A high-performance centralized coordination service is a critical component for any distributed application. Apache Zookeeper is such a component and it has been in the past few years a practical solution to cluster coordination. When running ZooKeeper in production, however, there are many issues that one needs to be aware of, like proper connection management, number of direct children of a single node, herd effect, watcher implementations etc. In this presentation, I will cover some of my experience while running ZooKeeper in production.