --> select
explain --> select
select --> expression
expression --> as
as --> colSep
expression --> colSep
colSep --> expression
expression --> from
from --> table

Loading

CREATE

Create supports the PRIMARY KEY column constraint for a single integer column.

graph LR
begin(( ))
explain([EXPLAIN])
queryPlan([QUERY PLAN])
create([CREATE])
table([TABLE])
colTypeInt([INTEGER])
colTypeText([TEXT])
lparen["("]
colSep[","]
rparen[")"]
tableIdent["Table Identifier"]
colIdent["Column Identifier"]
pkConstraint["PRIMARY KEY"]

begin --> explain
explain --> queryPlan
queryPlan --> create
begin --> create
explain --> create
create --> table
table --> tableIdent
tableIdent --> lparen
lparen --> colIdent
colIdent --> colTypeInt
colIdent --> colTypeText
colTypeInt --> pkConstraint
colTypeInt --> colSep
colTypeInt --> rparen
colTypeText --> colSep
colTypeText --> rparen
pkConstraint --> colSep
pkConstraint --> rparen
colSep --> rparen
colSep --> colIdent

Loading

INSERT

graph LR
begin(( ))
explain([EXPLAIN])
queryPlan([QUERY PLAN])
insert([INSERT])
into([INTO])
tableIdent["Table Identifier"]
lparen["("]
rparen[")"]
colSep[","]
values([VALUES])
lparen2["("]
rparen2[")"]
colSep2[","]
literal["literal"]
valSep[","]

begin --> explain
explain --> queryPlan
queryPlan --> insert
begin --> insert
explain --> insert
insert --> into
into --> tableIdent
tableIdent --> lparen
lparen --> colIdent
colIdent --> colSep
colSep --> colIdent
colSep --> rparen
rparen --> values
values --> lparen2
lparen2 --> literal
literal --> colSep2
colSep2 --> literal
literal --> rparen2
colSep2 --> rparen2
rparen2 --> valSep
valSep --> lparen2

Loading

Flags

Run cdb -h for command line flags.

Architecture

---
title: Packages
---
graph LR
    subgraph Adapters
    Driver
    REPL
    end
    Driver --> DB
    REPL --> DB
    DB --> Compiler
    subgraph Compiler
    Lexer --> Parser --> AST
    end
    AST --> Planner
    Planner --> VM
    Planner --> Catalog
    VM --> KV
    subgraph KV
    Cursor
    Catalog
    Encoder
    end
    KV --> Pager
    subgraph Pager
    Storage
    end

Loading

REPL (Read Eval Print Loop)

The REPL works with the DB (Database) layer and is responsible for two things. Passing down the SQL strings that are read by the REPL to the DB. Printing out execution results that are returned from the DB layer. The REPL can be thought of as an adapter layer.

Driver

The Driver plays the same role as the REPL in that it adapts the DB to be used in a Go program. This is done by implementing the Go standard library database/sql/driver.Driver interface.

DB (Database)

The DB (Database) layer is an interface that is called by adapters like the REPL. In theory, the DB layer could be called directly by a consumer of the package or by something like a TCP connection adapter.

Compiler

The Compiler is responsible for converting a raw SQL string to a AST (Abstract syntax tree). In doing this, the compiler performs two major steps known as lexing and parsing.

Planner

The Planner is what is known as a query planner. The planner takes the AST generated by the compiler and performs steps to generate an optimal "byte code" routine consisting of commands defined in the VM. This routine can be examined by prefixing any SQL statement with the EXPLAIN keyword.

VM (Virtual Machine)

The VM defines a set of commands that can be executed or explained. Each command performs basic calls into the KV layer that make up a query execution. This mechanism makes queries predictable and consistent.

KV (Key Value)

The KV layer implements a data structure known as a B+ tree this tree enables the database to perform fast lookups. At this layer, data is encoded into byte slices by the Encoder. The KV layer implements a cursor abstraction, which enables queries to scan and seek the B trees associated with a table or index. Additionally this layer maintains the Catalog, an in memory representation of the database schema.

Pager

The Pager sits on top of a contiguous block of bytes defined in the Storage interface. This block is typically a single file enabling the database to persist data, but it can be an in memory representation. The pager abstracts this block into pages which represent nodes in the KV layer's B tree. The pager is capable of caching the pages. The pager implements a read write mutex for concurrency control. The pager implements atomic writes to its storage through what is known as the journal file.