MyDuck Server

MyDuck Server unlocks serious power for your MySQL & Postgres analytics. Imagine the simplicity of (MySQL|Postgres)’s familiar interface fused with the raw analytical speed of DuckDB. Now you can supercharge your analytical queries with DuckDB’s lightning-fast OLAP engine, all while using the tools and dialect you know.

📑 Table of Contents

• Why MyDuck
• Key Features
• Performance
• Getting Started
• Prerequisites
• Installation
• Usage
• Replicating Data
• Connecting to Cloud MySQL & Postgres
• HTAP Setup
• Customizing the Docker Container
• Query Parquet Files
• Already Using DuckDB?
• Backup and Restore with Object Storage
• LLM Integration
• Access from Python
• Roadmap
• Contributing
• Acknowledgements
• License

❓ Why MyDuck ❓

While MySQL and Postgres are the most popular open-source databases for OLTP, their performance in analytics often falls short. DuckDB, on the other hand, is built for fast, embedded analytical processing. MyDuck Server lets you enjoy DuckDB's high-speed analytics without leaving the (MySQL|Postgres) ecosystem.

With MyDuck Server, you can:

• Set up an isolated, fast, and real-time replica dedicated to ad-hoc analytics, batch jobs, and LLM-generated queries, without exhausting or corrupting your primary database 🔥
• Accelerate existing MySQL & Postgres analytics to new heights through DuckDB's high-speed engine with minimal changes 🚀
• Enable richer & faster connectivity between modern data manipulation & analysis tools and your MySQL & Postgres data 🛠️
• Go beyond MySQL & Postgres syntax with DuckDB's advanced SQL features to expand your analytics potential 🦆
• Run DuckDB in server mode to share a DuckDB instance with your team or among your applications 🌩️
• Build HTAP systems by combining (MySQL|Postgres) for transactions with MyDuck for analytics 🔄
• and much more! See below for a full list of feature highlights.

MyDuck Server isn't here to replace MySQL & Postgres — it's here to help MySQL & Postgres users do more with their data. This open-source project provides a convenient way to integrate high-speed analytics into your workflow while embracing the flexibility and efficiency of DuckDB.

✨ Key Features

• Blazing Fast OLAP with DuckDB: MyDuck stores data in DuckDB, an OLAP-optimized database known for lightning-fast analytical queries. DuckDB enables MyDuck to execute queries up to 1000x faster than traditional MySQL & Postgres setups, making complex analytics practical that were previously unfeasible.

• MySQL-Compatible Interface: MyDuck implements the MySQL wire protocol and understands MySQL syntax, allowing you to connect with any MySQL client and run MySQL-style SQL. MyDuck automatically translates your queries and executes them in DuckDB.

• Postgres-Compatible Interface: MyDuck implements the Postgres wire protocol, enabling you to send DuckDB SQL directly using any Postgres client. Since DuckDB's SQL dialect closely resembles PostgreSQL, you can speed up existing Postgres queries with minimal changes.

• Raw DuckDB Power: MyDuck provides full access to DuckDB's analytical capabilities through raw DuckDB SQL, including friendly SQL syntax, advanced aggregates, remote data source access, nested data types, and more.

• Zero-ETL: Simply start replication and begin querying! MyDuck can function as a MySQL replica or Postgres standby, replicating data from your primary server in real-time. It works like standard MySQL & Postgres replication - using MySQL's START REPLICA or Postgres' CREATE SUBSCRIPTION commands, eliminating the need for complex ETL pipelines.

• Consistent and Efficient Replication: Thanks to DuckDB's solid ACID support, we've carefully managed transaction boundaries in the replication stream to ensure a consistent data view — you'll never see dirty data mid-transaction. Plus, MyDuck's transaction batching collects updates from multiple transactions and applies them to DuckDB in batches, significantly reducing write overhead (since DuckDB isn’t designed for high-frequency OLTP writes).

• HTAP Architecture Support: MyDuck works well with database proxy tools to enable hybrid transactional/analytical processing setups. You can route DML operations to (MySQL|Postgres) and analytical queries to MyDuck, creating a powerful HTAP architecture that combines the best of both worlds.

• Bulk Upload & Download: MyDuck supports fast bulk data loading from the client side with the standard MySQL LOAD DATA LOCAL INFILE command or the PostgreSQL COPY FROM STDIN command. You can also extract data from MyDuck using the PostgreSQL COPY TO STDOUT command.

• End-to-End Columnar IO: In addition to the traditional row-oriented data transfer in MySQL & Postgres protocol, MyDuck can also send query results and receive data uploads in columnar format, which can be significantly faster for high-volume data. This is implemented on top of the standard Postgres COPY protocol with extended columnar format support, e.g., COPY ... TO STDOUT (FORMAT parquet | arrow), allowing you to use the standard Postgres client library to interact with MyDuck in an optimized way.

• Standalone Mode: MyDuck can run in standalone mode without replication. In this mode, it is a drop-in replacement for (MySQL|Postgres), but with a DuckDB heart. You can CREATE TABLE, transactionally INSERT, UPDATE, and DELETE data, and run blazingly fast SELECT queries.

• DuckDB in Server Mode: If you aren't interested in MySQL & Postgres but just want to share a DuckDB instance with your team or among your applications, MyDuck is also a great solution. You can deploy MyDuck to a server, connect to it with the Postgres client library in your favorite programming language, and start running DuckDB SQL queries directly.

• Seamless Integration with Dump & Copy Utilities: MyDuck plays well with modern MySQL & Postgres data migration tools, especially the MySQL Shell and pg_dump. For MySQL, you can load data into MyDuck in parallel from a MySQL Shell dump, or leverage the Shell’s copy-instance utility to copy a consistent snapshot of your running MySQL server to MyDuck. For Postgres, MyDuck can load data from a pg_dump archive.

📊 Performance

Typical OLAP queries can run up to 1000x faster with MyDuck Server compared to MySQL & Postgres alone, especially on large datasets. Under the hood, it's just DuckDB doing what it does best: processing analytical queries at lightning speed. You are welcome to run your own benchmarks and prepare to be amazed! Alternatively, you can refer to well-known benchmarks like the ClickBench and H2O.ai db-benchmark to see how DuckDB performs against other databases and data science tools. Also remember that DuckDB has robust support for transactions, JOINs, and larger-than-memory query processing, which are unavailable in many competing systems and tools.

🏃‍♂️ Getting Started

Prerequisites

• Docker (recommended) for setting up MyDuck Server quickly.
• MySQL or PostgreSQL CLI clients for connecting and testing your setup.

Installation

Get a standalone MyDuck Server up and running in minutes using Docker:

docker run -p 13306:3306 -p 15432:5432 apecloud/myduckserver:latest

This setup exposes:

• Port 13306 for MySQL wire protocol connections.
• Port 15432 for PostgreSQL wire protocol connections, allowing direct DuckDB SQL.

Usage

Connecting via MySQL client

Connect using any MySQL client to run MySQL-style SQL queries:

mysql -h127.0.0.1 -P13306 -uroot

[!NOTE] MySQL CLI clients version 9.0 and above are not yet supported on macOS. Consider brew install mysql-client@8.4.

Connecting via PostgreSQL client

For full analytical power, connect to the Postgres port and run DuckDB SQL queries directly:

psql -h 127.0.0.1 -p 15432 -U postgres

Replicating Data

We have integrated a setup tool in the Docker image that helps replicate data from your primary (MySQL|Postgres) server to MyDuck Server. The tool is available via the SETUP_MODE environment variable. In REPLICA mode, the container will start MyDuck Server, dump a snapshot of your primary (MySQL|Postgres) server, and start replicating data in real-time.

[!NOTE] Supported primary database versions: MySQL>=8.0 and PostgreSQL>=13. In addition to the default settings, logical replication must be enabled for PostgreSQL by setting wal_level=logical. For MySQL, GTID-based replication (gtid_mode=ON and enforce_gtid_consistency=ON) is recommended but not required.

docker run -d --name myduck \
  -p 13306:3306 \ 
  -p 15432:5432 \
  --env=SETUP_MODE=REPLICA \
  --env=SOURCE_DSN="<postgres|mysql>://<user>:<password>@<host>:<port>/<dbname>"
  apecloud/myduckserver:latest

SOURCE_DSN specifies the connection string to the primary database server, which can be either MySQL or PostgreSQL.

• MySQL Primary: Use the mysql URI scheme, e.g.,
  --env=SOURCE_DSN=mysql://root:password@example.com:3306

• PostgreSQL Primary: Use the postgres URI scheme, e.g.,
  --env=SOURCE_DSN=postgres://postgres:password@example.com:5432/db01

[!NOTE] To replicate from a server running on the host machine, use host.docker.internal as the hostname instead of localhost or 127.0.0.1. On Linux, you must also add --add-host=host.docker.internal:host-gateway to the docker run command.

Connecting to Cloud MySQL & Postgres

MyDuck Server supports setting up replicas from common cloud-based MySQL & Postgres offerings. For more information, please refer to the replica setup guide.

HTAP Setup

With MyDuck's powerful analytics capabilities, you can create an hybrid transactional/analytical processing system where high-frequency data writes are directed to a standard MySQL or Postgres instance, while analytical queries are handled by a MyDuck Server instance. Follow our HTAP setup instructions to easily set up an HTAP demonstration: * Provisioning a MySQL HTAP cluster based on ProxySQL or MariaDB MaxScale. * Provisioning a PostgreSQL HTAP cluster based on PGPool-II

Customizing the Docker Container

To rename the default database, pass the DEFAULT_DB environment variable to the Docker container:

docker run -d -p 13306:3306 -p 15432:5432 \
    --env=DEFAULT_DB=mydbname \
    apecloud/myduckserver:latest

To set the superuser password, pass the SUPERUSER_PASSWORD environment variable to the Docker container:

docker run -d -p 13306:3306 -p 15432:5432 \
    --env=SUPERUSER_PASSWORD=mysecretpassword \
    apecloud/myduckserver:latest

To initialize MyDuck Server with custom SQL statements, mount your .sql file to either /docker-entrypoint-initdb.d/mysql/ or /docker-entrypoint-initdb.d/postgres/ inside the Docker container, depending on the SQL dialect you're using.

For example:

# Execute `init.sql` via MySQL protocol
docker run -d -p 13306:3306 --name=myduck \
    -v ./init.sql:/docker-entrypoint-initdb.d/mysql/init.sql \
    apecloud/myduckserver:latest

# Execute `init.sql` via PostgreSQL protocol
docker run -d -p 15432:5432 --name=myduck \
    -v ./init.sql:/docker-entrypoint-initdb.d/postgres/init.sql \
    apecloud/myduckserver:latest

Query Parquet Files

Looking to load Parquet files into MyDuck Server and start querying? Follow our Parquet file loading guide for easy setup.

Already Using DuckDB?

Already have a DuckDB file? You can seamlessly bootstrap MyDuck Server with it. See our DuckDB file bootstrapping guide for more details.

Managing Multiple Databases

Easily manage multiple databases in MyDuck Server, same as Postgres. For step-by-step instructions and detailed guidance, check out our Database Management Guide.

Backup and Restore with Object Storage

To back up and restore your databases inside MyDuck Server using object storage, refer to our backup and restore guide for detailed instructions.

LLM Integration

MyDuck Server can be integrated with LLM applications via the Model Context Protocol (MCP). Follow the MCP integration guide to set up MyD