Fifteen-Minute Buckets — Solution & ExplanationProblems / Fifteen-Minute Buckets / EditorialApproach

Floor each timestamp to its 15-minute window with time_bucket(INTERVAL '15 minutes', event_time). The function returns the window *start*, so 14:07 -> 14:00 and 14:59:59 -> 14:45, while a timestamp already on a boundary (14:15:00) maps to itself.

Group and count. GROUP BY bucket_start with COUNT(*) gives events per bucket; only buckets with data appear, which is exactly what the problem asks (no zero-fill).

Order by bucket_start.

The arithmetic alternative, date_trunc('hour', ts) + INTERVAL 15 MINUTE * (EXTRACT(minute FROM ts) // 15), does the same thing and shows what the built-in encapsulates: truncate to the hour, then add back the floored quarter-hours.Key InsightBucketing timestamps is *flooring*, not rounding — 14:22:59 belongs to 14:15, never 14:30. Getting the boundary semantics right (14:15:00 stays in 14:15) matters because downstream joins on bucket_start silently misalign if two systems round differently. In PySpark-on-DuckDB, F.expr lets you reach the same native time_bucket function when no built-in F equivalent exists.Hints

Flooring to 15 minutes means keeping the hour and replacing the minutes with the largest multiple of 15 that is <= the actual minutes; seconds go to zero.

DuckDB has this built in: time_bucket(INTERVAL '15 minutes', event_time) returns the start of the window a timestamp falls in.

Without the built-in you can construct it: date_trunc('hour', event_time) + INTERVAL 15 MINUTE * (EXTRACT(minute FROM event_time) // 15).

Watch the boundaries: 14:15:00 must land in the 14:15 bucket, and 14:59:59 in the 14:45 bucket.

Solve Fifteen-Minute Buckets yourself →