Watermarks, Tables, Event Time, and the Dataflow Model | Confluent

This post critiques the Google Dataflow model's use of watermarks and triggers for handling out-of-order data and proposes Kafka Streams' table-centric approach as a more general and less complex alternative. It introduces the concept of tables as streams of updates, applicable to both mutable data (like customer accounts) and windowed computations. The post details how Kafka Streams uses commit interval and record cache size as operational tuning parameters to manage output volume and update lag, contrasting this with the complexity of Dataflow's triggers. It also touches upon interactive queries for accessing the latest results on demand.