Kafka: The Definitive Guide

Chapter 11. Stream Processing

Kafka was traditionally seen as a powerful message bus, capable of delivering streams of events but without processing or transformation capabilities. Kafka’s reliable stream delivery capabilities make it a perfect source of data for stream-processing systems. Apache Storm, Apache Spark Streaming, Apache Flink, Apache Samza, and many more stream-processing systems were built with Kafka often being their only reliable data source.

Industry analysts sometimes claim that all those stream-processing systems are just like the complex event processing (CEP) systems that have been around for 20 years. We think stream processing became more popular because it was created after Kafka and therefore could use Kafka as a reliable source of event streams to process. With the increased popularity of Apache Kafka, first as a simple message bus and later as a data integration system, many companies had a system containing many streams of interesting data, stored for long amounts of time and perfectly ordered, just waiting for some stream-processing framework to show up and process them. In other words, in the same way that data processing was significantly more difficult before databases were invented, stream processing was held back by lack of a stream-processing platform.

Starting from version 0.10.0, Kafka does more than provide a reliable source of data streams to every popular stream-processing framework. Now Kafka includes a powerful stream-processing library ...