sql-server - 如何在不使用 SQL Server 中的子查询的情况下连接 GROUP BY 子句中的字符串而无需额外的查询？

Question

我正在寻找 SQL Server 2012 中 GROUP_CONCAT() MySQL 函数的等效项 - 它不使用子查询，解释如下:

CREATE TABLE Temp
( 
ID INT PRIMARY KEY NOT NULL IDENTITY(1,1),
ColA varchar(900) NULL,
ColB varchar(900) NULL
)

INSERT INTO Temp (ColA, ColB)
SELECT 'A', 'some' UNION ALL
SELECT 'A', 'thing' UNION ALL
SELECT 'A', 'and' UNION ALL
SELECT 'B', 'some' UNION ALL
SELECT 'B', 'more' UNION ALL
SELECT 'B', 'and' UNION ALL
SELECT 'B', 'more' UNION ALL
SELECT 'C', 'things' UNION ALL
SELECT 'C', 'things'

-- Desired Output. Note that the lists are in descending order of frequency ('more' appears twice)
ColA, Frequency, ColBs
'B', 4, 'more, some, and'
'A', 3, 'some, thing, and'
'C', 2, 'things'

SELECT 
    ColA, 
    COUNT(*) as Frequency, 
    GROUP_CONCAT(ColB) --Would be nice
FROM Temp
GROUP BY ColA
ORDER BY Frequency DESC

在 SQL Server 中对此问题的常见答案是在子查询上使用 STUFF()。就我而言，性能根本无法接受(2 亿条记录，每个子查询 26 秒 * 2 亿 = 164 年)。

SELECT 
    ColA, 
    COUNT(*) as Frequency, 
    ISNULL(
        STUFF((
            SELECT ', ' + ColBs FROM
                (SELECT ColBs, Count(*) as Frequency
                FROM Temp sub
                WHERE sub.ColA = t.ColA
                GROUP BY ColB
                ORDER BY Frequency DESC)
            FOR XML PATH('')
        ), 1, 2, '')
    ), '') as ColBs --Would take 164 years on the entire data set
FROM Temp t
GROUP BY ColA
ORDER BY Frequency DESC

所需的输出是每个唯一 ColA 的 ColB 值，按出现次数降序分组在一起，如上所示。但是，这需要通过表进行单个查询来完成。

我需要自己构建这个并放弃“GROUP BY”调用吗？手动迭代数据集并在控制台应用程序中构建新表？还是我遗漏了什么？

Answer 1

试试这个:

WITH prelim
AS
(
   SELECT
     cola
    ,colb
    ,count(*) AS recs
    ,row_number() over (partition BY cola ORDER BY count(*) DESC ,colb) AS recno
    ,Count(*) over (partition BY cola ) AS cnt
  FROM TEMP
  GROUP BY cola,colb ),
Group_Concat (recno,cnt,recs,cola,colbs)
AS
(
SELECT
    recno
    ,cnt
    ,recs
    ,cola
    ,CAST (colb AS varchar(MAX)) AS colbs
FROM
    prelim
WHERE
    recno=1
UNION ALL
SELECT
    p.recno
    ,p.cnt
    ,g.recs+p.recs
    ,p.cola
    , g.colbs + ', ' + CAST (p.colb AS varchar(MAX)) AS colbs
FROM
    prelim p
    JOIN Group_Concat g ON p.cola=g.cola AND p.recno=g.recno+1
)

SELECT COLA,Recs as Frequency,COLBS 
FROM Group_Concat
where recno=cnt
order by cola