我有一张这样的table

+-----+-----+-----+-----+
|att1 | att2| att3|  id |
+-----------------------+
| a   |  b  |  a  |  1  |
| b   |  c  |  c  |  2  |
| a   |  b  |  a  |  3  |
+-----------------------+

如果att1、att2和att3相等，我想用一个相等的新id替换该id，如下所示：

+-----+-----+-----+-----+
|att1 | att2| att3|  id |
+-----------------------+
| a   |  b  |  a  |  1  |
| b   |  c  |  c  |  2  |
| a   |  b  |  a  |  1  |
+-----------------------+

我尝试了经典的merge-into语句，在属性上使用密集的\u rank（）函数，如下所示：

merge into the_table t
using (
  select att1, att2, att3, dense_rank() over (partition by att1, att2, att3 order by att1) as rnk
  from the_table 
) x on (x.att1 = t.att1 and x.att2 = t.att2 and t.att3 = x.att3)
when matched then update
   set old_id = x.new_id;

但这行不通，因为 select att1, att2, att3, dense_rank() over (partition by att1, att2, att3 order by att1) as rnk 并没有像我预期的那样给att1、att2和att3的每个分区分配自己的等级，而是所有的分区都得到了等级1。
我做错什么了？