A streaming application is not just about doing some real-time computations on a stream of data. Generally, streaming will be part of a larger application that includes real-time, batch, and serving layers with machine learning, and so on. A continuous application is an end-to-end application that combines all these features in one application.
In Spark 2.0, the Structured Streaming API is introduced for building continuous applications. The Structured Streaming API addresses the following concerns of a typical streaming application:
Node delays: Delay in a specific node can cause data inconsistency at the database layer. Ordering the guarantee of events is achieved using systems like Kafka in which events on the same key always go to the same Kafka partition. Streaming applications pull data from this partition and process them in the order they are received. However, while applying operations on streaming data, if one of the node delays processing, data...