Chapter 3: Watermarks

Watermark

"The watermark is a monotonically increasing timestamp of the oldest work not yet completed."

Watermarks assert that all earlier events have (probably) arrived.

Perfect Watermark

When stream elements are in order (or in order by key), we can assert that a watermark at some time is accurate.

Heuristic Watermark

Watermarks determined by a set of rules to capture potentially late data from heterogeneous sources.

An Example

Propagation

Propagation Example

Percentile Watermarks

Processing time Watermarks

Flink Watermark Model

Flink Kafka Example

Fin

Chapter 3: Watermarks

By Jowanza Joseph

Chapter 3: Watermarks

  • 636