Airflow will run tasks again: retries after a failure, a clear after a fix, a backfill over days that already ran. A task is idempotent when running it twice for the same logical date leaves the same result as running it once. An append-only load is not: each rerun adds the day's rows again, and revenue doubles.
The usual pattern is delete then insert for the run's own slice: delete the rows for ds, then insert that day's rows. Only the slice is touched, so other days stay as they are. Other forms are a merge (upsert) on a key, or writing a file whose name contains the date, which a rerun overwrites.
Two more habits: derive dates from the run (ds, the data interval), never from datetime.now(), and make each task's output depend only on its inputs and its date.
Delete the run's day, then insert it: a rerun replaces that day and nothing else.