在 SQL Server 中，有没有一种方法可以在聚合时使用 GROUPING SETS 删除重复项，而无需进行巨大的基数扩展？

Question

一个简化的（有点傻）的例子将使我的要求更清楚一些。 Bob/Tim 应该是这里的唯一标识符。

DROP TABLE IF EXISTS #People

CREATE TABLE #People 
(
    Person varchar(10), 
    City varchar(20), 
    Department varchar(1), 
    PersonAge int
);

INSERT INTO #People VALUES ('Bob', 'New York', 'A', 40),
                           ('Tim', 'New York', 'A', 30),
                           ('Tim', 'New York', 'B', 30)

;WITH InitialGrouping AS 
(
    SELECT 
        Person,
        City,
        Department,
        MAX(PersonAge) PersonAge,
        COUNT(*) NumRows
    FROM
        #People
    GROUP BY 
        Person, City, Department
) --SELECT * FROM InitialGrouping
SELECT 
    City,
    Department,
    AVG(PersonAge) AveragePersonAge,
    COUNT(DISTINCT Person) PersonCount
FROM
    InitialGrouping
GROUP BY 
    GROUPING SETS (City, Department, (City, Department))
ORDER BY 
    City, Department

;WITH InitialGrouping AS 
(
    SELECT 
        GROUPING_ID(Person,City,Department) GROUPID,
        Person,
        City,
        Department,
        MAX(PersonAge) PersonAge,
        COUNT(*) NumRows
    FROM
        #People
    GROUP BY  
        Person, GROUPING SETS (City, Department, (City, Department))
) --SELECT * FROM InitialGrouping
SELECT 
    City,
    Department,
    AVG(PersonAge) AveragePersonAge,
    COUNT(DISTINCT Person) PersonCount
FROM 
    InitialGrouping
GROUP BY 
    GROUPID, City, Department
ORDER BY 
    City, Department

这两个例子在功能上并不等价。如果您查看纽约的汇总行，您会发现第一个 (33) 的平均年龄是错误的，而第二个 (35) 的平均年龄是正确的。第二个在功能上是正确的，但是，对于具有大量分组集的更复杂的示例，它绝对可以在查询计划的串联步骤中爆炸基数，从而导致性能无法接受。在这个简单的示例中，您可以看到第一个查询将内容保持在 5 行，而第二个查询扩展到 8 行。在具有大量分组集的更复杂的情况下，我已经看到基数在数百万行中扩展了很多次。

有没有办法解决这个问题，防止事情扩展超过最终输出的目标粒度？

Answer 1

我使用第一种方式，但更改方法类型顺序聚合并使用窗口函数

 select  

    City,
    Department,
    max(PersonAge) AveragePersonAge,
    max( Person) PersonCount 

from (

    select 
             City
            ,Department
            ,AVG(PersonAge) over(partition by City,Department )  as PersonAge
            ,count( *) over(partition by City,Department )  as Person    

     from #People
)a
GROUP BY 
    GROUPING SETS (Department,City,(Department,City) )
ORDER BY 
    City, Department