As software becomes less monolithic and more service-oriented, log collection becomes a real problem. How can we stop hacking together brittle log parsing scripts and start building a unified logging layer? Fluentd is a data collector written in Ruby to solve this problem. Unlike other log management tools that are designed for a single backend system, Fluentd aims to connect many input sources into several output systems. This talk surveys Fluentd's architecture and shows real-world use cases of how different users extend it to do more with their logs with less code.