T-SQL - 连接多个表会导致重复行
Posted
技术标签:
【中文标题】T-SQL - 连接多个表会导致重复行【英文标题】：T-SQL - Joining multiple tables results in duplicate rows 【发布时间】：2018-12-12 22:56:48 【问题描述】：我正在努力从下面的示例Fiddle 中得到以下结果。
结果：
源表：
Fiddle
这个想法是通过查找预算和预测 ID 的最大计数来为每个 RecordID 创建一行以创建其他列（如果该预算没有值，则将列单元格留空或预测 ID 存在）。
我尝试使用 PIVOT 函数，但甚至无法获得不错的结果。
更新： 请看下图，我试图解释预期的输出：
用词：对于每个 BudgetID 属于 RecordID 为 BDate 创建单独列、结果（百分比 * BAmount 记录表）和状态。
在我的示例中，RecordID 55 在预算表中有 两个 条目 - 因此需要 2x3 列来显示每个日期、result 和 status 在此 RecordID 的单行 中的自己的列中。
由于 RecordID 77 在预算表中具有最多（三个）条目，因此它用于在所有行上创建 3x3 列。
Forecast 也是如此。
我希望你能帮助我。
谢谢。
【问题讨论】：
您能否为此添加预期输出 我不清楚 详细解释请看更新。 您必须将我的答案更改为添加column names。你需要什么列。让我知道你得到了什么......！
SQL 没有根据新数据添加新列的语法。您是否知道每个 RecordID 总是最多有 3 个预算 ID？如果是这样，您可以编写一个透视样式查询来添加 Result1、Result2 和 Result3 列。如果没有，并且在运行查询之前不知道需要多少 Result 列，那么您必须编写动态 SQL。也就是说，读取数据的代码，然后根据数据编写新查询。
【参考方案1】：
早安，
注意！我的目标是帮助读者学习，而不是提供最终查询作为其作业的解决方案。因此，我将分两步介绍解决方案，并在动态解决方案中添加几个“PRINT”命令，以便读者可以选择检查工作中的中间步骤。
OP提供的DDL+DML：
DROP TABLE IF EXISTS Budget;
CREATE TABLE Budget
(BudgetID int, RecordID int, BDate date,Percentage int, [Status] varchar(50));
INSERT INTO Budget
(BudgetID, RecordID, BDate,Percentage,Status)
VALUES
(1, 55, '2017-01-01', 60, 'ordered'),
(2, 55, '2017-03-24', 40, 'ordered'),
(3, 66, '2018-08-15', 100, 'invoiced'),
(4, 77, '2018-12-02', 25, 'paid'),
(5, 77, '2018-09-10', 35, 'ordered'),
(6, 77, '2019-07-13', 40, 'ordered')
GO
DROP TABLE IF EXISTS Forecast;
CREATE TABLE Forecast
(ForecastID int, RecordID int, FDate date, Percentage int);
INSERT INTO Forecast
(ForecastID, RecordID, FDate,Percentage)
VALUES
(1, 55, '2020-12-01', 100),
(2, 77, '2023-05-17', 25),
(3, 77, '2024-11-28', 75)
GO
DROP TABLE IF EXISTS Records;
CREATE TABLE Records
(RecordID int, BAmount int, FAmount int, Name varchar(40), Description varchar(40) )
;
INSERT INTO Records
(RecordID, BAmount,FAmount,Name, Description)
VALUES
(55, 15000, 33000, 'Prod1', 'Desc1' ),
(66, 22000, 17500, 'Prod2', 'Desc2' ),
(77, 40000, 44000, 'Prod3', 'Desc3' )
GO
select * from Budget
select * from Forecast
select * from Records
我们先展示一个简单的静态解决方案
这是基于这样的知识：预算表中的每个 RecordID 最多三行，而预测表中的每个 RecordID 最多两行。这将有助于理解我接下来要展示的动态解决方案
;With CteBudget as (
select
b.BDate, b.BudgetID, b.Percentage, b.RecordID, b.Status
,RN = ROW_NUMBER() OVER (partition by b.RecordID order by b.BudgetID)
from Budget b
),
CteForecast as (
select
f.FDate, f.ForecastID, f.Percentage, f.RecordID
,RN = ROW_NUMBER() OVER (partition by f.RecordID order by f.ForecastID)
from Forecast f
)
select
r.RecordID, r.Name, r.Description,
b1.BDate BDate1, (b1.Percentage * r.BAmount)/100 BResult1, b1.Status BStatus1,
b2.BDate BDate2, (b2.Percentage * r.BAmount)/100 BResult2, b2.Status BStatus2,
b3.BDate BDate3, (b3.Percentage * r.BAmount)/100 BResult3, b3.Status BStatus3,
f1.FDate FDate1, (f1.Percentage * r.BAmount)/100 FResult1,
f2.FDate FDate2, (f2.Percentage * r.BAmount)/100 FResult2
from Records r
left join CteBudget b1 on r.RecordID = b1.RecordID and b1.RN = 1
left join CteBudget b2 on r.RecordID = b2.RecordID and b2.RN = 2
left join CteBudget b3 on r.RecordID = b3.RecordID and b3.RN = 3
left join CteForecast f1 on r.RecordID = f1.RecordID and f1.RN = 1
left join CteForecast f2 on r.RecordID = f2.RecordID and f2.RN = 2
--where r.RecordID = 77
GO
注意！ 对于静态解决方案并且没有索引（我将在最后添加），上述解决方案在性能方面非常糟糕，但是一旦我们添加了正确的索引并作为基础动态解决方案这个选项应该很合适。
现在我们可以提出动态解决方案了。
-- Get number of columns
Declare @NumBudget tinyint
Declare @NumForecast tinyint
SELECT @NumBudget = MAX(C) FROM (
select COUNT(RecordID) C
from Budget
GROUP BY RecordID
) t
SELECT @NumForecast = MAX(C) FROM (
select COUNT(RecordID) C
from Forecast
GROUP BY RecordID
) t
---------------------------------------------
DECLARE @SQLString1 nvarchar(MAX) = '';
DECLARE @SQLString2 nvarchar(MAX) = '';
DECLARE @loop int = 1;
WHILE @loop <= @NumBudget BEGIN
SET @SQLString1 = @SQLString1 + N'
b' + CONVERT(VARCHAR(2),@loop) + '.BDate BDate' + CONVERT(VARCHAR(2),@loop) + ', (b' + CONVERT(VARCHAR(2),@loop) + '.Percentage * r.BAmount)/100 BResult' + CONVERT(VARCHAR(2),@loop) + ', b' + CONVERT(VARCHAR(2),@loop) + '.Status BStatus' + CONVERT(VARCHAR(2),@loop) + ', '
SET @SQLString2 = @SQLString2 + N'
left join CteBudget b' + CONVERT(VARCHAR(2),@loop) + ' on r.RecordID = b' + CONVERT(VARCHAR(2),@loop) + '.RecordID and b' + CONVERT(VARCHAR(2),@loop) + '.RN = 1'
SET @loop = @loop + 1
END
SET @loop = 1
WHILE @loop <= @NumForecast BEGIN
SET @SQLString1 = @SQLString1 + N'
f' + CONVERT(VARCHAR(2),@loop) + '.FDate FDate' + CONVERT(VARCHAR(2),@loop) + ', (f' + CONVERT(VARCHAR(2),@loop) + '.Percentage * r.BAmount)/100 FResult' + CONVERT(VARCHAR(2),@loop) + ','
SET @SQLString2 = @SQLString2 + N'
left join CteForecast f' + CONVERT(VARCHAR(2),@loop) + ' on r.RecordID = f' + CONVERT(VARCHAR(2),@loop) + '.RecordID and f' + CONVERT(VARCHAR(2),@loop) + '.RN = 1'
SET @loop = @loop + 1
END
SET @SQLString1 = STUFF (@SQLString1, LEN(@SQLString1) , 1 , '')
PRINT '/************************************************/'
PRINT @SQLString1
PRINT @SQLString2
PRINT '/************************************************/'
DECLARE @SQLString nvarchar(MAX);
SET @SQLString = N'
;With CteBudget as (
select
b.BDate, b.BudgetID, b.Percentage, b.RecordID, b.Status
,RN = ROW_NUMBER() OVER (partition by b.RecordID order by b.BudgetID)
from Budget b
),
CteForecast as (
select
f.FDate, f.ForecastID, f.Percentage, f.RecordID
,RN = ROW_NUMBER() OVER (partition by f.RecordID order by f.ForecastID)
from Forecast f
)
select
r.RecordID, r.Name, r.Description,
'
+ @SQLString1
+ N'
from Records r'
+ @SQLString2
print @SQLString
EXECUTE sp_executesql @SQLString
GO
重要提示！ 此解决方案不一定是性能最佳的解决方案，但可能是最容易遵循和理解的解决方案。在生产中，一旦我们将拥有真正的 DDL+DML 并且服务器将拥有统计信息，那么我们将能够提高性能并为我们的特定情况选择最佳解决方案。
索引
注意！ 上述解决方案可能会导致对数据进行大量排序，在这里拥有正确的索引非常重要！测试几个不同的选项并选择最佳选项非常重要。
为了论坛（或任何你称之为 *** 的东西，我认为它不是讨论论坛界面而是问答界面），我添加了查询来创建我假设你在生产中拥有的 CLUSTERED INDEX，以及一个可选的你应该测试的非集群索引（我没有测试其他选项，这意味着我想到的第一个，所以建议使用真正的 DDL+DML 继续检查正确的索引）。
-- CLUSTERED INDEX
CREATE CLUSTERED INDEX IX_Budget_BudgetID
ON dbo.Budget (BudgetID);
GO
CREATE CLUSTERED INDEX IX_Forecast_ForecastID
ON dbo.Forecast (ForecastID);
GO
CREATE CLUSTERED INDEX IX_Records_RecordID
ON dbo.Records (RecordID);
GO
-- NONCLUSTERED INDEX
CREATE NONCLUSTERED INDEX NX_Budget_RecordID_BudgetID
ON dbo.Budget (RecordID,BudgetID);
GO
CREATE NONCLUSTERED INDEX NX_Forecast_RecordID_ForecastID
ON dbo.Forecast (RecordID,ForecastID);
GO
CREATE NONCLUSTERED INDEX NX_Records_RecordID_RecordID
ON dbo.Records (RecordID);
GO
【讨论】：
非常感谢您的详细解答。我能够根据生产数据调整样本。 欢迎您@ratanmalko。我很高兴听到我可以提供帮助:-)【参考方案2】：我想我会这样做：
select Max(BudgetId) as BudgetID, RecordID
into #MBudget
from Budget
group by RecordID
select B.*
into #MaxB
from #MBudget M
inner join Budget B
on M.BudgetID = B.BudgetID
and M.RecordID = B.RecordID
--The above will then only have the "maximum" BudgetID data from the table.
-- You need to then do the same with the Forecast table
select Max(ForecastId) as ForecastID, RecordID
into #MForecast
from Forecast
group by RecordID
select F.*
into #MaxF
from #MForecast M
inner join Forecast F
on M.ForecastID = F.ForecastID
and M.RecordID = F.RecordID
-- Join them together on the RecordID
select * -- you will need to pick the required fields
from #MaxF F
inner join #MaxB B
on F.RecordID = B.RecordID
inner join Record R
on F.RecordID = R.RecordID
我认为这会让你到达那里，我不是小提琴的忠实粉丝，所以在我的脑海中也是如此。
我确实有点同意这些家伙的观点，你本来可以更清楚，但希望这能让你到达你需要的地方。
【讨论】：
【参考方案3】：最后，我们将动态查询创建为多重连接。
输入表：
CREATE TABLE #Budget (BudgetID int, RecordID int, BDate date,Percentage int, [Status] varchar(50));
INSERT INTO #Budget(BudgetID, RecordID, BDate,Percentage,Status)
VALUES(1, 55, '2017-01-01', 60, 'ordered'),(2, 55, '2017-03-24', 40, 'ordered'),(3, 66, '2018-08-15', 100, 'invoiced'),(4, 77, '2018-12-02', 25, 'paid'),(5, 77, '2018-09-10', 35, 'ordered'),(6, 77, '2019-07-13', 40, 'ordered')
CREATE TABLE #Forecast(ForecastID int, RecordID int, FDate date, Percentage int)
INSERT INTO #Forecast(ForecastID, RecordID, FDate,Percentage)
VALUES(4, 77, '2018-07-18', 24),(1, 55, '2020-12-01', 100),(2, 77, '2023-05-17', 25),(3, 77, '2024-11-28', 75)
CREATE TABLE #Records(RecordID int, BAmount int, FAmount int, Name varchar(40), Description varchar(40) )
INSERT INTO #Records(RecordID, BAmount,FAmount,Name, Description)
VALUES(55, 15000, 33000, 'Prod1', 'Desc1' ),(66, 22000, 17500, 'Prod2', 'Desc2' ),(77, 40000, 44000, 'Prod3', 'Desc3' )
最终加入查询
select * from (select r1.RecordID, f1.FDate fdate1, b1.BDate bdate1
, ROW_NUMBER() over(partition by r1.recordid order by (select NULL)) rn from #Records r1
join #Forecast f1
on r1.RecordID = f1.RecordID
and r1.RecordID = 77
join #Budget b1
on r1.RecordID = b1.RecordID
)t1 join (select r2.RecordID, f2.FDate fdate2, b2.BDate bdate2
, ROW_NUMBER() over(partition by r2.recordid order by (select NULL)) rn from #Records r2
join #Forecast f2
on r2.RecordID = f2.RecordID
and r2.RecordID = 77
join #Budget b2
on r2.RecordID = b2.RecordID
)t2 on t1.RecordID = t2.RecordID and t1.rn < t2.rn join (select r3.RecordID, f3.FDate fdate3, b3.BDate bdate3
, ROW_NUMBER() over(partition by r3.recordid order by (select NULL)) rn from #Records r3
join #Forecast f3
on r3.RecordID = f3.RecordID
and r3.RecordID = 77
join #Budget b3
on r3.RecordID = b3.RecordID
)t3 on t2.RecordID = t3.RecordID and t2.rn < t3.rn join (select r4.RecordID, f4.FDate fdate4, b4.BDate bdate4
, ROW_NUMBER() over(partition by r4.recordid order by (select NULL)) rn from #Records r4
join #Forecast f4
on r4.RecordID = f4.RecordID
and r4.RecordID = 77
join #Budget b4
on r4.RecordID = b4.RecordID
)t4 on t3.RecordID = t4.RecordID and t3.rn < t4.rn join (select r5.RecordID, f5.FDate fdate5, b5.BDate bdate5
, ROW_NUMBER() over(partition by r5.recordid order by (select NULL)) rn from #Records r5
join #Forecast f5
on r5.RecordID = f5.RecordID
and r5.RecordID = 77
join #Budget b5
on r5.RecordID = b5.RecordID
)t5 on t4.RecordID = t5.RecordID and t4.rn < t5.rn join (select r6.RecordID, f6.FDate fdate6, b6.BDate bdate6
, ROW_NUMBER() over(partition by r6.recordid order by (select NULL)) rn from #Records r6
join #Forecast f6
on r6.RecordID = f6.RecordID
and r6.RecordID = 77
join #Budget b6
on r6.RecordID = b6.RecordID
)t6 on t5.RecordID = t6.RecordID and t5.rn < t6.rn join (select r7.RecordID, f7.FDate fdate7, b7.BDate bdate7
, ROW_NUMBER() over(partition by r7.recordid order by (select NULL)) rn from #Records r7
join #Forecast f7
on r7.RecordID = f7.RecordID
and r7.RecordID = 77
join #Budget b7
on r7.RecordID = b7.RecordID
)t7 on t6.RecordID = t7.RecordID and t6.rn < t7.rn join (select r8.RecordID, f8.FDate fdate8, b8.BDate bdate8
, ROW_NUMBER() over(partition by r8.recordid order by (select NULL)) rn from #Records r8
join #Forecast f8
on r8.RecordID = f8.RecordID
and r8.RecordID = 77
join #Budget b8
on r8.RecordID = b8.RecordID
)t8 on t7.RecordID = t8.RecordID and t7.rn < t8.rn join (select r9.RecordID, f9.FDate fdate9, b9.BDate bdate9
, ROW_NUMBER() over(partition by r9.recordid order by (select NULL)) rn from #Records r9
join #Forecast f9
on r9.RecordID = f9.RecordID
and r9.RecordID = 77
join #Budget b9
on r9.RecordID = b9.RecordID
)t9 on t8.RecordID = t9.RecordID and t8.rn < t9.rn
首先从 IP 表和上述查询中验证 OP。如果答案被接受，您将使用以下 SP。
SP 1
exec rownumber
create procedure rownumber as --1st sp.
declare @r int = 1, @select nvarchar(100) = 'select * from ', @count int, @finalquery nvarchar(max) = N'', @out nvarchar(max)
begin
set @count = (select Count(*) from (select r.RecordID, f1.FDate fdate1, b1.BDate bdate1
, ROW_NUMBER() over(partition by r.recordid order by (select NULL)) rn from #Records r
join #Forecast f1
on r.RecordID = f1.RecordID
and r.RecordID = 77
join #Budget b1
on r.RecordID = b1.RecordID
)t
)
if @count > 2
begin
while @count >= @r
begin
exec multyJoin @r, @out output
set @finalquery += @out
set @r = @r + 1
end
end
else begin
select * from (
select r.RecordID, f1.FDate fdate1, b1.BDate bdate1
, ROW_NUMBER() over(partition by r.recordid order by (select NULL)) rn from #Records r
join #Forecast f1
on r.RecordID = f1.RecordID
and r.RecordID = 77
join #Budget b1
on r.RecordID = b1.RecordID
)t
end
set @finalquery = stuff(@finalquery,1, 6,@select)
set @finalquery = REPLACE(@finalquery, 'on t0.RecordID = t1.RecordID and t0.rn < t1.rn ','')
print @finalquery
exec sp_executesql @finalquery
end
SP 2
create procedure multyJoin (@r int, @join varchar(max) output) as --2nd sp. which is called form below sp
declare @cond nvarchar(100), @rvc varchar(3) = convert(int, @r) ----row number of current in varchar
, @rvp varchar(3) = convert(int, @r-1) ----row number of previous in varchar
begin
set @join = ' join (select r'+@rvc+'.RecordID, f'+@rvc+'.FDate fdate'+@rvc+', b'+@rvc+'.BDate bdate'+@rvc+ ---Here add your columns as like as 'Percentage'+@rvc+' * '+@rvc+'BAmount'
', ROW_NUMBER() over(partition by r'+@rvc+'.recordid order by (select NULL)) rn from #Records r'+@rvc+'
join #Forecast f'+@rvc+'
on r'+@rvc+'.RecordID = f'+@rvc+'.RecordID
and r'+@rvc+'.RecordID = 77
join #Budget b'+@rvc+'
on r'+@rvc+'.RecordID = b'+@rvc+'.RecordID
)t'+@rvc+
' on t'+@rvp+'.RecordID = t'+@rvc+'.RecordID and t'+@rvp+'.rn < t'+@rvc+'.rn' --Here multiple join has been created. Each table contains one row
end
【讨论】：
感谢脚本。使用数据透视函数的动态查询真的很棒。 你找到解决方案了吗..？ 尚无解决方案。我希望有一个动态查询的示例，如您的回答中所述。以上是关于T-SQL - 连接多个表会导致重复行的主要内容，如果未能解决你的问题，请参考以下文章