There has been much renewed interest in using log-centric architectures to scale distributed systems that provide efficient durability and high availability. In this approach, a collection of distributed servers can operate on a replicated log that record state changes in sequential ordering. The log itself can then be treated as the “source-of-truth”: when some of the servers fail and come back, their states can be deterministically reconstructed by replaying this log upon recovery.
Over the past years of developing and operating Kafka, we have envisioned and exercised the idea of extending its commit-log structured architecture into a replicated logging system in order to serve as the underlying data flow backbone for a wide scope of applications, such as data integration, commit log replication, and stream processing, etc. In this year’s Very Large Data Bases conference I will talk about our experience in building such a replicated logging system using Kafka and will present several of its use cases.
If you happen to be attending the VLDB conference and you’re interested in learning more about how to build a replicated log using Kafka, how to deploy it as your commit log replication layer underlying your distributed stores, etc., I invite you to attend my session or find me at the conference.
Building a Replicated Logging System with Apache Kafka
Guozhang Wang, Confluent
10:30am – 12:00pm, Thursday, September 3, 2015
41st International Conference on Very Large Data Bases
Hilton Waikoloa Hotel | Kohala Coast, Hawai’i | August 31 – September 4, 2015
You may also be interested in these blog posts by Jay Kreps (Kafka co-creator):