Grouping followed by count, reduce or aggregate builds a KTable backed by a store: genre-revenue keeps a running total per genre, genre-daily one per one-day tumbling window. Windows use event time (order_ts) and accept late records for a one-hour grace period. Hopping windows overlap, sliding ones group records close in time, session windows end after a gap.
KStream-KStream joins need a time window and buffer both sides. Stream-table joins look up the table's current value as each event arrives, so a later catalog change does not revisit old orders. KTable-KTable joins, including foreign-key joins, maintain a result table.