A streaming job uses event-time windows, and events sometimes arrive several minutes late. What does the watermark tell the pipeline?
Show answer and explanation
Answer: A
A watermark is the system's estimate of event-time progress. It asserts that input with event timestamps before a given point is believed to have arrived. Once the watermark passes a window's end, the runner can emit that window's results, and later elements are treated as late, subject to allowed lateness settings. The watermark is not a storage write time, an encryption key identifier, or a billing signal. Because it is an estimate, it can be wrong, which is why late data handling exists.
Why the other options are wrong
- B. A bucket write time is a processing moment, not an event-time progress marker. Watermarks depend on event timestamps, so this option describes a different concept.
- C. Encryption key versions identify which key protected data. They have no role in tracking how far event time has progressed through a streaming pipeline.
- D. Cost thresholds are billing or scaling policies, not timing markers in the data. A watermark describes event-time progress, not spend.