Vitess

Vitess is a set of servers and tools meant to facilitate scaling of MySQL databases for the web. It's been developed since 2011, and is currently used as a fundamental component of YouTube's MySQL infrastructure, serving thousands of QPS (per server). If you want to find out whether Vitess is a good fit for your project, please read our helicopter overview.

There are a couple of videos from sougou that you can watch: a short intro prepared for Google I/O 2014 and a more detailed presentation from @Scale '14.

Overview

Vitess consists of a number servers, command line utilities, and a consistent metadata store. Taken together, they allow you to serve more database traffic, and add features like sharding, which normally you would have to implement in your application.

vttablet is a server that sits in front of a MySQL database, making it more robust and available in the face of high traffic. Among other things, it adds a connection pool, has a row based cache, and it rewrites SQL queries to be safer and nicer to the underlying database.

vtgate is a very light proxy that routes database traffic from your app to the right vttablet, basing on the sharding scheme, latency required, and health of the vttablets. This allows the client to be very simple, as all it needs to be concerned about is finding the closest vtgate.

The topology is a metadata store that contains information about running servers, the sharding scheme, and replication graph. It is backed by a consistent data store, like Apache ZooKeeper. The topology backends are plugin based, allowing you to write your own if ZooKeeper doesn't fit your needs. You can explore the topology through vtctld, a webserver (not shown in the diagram).

vtctl is a command line utility that allows a human or a script to easily interact with the system.

All components communicate using a lightweight RPC system based on BSON. The RPC system is plugin based, so you can easily write your own backend (at Google we use a Protocol Buffers based protocol). We provide a client implementation for three languages: Python, Go, and Java. Writing a client for your language should not be difficult, as it's a matter of implementing only a few API calls (please send us a pull request if you do!).

Trying it out

Vitess is not entirely ready for unsupervised use yet. Some functionality is still under development, APIs may change, and parts of the code are undocumented. However, if you feel adventurous, you're more than welcome to try it. We know that there are some rough edges, so please don't hesitate to reach out to us through our mailing list if you run into any issues. Warnings aside, please take a look at our Getting Started guide.

Documentation

Intro

• Helicopter overview: high level overview of Vitess that should tell you whether Vitess is for you.
• Frequently Asked Questions.
• Vision: principles guiding the design of Vitess.

Using Vitess

• Getting Started: how to set your environment to work with Vitess.
• Tools: all Vitess tools and servers.
• vttablet/vtocc: information about the most important Vitess server.
• Reparenting: performing master failover.
• Resharding: adding more shards to your cluster.
• Preparing for production (wip).
• Schema management: managing your database schema using Vitess.

Reference

• General Concepts
• Zookeeper data
• Serving graph
• Replication Graph

License

Unless otherwise noted, the vitess source files are distributed under the BSD-style license found in the LICENSE file.