在 sql 中连接行的最佳方式

Question

目前，我正在使用以下代码来汇总已审计固定资产的报告。起初，这没问题，但 3 年后我发现系统现在非常慢。我重新访问代码，发现该代码导致系统运行变慢的原因。当我删除子字符串部分时，系统运行更快。这附近有工作吗？

SELECT        
ENTITY
, [ASSET NUMBER]
, [YEAR AUDITED]
, SUM(COUNT) AS AUDITED
, SUBSTRING
    (
        (
            SELECT       
                ', ' + [SCANNED BY].USERNAME
            FROM  dbo.vwAUDITED as [SCANNED BY]
            WHERE vwAUDITED.[ASSET NUMBER] = [SCANNED BY].[ASSET NUMBER] 
                AND vwAUDITED.ENTITY = [SCANNED BY].ENTITY 
                AND vwAUDITED.[YEAR AUDITED] = [SCANNED BY].[YEAR AUDITED]
            ORDER BY [SCANNED BY].[ASSET NUMBER] FOR XML PATH('')
        ), 2, 1000
    ) AS [SCANNED BY]
, MAX(DATE) AS [COMPLETION DATE]
FROM dbo.vwAUDITED
GROUP BY ENTITY
    , [ASSET NUMBER]
    , [YEAR AUDITED]

Answer 1

一种方法是在您的 MS SQL 服务器版本中添加 user-defined aggregate function, as an alternative for the missing STRING_AGG 功能。然后用函数替换相关子查询。

可以找到此类函数的示例here and here

另一种不涉及数据库管理员的方法是使用临时 table，在用于加入子查询的字段上使用复合索引 FOR XML.

在 db<>fiddle here

上测试

-- Just assuming the datatypes here, so change them to the correct types.
IF OBJECT_ID('tempdb..#tmpAUDITED') IS NOT NULL DROP TABLE #tmpAUDITED;
CREATE TABLE #tmpAUDITED 
(
  [ENTITY] INT NOT NULL,
  [ASSET NUMBER] INT NOT NULL,
  [YEAR AUDITED] INT NOT NULL,
  [COUNT] INT,
  [DATE] DATE,
  [USERNAME] VARCHAR(100),
  INDEX idx_1 NONCLUSTERED ([ENTITY], [ASSET NUMBER], [YEAR AUDITED])
);

INSERT INTO #tmpAUDITED (
[ENTITY], [ASSET NUMBER],[YEAR AUDITED], [COUNT], [DATE], [USERNAME]
)
SELECT 
[ENTITY], [ASSET NUMBER],[YEAR AUDITED], [COUNT], [DATE], [USERNAME]
FROM dbo.vwAUDITED;

-- Now using the temp table instead of the view.
SELECT
 [ENTITY],
 [ASSET NUMBER],
 [YEAR AUDITED],
 SUM([COUNT]) AS AUDITED,
 SUBSTRING
     (( 
        SELECT ', ' + s.[USERNAME]
        FROM #tmpAUDITED AS s
        WHERE s.[ENTITY] = [AUDITED].[ENTITY]
          AND s.[ASSET NUMBER] = [AUDITED].[ASSET NUMBER] 
          AND s.[YEAR AUDITED] = [AUDITED].[YEAR AUDITED]
        -- ORDER BY s.[ENTITY], s.[ASSET NUMBER], s.[YEAR AUDITED], s.[USERNAME]
        FOR XML PATH('')
     ), 2, 1000) AS [SCANNED BY],
 MAX([DATE]) AS [COMPLETION DATE]
FROM #tmpAUDITED AS [AUDITED]
GROUP BY [ENTITY], [ASSET NUMBER], [YEAR AUDITED];

此外，检查视图中的查询是否可以优化可能是值得的。或者在 table 上添加某些索引（在视图中使用）可以提高视图本身的性能。

Answer 2

最好的连接方式是：

declare @str nvarchar(max) = space(0);
select @str += [column] from [yourTable];
select @str;

但是，您的情况有点不同。在您的情况下，连接做得很好，并且由于视图结构可能会导致性能下降。如果您看不到视图结构，您可以通过使用主视图记录集作为 CTE 来最大限度地减少性能问题，然后在 CTE 而不是视图上执行资源成本高昂的连接：

;with [data] as (
    select [entity], [asset number], [year audited], [username], [date], [count] from [dbo].[vwAUDITED]
)
select 
     [entity]           =   [d].[entity]
    ,[asset number]     =   [d].[asset number]
    ,[year audited]     =   [d].[year audited]
    ,[audited]          =   sum([d].[count])
    ,[scanned by]       =   substring((
                                select 
                                    ', ' + [username]
                                from 
                                    [dbo].[vwAUDITED]
                                where 
                                        [asset number]  =   [d].[asset number]
                                    and [entity]        =   [d].[entity]
                                    and [year audited]  =   [d].[year audited]
                                order by 
                                     [asset number]
                                for xml path('')
                            ), 2, 1000) 
    ,[completion date] =    max([d].[date])
from 
    [data] as   [d]
group by 
     [d].[entity]
    ,[d].[asset number]
    ,[d].[year audited];

在 sql 中连接行的最佳方式

best way to concatenate rows in sql

sql

substring

sql-server-2014