If the distribution key column in a target table contains null values and you configure a mapping with an upsert operation for the same target table, the mapping might create duplicate rows. To avoid creating duplicate rows, you must perform one of the following tasks:

Replace the null value with a non-null value when you load data.
Do not configure the column as a distribution key if you expect null values in the distribution key column.
Remove the distribution key column from the target table temporarily when you load data.