After reading Introducing Kafka Streams: Stream Processing Made Simple, I thought it is a pretty nice progress. Kafka Stream makes Kafka not only a simple high-throughput, low-latency, and fault tolerance message queue but also a streaming processing engine. It means it could replace some of the jobs of Spark or Flink.
Normally, companies would use Spark or Flink to do ETL or more sophisticated and long-running analysis. But Kafka Stream seems could be used as part of real-time ETL and light-weight analysis. This point is I think a lot of company only need this capability - real time and relative light-weight processing with fault tolerance. That’s Kafka Stream could step into.
As Jay Kreps mentioned,
It clearly shows you the comparison of Kafka Stream with other related tech tools.
By the way, I think Kafka’s design documents is really really nice, it’s a pleasure to read, after I glimpsed one of KIPs(Kafka Improvement Proposals), Exactly Once Delivery and Transactional Messaging.