Exploring Top MySQL Middleware: Atlas, MySQL Router, Mycat, Cobar, and More

1. Introduction

MySQL is a ubiquitous database in internet companies; when performance issues arise, horizontal scaling via master‑slave replication improves read throughput, while sharding addresses write bottlenecks. This article focuses on MySQL data access middleware (DAL) implementations rather than high‑availability architectures.

2. Atlas

Atlas is a MySQL‑protocol data middleware developed by Qihoo 360. Built on MySQL‑Proxy 0.8.2, it fixes many bugs and adds features. It is widely used internally at 360 and by over 50 companies, handling tens of billions of read/write requests daily.

2.1. Atlas Overview

Atlas sits between applications and MySQL, acting as a client to the backend DB and as a server to the frontend application. It implements both MySQL client and server protocols and maintains a connection pool to reduce MySQL load.

Key functions include:

Read‑write separation

Slave load balancing

IP filtering

Automatic sharding

Graceful DBA‑controlled DB online/offline

Automatic removal of failed DB nodes

Advantages over official MySQL‑Proxy:

Core logic rewritten in C (Lua only for management)

Improved network and thread models

Real connection pool implementation

Optimized lock mechanism, delivering dozens‑fold performance gains

3. MySQL Router

Official MySQL Router is a lightweight proxy placed between client applications and DB servers. It detects, analyzes, and forwards queries, supporting read‑write separation, automatic failover, load balancing, and connection pooling.

3.1. Router Architecture

Router distinguishes read and write traffic by port: write queries go to the primary, while read‑only queries are round‑robin distributed among replicas.

3.2. Main Functions

Read‑write separation

Automatic primary‑replica failover

Load balancing

Connection pooling

4. Mycat

Mycat evolved from the open‑source Cobar project, rewriting the network module with NIO and enhancing buffer handling, aggregation, and join capabilities. Since version 1.4 it no longer depends on the original Cobar core.

4.1. Mycat Overview

Mycat is a high‑performance database middleware that supports SQL92, adheres to the native MySQL protocol, and works across languages, platforms, and databases.

4.2. Main Features

SQL92 compliance

Transparent MySQL protocol, cross‑language support

Heartbeat‑based automatic failover, read‑write separation, support for MySQL, Galera, Percona, MariaDB clusters

NIO‑based high‑concurrency handling

Automatic sharding, aggregation (SUM, COUNT, MAX, etc.)

Cross‑database joins, including multi‑table joins via CALTLET

Global tables and ER‑based sharding for efficient multi‑table queries

Multi‑tenant support

Weak XA distributed transactions

Global sequence numbers for primary‑key generation

Rich sharding rules and plugin architecture

Web and CLI monitoring

Support for Oracle, DB2, SQL Server, MongoDB, and other back‑ends via JDBC

Password encryption, service degradation, IP whitelist, SQL blacklist, injection protection

Table partitioning (since 1.6) and ZooKeeper‑managed clustering with online upgrade and scaling

5. Cobar

Cobar provides distributed MySQL services, acting as a transparent proxy that enables linear scaling of traditional databases.

5.1. Architecture

5.2. Status

Developed by Alibaba, Cobar has been stable for over three years, handling more than 3,000 MySQL schemas, processing over 50 billion SQL requests and terabytes of data daily. Development halted after 2013, and Mycat emerged as its successor.

6. Amoeba

Amoeba (released in 2008) is an open‑source MySQL front‑end proxy that provides load balancing, high availability, SQL filtering, read‑write separation, routing, and result merging across multiple databases.

6.1. Architecture

6.2. Status

The project is no longer maintained.

7. MySQL Proxy

MySQL Proxy sits between client and server, allowing monitoring, analysis, and modification of traffic via Lua scripts. It can implement load balancing, query filtering, and read‑write separation, though it is now deprecated in favor of MySQL Router.

8. Client‑Side Sharding

Client‑side sharding splits databases and tables within the application code, e.g., routing user IDs to different databases/tables based on the last digits of the UID. Advantages include no middleware overhead and fine‑grained control via code.

9. Conclusion

For production environments, the author recommends using Atlas or Mycat; client‑side sharding is suitable when strong development capability or DBA support exists. MySQL Router is still maturing, and other middleware that is no longer updated should be avoided unless legacy constraints require them.