Apache: Big Data North America 2017

Twitter collects more than Trillion messages per day. These messages are grouped into hundreds of categories which have different properties. Messages are routed based on category to various nodes in a cluster until they reach storage systems serving Analytics and Streaming workloads. The scale of messages with different delivery guarantees pose unique challenges at Twitter.

Twitter’s Log Collection framework has been built using Scribe over the years. Message delivery guarantee, priority and multiplexing add complexities to routing. Additionally Twitter scale introduces unique challenges for management of the logging framework. In this talk we discuss about challenges we face and effort to improving our logging framework using Apache Flume. Apache Flume with its pluggable architecture provides many building blocks to implement various features for our collection framework.