duckdb（duckdb 数据库）

DuckDB

DuckDB is an in-process analytical database system designed for fast analytical queries. It supports SQL, embedded operation, and seamless integration with data science tools.

Quick Start

# Install DuckDB CLI
curl https://install.duckdb.org | sh

# Start DuckDB shell
duckdb

# Run SQL query directly
duckdb -c "SELECT 42"

# Query a CSV file
duckdb -c "SELECT * FROM 'data.csv' LIMIT 10"

Installation

macOS/Linux

# Via install script
curl https://install.duckdb.org | sh

# Via Homebrew (macOS)
brew install duckdb

# Via conda
conda install -c conda-forge duckdb

Python

pip install duckdb

Other Platforms

• Windows: Download from https://duckdb.org/install/
• R: install.packages("duckdb")
• Node.js: npm install duckdb
• Java: Maven dependency org.duckdb:duckdb_jdbc

SQL Statements

SELECT

-- Basic query
SELECT * FROM users WHERE age > 25;

-- Aggregate
SELECT city, COUNT(*) FROM users GROUP BY city;

-- Join
SELECT a.*, b.name FROM orders a JOIN users b ON a.user_id = b.id;

-- FROM-first syntax (DuckDB extension)
FROM users SELECT * WHERE age > 25;

CREATE TABLE

-- Create table with schema
CREATE TABLE users (
    id INTEGER PRIMARY KEY,
    name VARCHAR,
    age INTEGER,
    created_at TIMESTAMP
);

-- Create table from query
CREATE TABLE users AS SELECT * FROM 'users.csv';

-- Create table from CSV (shortcut)
CREATE TABLE users AS FROM 'users.csv';

INSERT

-- Insert values
INSERT INTO users (id, name, age) VALUES (1, 'Alice', 30);

-- Insert from SELECT
INSERT INTO users SELECT * FROM new_users;

-- Insert from CSV
INSERT INTO users SELECT * FROM read_csv('new_users.csv');

COPY (Import/Export)

-- Import CSV to table
COPY users FROM 'users.csv';

-- Import with options
COPY users FROM 'users.csv' (DELIMITER '|', HEADER true);

-- Import Parquet
COPY users FROM 'users.parquet' (FORMAT parquet);

-- Import JSON
COPY users FROM 'users.json' (FORMAT json, AUTO_DETECT true);

-- Export to CSV
COPY users TO 'users.csv' (FORMAT csv, HEADER);

-- Export query result
COPY (SELECT * FROM users WHERE age > 25) TO 'adults.parquet' (FORMAT parquet, COMPRESSION zstd);

-- Copy entire database
COPY FROM DATABASE db1 TO db2;

Data Types

General-Purpose Types

Nested Types

-- ARRAY (fixed-length)
SELECT ARRAY[1, 2, 3];
CREATE TABLE t (arr INTEGER[3]);

-- LIST (variable-length)
SELECT [1, 2, 3];
CREATE TABLE t (lst INTEGER[]);

-- STRUCT
SELECT {'x': 1, 'y': 2};
CREATE TABLE t (s STRUCT(x INTEGER, y INTEGER));

-- MAP
SELECT MAP([1, 2], ['a', 'b']);
CREATE TABLE t (m MAP(INTEGER, VARCHAR));

-- UNION
CREATE TABLE t (u UNION(int_type INTEGER, str_type VARCHAR));

CSV Operations

Read CSV

-- Auto-detect options
SELECT * FROM 'data.csv';

-- With explicit options
SELECT * FROM read_csv('data.csv',
    delim = ',',
    header = true,
    columns = {
        'id': 'INTEGER',
        'name': 'VARCHAR'
    }
);

-- From stdin
cat data.csv | duckdb -c "SELECT * FROM read_csv('/dev/stdin')"

CSV Options

Parquet Operations

-- Read Parquet
SELECT * FROM 'data.parquet';

-- With options
SELECT * FROM read_parquet('data.parquet', hive_partitioning = true);

-- Write Parquet
COPY users TO 'users.parquet' (FORMAT parquet);

-- With compression
COPY users TO 'users.parquet' (FORMAT parquet, COMPRESSION zstd);

Python API

import duckdb

# Connect to database (in-memory)
con = duckdb.connect()

# Connect to file
con = duckdb.connect('mydb.db')

# Execute SQL
con.execute("CREATE TABLE users (id INTEGER, name VARCHAR)")
con.execute("INSERT INTO users VALUES (1, 'Alice')")

# Query and fetch
result = con.execute("SELECT * FROM users").fetchall()
print(result)  # [(1, 'Alice')]

# Fetch as DataFrame
df = con.execute("SELECT * FROM users").fetchdf()

# Query CSV directly
df = con.execute("SELECT * FROM 'data.csv'").fetchdf()

# Register DataFrame as table
con.register('my_df', df)
con.execute("SELECT * FROM my_df WHERE age > 25")

# Close connection
con.close()

CLI Usage

# Start interactive shell
duckdb

# Run SQL file
duckdb < script.sql

# Execute command
duckdb -c "SELECT 42"

# Open database file
duckdb mydb.db

# Import CSV and query
duckdb -c "SELECT * FROM read_csv_auto('data.csv') LIMIT 10"

# Output formats
.duckdb -c "SELECT * FROM users" -csv    # CSV output
.duckdb -c "SELECT * FROM users" -json   # JSON output

Common Workflows

Import CSV to Table

-- Method 1: Direct CREATE TABLE AS
CREATE TABLE users AS SELECT * FROM 'users.csv';

-- Method 2: Pre-create table, then COPY
CREATE TABLE users (id INTEGER, name VARCHAR, age INTEGER);
COPY users FROM 'users.csv' (HEADER);

-- Method 3: With explicit column mapping
COPY users FROM 'users.csv' (
    HEADER,
    COLUMNS = {'id': 'INTEGER', 'name': 'VARCHAR', 'age': 'INTEGER'}
);

Export Query Results

-- To CSV
COPY (SELECT * FROM users WHERE active = true) TO 'active_users.csv' (HEADER);

-- To Parquet
COPY (SELECT * FROM orders) TO 'orders.parquet' (FORMAT parquet);

-- To JSON
COPY (SELECT * FROM events) TO 'events.json' (FORMAT json);

Working with Multiple Files

-- Query multiple CSV files
SELECT * FROM read_csv_auto('data/*.csv');

-- With filename column
SELECT filename, * FROM read_csv_auto('data/*.csv');

-- Hive partitioning
SELECT * FROM read_parquet('data/*/*/*.parquet', hive_partitioning = true);

Extensions

-- Install and load extensions
INSTALL httpfs;
LOAD httpfs;

-- Popular extensions
INSTALL json;      LOAD json;      -- JSON support
INSTALL parquet;   LOAD parquet;   -- Parquet support
INSTALL httpfs;    LOAD httpfs;    -- HTTP/S3 support
INSTALL iceberg;   LOAD iceberg;   -- Apache Iceberg
INSTALL delta;     LOAD delta;     -- Delta Lake
INSTALL spatial;   LOAD spatial;   -- Geospatial data

Tips

• Auto-detection: DuckDB's CSV sniffer automatically detects delimiters, headers, and types. Use AUTO_DETECT = true.
• FROM-first syntax: DuckDB allows FROM table SELECT * instead of SELECT * FROM table.
• String literals: Single quotes for strings 'text', double quotes for identifiers "column".
• In-process: DuckDB runs embedded in your application - no server to manage.
• Zero-copy: Query Parquet and CSV files without loading them fully into memory.
• Parallel CSV: DuckDB automatically parallelizes CSV reading when possible.

Resources

• Documentation: https://duckdb.org/docs/
• Live Demo: https://shell.duckdb.org
• GitHub: https://github.com/duckdb/duckdb