Consistent, Complete Distributed Stream Processing ft. Guozhang Wang


Manage episode 298080531 series 2510642
By Confluent, original creators of Apache Kafka® and Original creators of Apache Kafka®. Discovered by Player FM and our community — copyright is owned by the publisher, not Player FM, and audio is streamed directly from their servers. Hit the Subscribe button to track updates in Player FM, or paste the feed URL into other podcast apps.

Stream processing has become an important part of the big data landscape as a new programming paradigm to implement real-time data-driven applications. One of the biggest challenges for streaming systems is to provide correctness guarantees for data processing in a distributed environment. Guozhang Wang (Distributed Systems Engineer, Confluent) contributed to a leadership paper, along with other leaders in the Apache Kafka® community, on consistency and completeness in streaming processing in Apache Kafka in order to shed light on what a reimagined, modern infrastructure looks like.

In his white paper titled Consistency and Completeness: Rethinking Distributed Stream Processing in Apache Kafka, Guozhang covers the following topics in his paper:

  • Streaming correctness challenges
  • Stream processing with Kafka
  • Exactly-once in Kafka Streams

For context, accurate, real-time data stream processing is more friendly to modern organizations that are composed of vertically separated engineering teams. Unlike in the past, stream processing was considered as an auxiliary system to normal batch processing oriented systems, often bearing issues around consistency and completeness. While modern streaming engines, such as ksqlDB and Kafka Streams are designed to be authoritative, as the source of truth, and are no longer treated as an approximation, by providing strong correctness guarantees. There are two major umbrellas of the correctness of guarantees:

  1. Consistency: Ensuring unique and extant records
  2. Completeness: Ensuring the correct order of records, also referred to as exactly-once semantics.

Guozhang also answers the question of why he wrote this academic paper, as he believes in the importance of knowledge sharing across the community and bringing industry experience back to academia (the paper is also published in SIGMOD 2021, one of the most important conference proceedings in the data management research area). This will help foster the next generation of industry innovation and push one step forward in the data streaming and data management industry. In Guozhang's own words, "Academic papers provide you this proof of concept design, which gets groomed into a big system."


179 episodes