在 SQL 服务器中聚合动态列

Question

我有一个窄 table 包含唯一键和源数据

Unique_Key	System
1	IT
1	ACCOUNTS
1	PAYROLL
2	IT
2	PAYROLL
3	IT
4	HR
5	PAYROLL

我希望能够选择一个系统作为基础 - 在本例中为 IT - 然后在重要的地方创建一个动态 SQL 查询：

所选系统中不同的唯一键
与其他系统共享唯一密钥的比例。这些系统可能是动态的，而且数量远不止 4

我正在考虑使用动态 SQL 和 PIVOT 来首先挑选出 IT 之外的所有系统名称。然后以 IT 为基础，加入那个 table 以获取信息。

select distinct Unique_Key, System_Name
into #staging
from dbo.data
where System_Name <> 'IT'

DECLARE @cols AS NVARCHAR(MAX),
    @query  AS NVARCHAR(MAX);

SET @cols = STUFF((SELECT distinct ',' + QUOTENAME(System_Name) 
            FROM #staging
            FOR XML PATH(''), TYPE
            ).value('.', 'NVARCHAR(MAX)') 
        ,1,1,'')

set @query = 'SELECT Unique_Key, ' + @cols + ' into dbo.temp from 
            (
                select Unique_Key, System_Name
                from #staging
           ) x
            pivot 
            (
                 count(System_Name)
                for System_Name in (' + @cols + ')
            ) p '

execute(@query)

  select *
  from 
  (
    select distinct Unique_Key
  from dbo.data
  where System_Name = 'IT'
  ) a
  left join dbo.temp b
  on a.Unique_Key = b.Unique_Key

所以结果 table 是：

Unique_Key	PAYROLL	ACCOUNTS
1	1	1
2	1	0
3	0	0

我要的是更进一步：

Distinct Count IT Key	PAYROLL	ACCOUNTS	HR
3	67%	33%	0%

我可以用特定的 case when/sum 语句进行简单的连接，但想知道是否有办法动态地做到这一点，所以我不需要指定每个系统名称。

感谢任何tips/hints。

Answer 1

您可以尝试使用动态 SQL，如下所示，我会使用条件聚合函数获取枢轴值，然后我们可能会在动态 SQL 中添加 OUTER JOIN 或 EXISTS 条件。

我会使用 sp_executesql 而不是 exec 来避免 sql-injection。

DECLARE @System_Name NVARCHAR(50) = 'IT' 
DECLARE @cols AS NVARCHAR(MAX),
    @query  AS NVARCHAR(MAX),
    @parameter AS NVARCHAR(MAX);
    
SET @parameter = '@System_Name NVARCHAR(50)'

select DISTINCT System_Name
into #staging
from dbo.data t1
WHERE t1.System_Name <> @System_Name



SET @cols = STUFF((SELECT distinct ', SUM(IIF(System_Name = '''+ System_Name+''',1,0))  * 100.0 / SUM(IIF(System_Name = @System_Name,0,1)) ' + QUOTENAME(System_Name) 
            FROM #staging
            FOR XML PATH(''), TYPE
            ).value('.', 'NVARCHAR(MAX)') 
        ,1,1,'')
        
set @query = 'SELECT SUM(IIF(System_Name = @System_Name,0,1)) [Distinct Count IT Key], ' + @cols + ' from  dbo.data t1
 WHERE EXISTS (
  SELECT 1 
  FROM dbo.data tt 
  WHERE tt.Unique_Key = t1.Unique_Key
  AND tt.System_Name = @System_Name
 ) '
 
EXECUTE sp_executesql @query, @parameter,  @System_Name

sqlfiddle

Answer 2

这个方案改变了PIVOT本身的聚合函数。

首先，让我们向 #staging 添加一个列 [has_it]，以跟踪每个 Unique_Key 是否有一个 IT 行：

select Unique_Key, System_Name, case when exists(select 1 from data d2 where d2.Unique_Key=d1.Unique_Key and d2.System_Name='IT') then 1 else 0 end as has_it
into #staging
from data d1
where System_Name <> 'IT'
group by Unique_Key, System_Name

现在，此列的 per-System 聚合（总和）除以所需的最终唯一键总数（示例案例=3）returns 请求数。将 PIVOT 更改为以下内容并准备就绪 as-is，无需进一步查询：

set @query = '  select * 
    from 
        (
            select System_Name,cnt as [Distinct Count IT Key],has_it*1.0/cnt as divcnt
            from #staging
            cross join
            (   
                select count(distinct Unique_Key) as cnt
                from dbo.data
                where System_Name = ''IT''
            )y
        ) x
        pivot 
        (
                sum(divcnt)
            for System_Name in (' + @cols + ')
        ) p'

Answer 3

编写动态查询时，您从 non-dynamic 查询开始。在转换为动态查询之前，请确保您获得的查询结果是正确的。

对于您需要的结果，查询将是

with cte as
(
    select it.Unique_Key, ot.System_Name
    from   data it
           left join data ot on it.Unique_Key = ot.Unique_Key
                            and ot.System_Name <> 'IT'
    where  it.System_Name = 'IT'
)
select  [ITKey]    = count(distinct Unique_Key), 
        [ACCOUNTS] = count(case when System_Name = 'ACCOUNTS' then Unique_Key end) * 100.0 
                   / count(distinct Unique_Key),
        [HR]       = count(case when System_Name = 'HR'       then Unique_Key end) * 100.0 
                   / count(distinct Unique_Key),
        [PAYROLL]  = count(case when System_Name = 'PAYROLL'  then Unique_Key end) * 100.0 
                   / count(distinct Unique_Key)
from   cte;

一旦得到正确的结果，转换为动态查询就不难了。使用 string_agg() 或 xml 那些重复行的路径

declare @sql nvarchar(max);

; with cte as
(
    select distinct System_Name
    from   data
    where  System_Name <> 'IT'
)
select  @sql = string_agg(sql1 + ' / ' + sql2, ',' + char(13))
from    cte
        cross apply
        (
            select sql1 = char(9) + quotename(System_Name) + ' = '
                        + 'count(case when System_Name = ''' + System_Name + ''' then Unique_Key end) * 100.0 ',
                   sql2 = 'count(distinct Unique_Key)'
        ) a

select  @sql = 'with cte as' + char(13)
             + '(' + char(13)
             + '    select it.Unique_Key, ot.System_Name' + char(13)
             + '    from   data it' + char(13)
             + '           left join data ot on it.Unique_Key = ot.Unique_Key' + char(13)
             + '                            and ot.System_Name <> ''IT''' + char(13)
             + '    where  it.System_Name = ''IT''' + char(13)
             + ')' + char(13)
             + 'select [ITKey] = count(distinct Unique_Key), ' + char(13)
             + @sql + char(13)
             + 'from   cte;' + char(13)

print @sql;

exec sp_executesql @sql;

db<>fiddle demo

在 SQL 服务器中聚合动态列

Aggregate dynamic columns in SQL Server

sql

sql-server