Sensors emit noise levels irregularly; you must compute a sensor's average when there’s continuous data for >30 minutes, but end the window after 15 minutes of inactivity. Which windowing do you use?

Session windows are ideal for handling irregular, event-driven data streams like sensor noise. They group elements that arrive within a specified gap duration, effectively defining a "session" of activity. Here, a 15-minute gap duration means that if no new data arrives for 15 minutes, the current session window closes. This perfectly matches the requirement to end the window after 15 minutes of inactivity. The condition of needing continuous data for 30 minutes before computing the average is handled by the aggregation logic within the session window, not by the windowing itself. Using a 30-minute gap duration would keep sessions open too long, violating the 15-minute inactivity rule. Hopping windows and tumbling windows are fixed-time-interval windows and are not suitable for dynamically grouping irregular data based on inactivity. Tumbling windows with withAllowedLateness are for handling late-arriving data in fixed windows, not for defining activity-based windows.