🎧 Optimizing Apache Kafka's Internals with Its Co-Creator Jun Rao

There’s a new Streaming Audio episode - check it out!

You already know Apache Kafka® is a distributed event streaming system for setting your data in motion, but how does its internal architecture work? No one can explain Kafka’s internal architecture better than Jun Rao, one of its original creators and Co-Founder of Confluent. Jun has an in-depth understanding of Kafka that few others can claim—and he shares that with us in this episode, and in his new Kafka Internals course on Confluent Developer.

One of Jun's goals in publishing the Kafka Internals course was to cover the evolution of Kafka since its initial launch. In line with that goal, he discusses the history of Kafka development, including the original thinking behind some of its design decisions, as well as how its features have been improved to better meet its key goals of durability, scalability, and real-time data.

With respect to its initial design, Jun relates how Kafka was conceived from the ground up as a distributed system, with compute and storage always maintained as separate entities, so that they could scale independently. Additionally, he shares that Kafka was deliberately made for high throughput since many of the popular messaging systems at the time of its invention were single node, but his team needed to process large volumes of non-transactional data, such as application metrics, various logs, click streams, and IoT information.

As regards the evolution of its features, in addition to others, Jun explains these two topics at great length:

  • Consumer rebalancing protocol: The original "stop the world" approach to Kafka's consumer rebalancing—although revolutionary at the time of its launch, was eventually improved upon to take a more incremental approach.
  • Cluster metadata: Moving from the external ZooKeeper to the built-in KRaft protocol allows for better scaling by a factor of ten. according to Jun, and it also means you only need to worry about running a single binary.

The Kafka Internals course consists of eleven concise modules, each dense with detail—covering Kafka fundamentals in technical depth. The course also pairs with four hands-on exercise modules led by Senior Developer Advocate Danica Fine.


:headphones: Listen to the episode

1 Like