向主键添加列会消除唯一性

Question

我有大约 3400 万行，每行在 tpc-ds dataset 的 store_sales 表中有 23 列。

我有一个 composite primary key 列 ss_item_sk 和 ss_ticket_number。

在运行查询 SELECT count(DISTINCT <primary key>) .. 时，我可以看到它输出了表中存在的总行数。

现在我在primary key 旁边添加另一列，即ss_sold_date_sk。

在此之后，如果我运行countquery，我得到的打印行数比以前少。有人可以通过示例向我解释为什么会发生这种情况吗？

TL;DR

向复合主键添加列何时会停止使其唯一？

Answer 1

演示

create table mytable (c1 string,c2 string);
insert into mytable values ('A','A'),('B',null),('C',null);

---

select count(distinct c1) as count_distinct from mytable;

+----------------+
| count_distinct |
+----------------+
|              3 |
+----------------+

如预期 - 3 个不同的值 - 'A'、'B' 和 'C'

---

select count(distinct concat(c1,c2)) as count_distinct from mytable;

+----------------+
| count_distinct |
+----------------+
|              1 |
+----------------+
    

正如预期的那样。为什么？ - 查看下一个查询

---

select c1,c2,concat(c1,c2) as concat_c1_c2 from mytable;

+----+------+--------------+
| c1 |  c2  | concat_c1_c2 |
+----+------+--------------+
| A  | A    | AA           |
| B  | NULL | NULL         |
| C  | NULL | NULL         |
+----+------+--------------+

与 NULL 连接产生 NULL

---

select count(distinct c1,c2) as count_distinct from mytable;

+----------------+
| count_distinct |
+----------------+
|              1 |
+----------------+

BUG！！

---

这是一个解决该错误的方法：

select count(distinct struct(c1,c2)) as count_distinct from mytable;

+----------------+
| count_distinct |
+----------------+
|              3 |
+----------------+