The Dataflow Model Revisited

This article argues that the original Dataflow Model correctly anticipated the need to process incomplete, out-of-order data in real time, and that its core ideas—event-time processing, strong consistency, and not waiting for completeness—remain strategically sound. However, it also concludes that the industry’s most practical advances came from database-style approaches like SQL, incremental view maintenance, and freshness contracts, suggesting that IT organizations should favor simpler, declarative analytics platforms over highly complex streaming machinery whenever possible. For CIOs and technology leaders, the implication is that real-time analytics strategy should center on business-relevant freshness, consistency, and operational simplicity rather than on streaming for its own sake.