SQL SERVER - 有没有办法循环表中的列并在另一个 select 语句的 where 子句中使用它？答案

【问题标题】：SQL SERVER - is there a way to loop a column in a table and use that in the where clause of another select statement?SQL SERVER - 有没有办法循环表中的列并在另一个 select 语句的 where 子句中使用它？
【发布时间】：2021-12-10 08:59:41
【问题描述】：

我正在尝试在我的数据库中获取分层数据。这是我的初始代码。

        SELECT TOP 5
            year, location_state, location_city,
            COUNT(tf.customer_key) Number_of_Customers
        FROM TransactionFact tf
        JOIN LocationDim as ld
            ON ld.location_key = tf.seller_location_key
        JOIN DateDim dd
            ON dd.date_key = tf.order_date_key
        WHERE dd.year = 2016 and location_state = 'SP'
        GROUP BY dd.year, ld.location_state, ld.location_city
        ORDER BY dd.year DESC, Number_of_Customers DESC

这是结果，result。

基本上，在查询中，我想做的是不对WHERE 子句中的location_state 进行硬编码。我想让它动态化，以便我得到每个州的前 5 个 城市。 p>

这是 LocationDim 表的列名

location_key
location_zip_code_prefix
location_state
location_city

编辑：我需要的是这样的东西。

+------+----------------+---------------+---------------------+
| year | location_state | location_city | Number_of_Customers |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_1     |    city_1     |       100           |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_1     |    city_2     |       90            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_1     |    city_3     |       89            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_1     |    city_4     |       88            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_1     |    city_5     |       20            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_2     |    city_1     |       100           |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_2     |    city_2     |       45            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_2     |    city_3     |       23            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_2     |    city_4     |       10            |
+------+----------------+---------------+---------------------+
| 2016 |    STATE_2     |    city_5     |       5             |
+------+----------------+---------------+---------------------+

PS：抱歉，这是我在 stackoverflow 中的第一个问题。如果这个问题是重复的，请删除链接，我会试一试。提前谢谢你。

【问题讨论】：

这能回答你的问题吗？ Get top 1 row of each group
@Charlieface，我想我很难读懂那个，因为我不知道 CTE 是如何工作的。我仍然需要弄清楚这一点。我希望有一个更简单的方法。谢谢
标准答案与@LukStorms 答案几乎相同：只需添加一个分区行号并按此过滤即可。您将整个查询放入带有行号的派生表中，然后在该表之外进行过滤（因为您无法从同一级别过滤ROW_NUMBER）。旁白：COUNT(tf.customer_key) 与COUNT(*) 相同，它只计算非空行。如果您想要不同 customer_key 的计数，则需要 COUNT(DISTINCT tf.customer_key)
@Charlieface 是的，它解决了我的问题。我已经对他的回答投了赞成票，但我不确定为什么他的回答中没有绿色。

标签： sql-server tsql

【解决方案1】：

如果您计算 row_number，则可以对其进行过滤。

SELECT *
FROM 
(
    SELECT 
      dd.[year]
    , ld.location_state
    , ld.location_city
    , COUNT(tf.customer_key) AS total_customers
    , rn = ROW_NUMBER() OVER (PARTITION BY ld.location_state, dd.[year] 
                              ORDER BY COUNT(tf.customer_key) DESC)
    FROM TransactionFact AS tf
    JOIN LocationDim AS ld
      ON ld.location_key = tf.seller_location_key
    JOIN DateDim AS dd
      ON dd.date_key = tf.order_date_key
    WHERE dd.[year] = 2016
    GROUP BY ld.location_state, ld.location_city, dd.[year] 
) q
WHERE rn <= 5
ORDER BY location_state, [year], rn

【讨论】：

@LukeStorms 你能解释一下 rn 是如何工作的吗？因为这确实解决了我的问题。
它为每个 PARTITIONend 列生成一个没有间隙的序列号。同样流行的是 DENSE_RANK，它将按顺序为相同的值分配相等的数字。 F.e.具有相同 total_customers 的 2 将获得相同的 dense_rank，但不同的 row_number。
@LukeStorms，好的，谢谢。我想我必须阅读更多关于该功能的内容。我已经赞成您的解决方案。谢谢！
它的工作原理基本上是因为window functions这样的东西是在group by之后处理的。

【解决方案2】：

您可以使用ROW_NUMBER() OVER (PARTITION BY ORDER BY )添加行号

下面的查询按location_state分区记录，并以number_of_customer顺序添加行号：

Select * from 
inn.*,  ROW_NUMBER() OVER (PARTITION BY location_state ORDER BY Number_of_Customers DESC) AS rn
( 
SELECT year,
       location_state,
       location_city,
       COUNT(tf.customer_key) Number_of_Customers
  FROM TransactionFact tf
  JOIN LocationDim as ld
    ON ld.location_key = tf.seller_location_key
  JOIN DateDim dd
    ON dd.date_key = tf.order_date_key
 WHERE dd.year = 2016
 GROUP BY dd.year, ld.location_state, ld.location_city
 ) inn

在此之后，您可以轻松过滤并选择top5或任何其他顺序...

注意：我将您的查询用作内部查询。由于没有小提琴，我没有机会测试它

【讨论】：

【解决方案3】：

您可以尝试使用 cross apply 运算符将每个 location_state 的值传递给您的查询，如下所示：

;With STATES As (
SELECT location_state 
FROM LocationDim 
GROUP BY location_state)
SELECT T.[year], T.location_state, T.location_city, T.Number_of_Customers
FROM STATES CROSS APPLY (
            SELECT TOP 5
                [year], location_state, location_city,
                COUNT(tf.customer_key) AS Number_of_Customers
            FROM TransactionFact tf
            JOIN LocationDim as ld
                ON ld.location_key = tf.seller_location_key
            JOIN DateDim dd
                ON dd.date_key = tf.order_date_key
            WHERE dd.[year] = 2016 and location_state = STATES.location_state
            GROUP BY dd.[year], ld.location_state, ld.location_city
            ORDER BY dd.[year] DESC, Number_of_Customers DESC) As T

【讨论】：

【解决方案4】：

所以，基本上这里发生了什么：

    SELECT TOP 5
        dd.[year], ld.location_state, ld.location_city, COUNT(tf.customer_key) Number_of_Customers
    FROM TransactionFact tf
    JOIN LocationDim as ld
        ON ld.location_key = tf.seller_location_key
    JOIN DateDim dd
        ON dd.date_key = tf.order_date_key
    WHERE dd.year = 2016 and location_state = 'SP'
    GROUP BY dd.year, ld.location_state, ld.location_city
    ORDER BY dd.year DESC, Number_of_Customers DESC;

您是否只选择该查询的前 5 个结果。

但你想要的是得到所有结果，按年份和州排名，并且只取每个州的前 5 名？

我会使用RANK() function，它是专为您正在查看的场景而设计的。我将在您的查询中将其显示为添加的列：

  SELECT * FROM ( SELECT dd.[year], ld.location_state, ld.location_city, COUNT(tf.customer_key) Number_of_Customers,
           RANK() OVER(PARTITION BY dd.[year], ld.location_state, ld.location_city 
                       ORDER BY Number_of_Customers DESC) r
    FROM TransactionFact tf
    JOIN LocationDim as ld
        ON ld.location_key = tf.seller_location_key
    JOIN DateDim dd
        ON dd.date_key = tf.order_date_key
    WHERE dd.year = 2016 and location_state = 'SP' 
    GROUP BY dd.year, ld.location_state, ld.location_city
     ) x WHERE x.r <= 5
     ORDER BY x.[year] desc, x.location_state, x.r

或者，您可以在应用RANK 之前使用CTE（公用表表达式）来保存第一个查询的结果：

;WITH cte AS(
   SELECT dd.[year], ld.location_state, ld.location_city, COUNT(tf.customer_key) Number_of_Customers, 
       RANK() OVER(PARTITION BY [year], location_state, location_city
                   ORDER BY Number_Of_Customers DESC) r
   FROM TransactionFact tf
        JOIN LocationDim as ld
            ON ld.location_key = tf.seller_location_key
        JOIN DateDim dd
            ON dd.date_key = tf.order_date_key
   WHERE dd.year = 2016 and location_state = 'SP'
   GROUP BY dd.year, ld.location_state, ld.location_city
)
SELECT *
FROM cte
WHERE r <= 5;

作为免责声明，我只是在WITH之前放置一个分号，以表明如果事先有一个不以分号结尾的语句，那么该语句将引发错误。

编辑：添加，使用RANK 意味着结果按值排序。因此，如果两个城市有 30,000 名客户，他们将获得相同的 RANK 值（类似于他们在一轮高尔夫球中打成平手时在排行榜上所做的事情）。这意味着您将从每个州获得至少 5 个结果 - 如果您想要正好 5 个，而不管绑定值如何，那么您可以以相同的方式使用 ROW_NUMBER。

【讨论】：

它抛出一个错误“无效的列名'r'”。我猜它不知道where 中的“r”，即使 r 是在 RANK() 行之后定义的
刚刚编辑将 RANK 放在内部查询中，将 WHERE 放在外部。