如何在SQL Server 2014中的數據發生更改時提取列名稱？

Question

我有一張桌子，插入到每個夜晚都是數據的快照。在任何時候，數據可能在列中發生變化（AccountNo保持不變，RunKey增加1，RunDate增加1天;所有其他列可以改變特別）。下面是數據的外觀的例子：

|AccountNo | RunKey | RunDate |  Address  | Salary | PromotionDate| 
---------------------------------------------------------------------------- 
| 12345 | 2 | 06/20/2017 | 123 Main Street | 60,000 | 01/15/2017 | 
| 12345 | 3 | 06/21/2017 | 123 Main Street | 60,000 | 01/15/2017 | 
| 12345 | 4 | 06/22/2017 | 123 Main Street | 65,000 | 06/21/2017 | 

使用LAG函數和CASE表達式我能確定何時有變化（1意味着它改變標誌）：

|AccountNo | RunKey | RunDate |  Address  | AddressLAG |AddressFlag| Salary | SalaryLAG |SalaryFlag| PromotionDate|PromotionDateLag|PromotionFlag| 
----------------------------------------------------------------------------------------------------------------------------------------------------------------- 
| 12345 | 2 | 06/20/2017 | 123 Main Street | 123 Main Street |  0  | 60,000 | 60,000 | 0  | 01/15/2017 | 01/15/2017 |  0  | 
| 12345 | 3 | 06/21/2017 | 123 Main Street | 123 Main Street |  0  | 60,000 | 60,000 | 0  | 01/15/2017 | 01/15/2017 |  0  | 
| 12345 | 4 | 06/22/2017 | 123 Main Street | 123 Main Street |  0  | 65,000 | 60,000 | 1  | 01/15/2017 | 06/21/2017 |  1  | 

我只需要更改的記錄插入到一​​個新表，新表如下所示：

| RunKey | AccountNo | ChangedCol | PrevRunDate | RunDate | PrevValue | NewValue | 
------------------------------------------------------------------------------------------- 
| 4 | 12345 | Salary  | 06/21/2017 | 06/22/2017 | 60,000 | 65,000  | 
| 4 | 12345 | PromotionDate | 06/21/2017 | 06/22/2017 | 01/15/2017 | 06/21/2017 | 

將有每列變化的新紀錄。因此，如果多列更改，每個更改都會記錄在一個新行中。這是我需要幫助的地方，我不知道如何動態地只將已更改的列插入到新表中。

Answer 1

所以，通常這是通過觸發器完成的。任何時候在表上完成插入或更新操作時，都會對審計表執行後續插入操作。所以，我真的會研究這一點。但是如果你不想走這條路線，或者這是一個你不能添加觸發器的第三方系統，你可以用幾種方式插入更改。一個快速的方法是使用除了。基本上，它將來自源表的記錄插入審計表中，當它們不完全匹配時。這是一個例子。

declare @source table (
         AccountNo int 
         ,Address varchar(256) 
         ,Salary decimal(16,4) 
         ,PromotionDate datetime) 
insert into @source 
values 
(12345,'123 Main Street',60000,'20170115') 


declare @audit table (
         AccountNo int 
         ,Address varchar(256) 
         ,Salary decimal(16,4) 
         ,PromotionDate datetime 
         ,RunDate datetime) 

--load the audit table with the current version of the source table 
insert into @audit 
select *, getdate() from @source 

--show that the tables match currently 
select * from @source 
select * from @audit 

--insert into @audit if there are any changes (notice we haven't made any updates yet) 
insert into @audit 
select AccountNo, Address, Salary, PromotionDate, getdate() from @source 
except 
select AccountNo, Address, Salary, PromotionDate, getdate() from @audit 

--show that a record WAS NOT inserted since there was no change. There is only 1 record, the orignal version 
select * from @audit 

--update the promotion and salary 
update @source 
set 
    PromotionDate = '20170331' 
    ,Salary = '65000' 

--insert into @audit if there are any changes 
insert into @audit 
select AccountNo, Address, Salary, PromotionDate, getdate() from @source 
except 
select AccountNo, Address, Salary, PromotionDate, getdate() from @audit 

--show that a record was inserted since there was a change 
select * from @audit 

然後，所有你需要做的就是從@audit表和order by RunDate選擇，你可以很容易地看到什麼是迅速改變，對旋轉的數據並具有用於每一個變化，每帳戶1行。在這個例子中，儘管薪水和促銷數據發生變化，但您只能看到1行。您可以使用您的LEAD和LAG函數，或者在RunDate < RunDate的前1處自行加入以標記列是否更改，但實際上沒有必要。

Answer 2

下可能會有幫助，但你必須定義字段中Cross Apply

例跟蹤或dbFiddle

Declare @YourTable Table ([AccountNo] int,[RunKey] int,[RunDate] date,[Address] varchar(50),[Salary] int,[PromotionDate] date) 
Insert Into @YourTable Values 
(12345,2,'06/20/2017','123 Main Street',60000,'01/15/2017') 
,(12345,3,'06/21/2017','123 Main Street',60000,'01/15/2017') 
,(12345,4,'06/22/2017','123 Main Street',65000,'06/21/2017') 


;with cte as (
    Select A.AccountNo 
      ,A.RunKey 
      ,A.RunDate 
      ,B.* 
     ,PreValue=Lag(Value) over (Partition By AccountNo,Item Order by RunDate) 
     ,PreDate =Lag(RunDate) over (Partition By AccountNo,Item Order by RunDate) 
    From @YourTable A 
    Cross Apply (values ('Address'  ,cast(A.[Address] as varchar(max))) 
         ,('Salary'  ,cast(A.[Salary] as varchar(max))) 
         ,('PromotionDate',cast(A.[PromotionDate] as varchar(max))) 
       ) B (Item,Value) 
) 
Select * 
From cte 
Where Value<>PreValue and PreValue is not null 

返回

如果與可視化幫助

熱膨脹係數生成

Answer 3

我不知道我已經清楚地理解你的要求，但我會找一個使用INSERT/UPDATE觸發器解決這種審計試驗任務的解決方案。也許這會讓事情變得更簡單。