Percona Live 2017 Open Source Database Conference

April 24 - 27, 2017

Santa Clara, California

Bootstrapping a New Datacenter Made Easy

Bootstrapping a New Datacenter Made Easy

 25 April - 1:20 PM - 2:10 PM @ Room 210
Experience level: 
50 minutes conference
Tools and Techniques


Yelp has grown rapidly over the last several years, and we were forced to quickly scale our infrastructure to keep up. While every company experiences growth in a different way, there are some realities that all growing and successful organizations will face: geographical expansion will become a requirement, launching new database clusters will become vital, and all hosts will become replaceable. You must have automation and a powerful toolset, or be forced to put in a lot of hours and manual labor to accomplish these goals. Yelp has endeavored to run efficiently and cost effectively, and we have put the time in to automate away the time consuming and manual portions of this work. This talk will be about the various tools we've built, adopted, and perfected to handle the seamless launching of a new datacenter. We will describe how we bootstrap a new host, how we've automated the creation of new schemas, and our tooling for adding and removing hosts from the pool via smartstack. Attendees should have a general understanding of configuration management, and load balancing techniques.


Matt Ullmer's picture

Matt Ullmer

MySQL Database Administrator, Yelp


Worked with databases since 2005, first as a developer, and later as an Administrator. I've specialized in MySQL since 2011, and joined Yelp in 2013 as the second DBA. Since then I've specialized in performance monitoring, benchmarking, and configuration management both with physical hardware (FusionIO) and in the cloud. My work has extended to implementing many opensource projects internally to advance our ability to scale and manage an ever growing infrastructure footprint. Recent projects include quantifying impacts of traffic patterns on database performance on various EC2 instance types, and identifying failure points with the use of a multidimensional metrics system based around cluster membership and database server roles.

Share this talk