Question

我正在使用SQL Server 2005。

每次更改某个字段时，我都有一个包含行存档的表。我必须生成一个报告，显示每个员工更改的字段。

我的表架构：

tblEmp(empid, name, salary, createddate)

我的表格数据：

Row 1: 1, peter, 1000, 11/4/2012
Row 2: 1, peter, 2000, 11/5/2012
Row 3: 1, pete, 2000, 11/6/2012
Row 4: 1, peter, 4000, 11/7/2012

根据员工Peter（员工ID 1）的上述数据，输出（更改）将为：

结果集：

1, oldsalary: 1000 newsalary: 2000 (changed on 11/5/2012)
1, oldname: peter newname: pete (changed on 11/6/2012)
1, oldname: pete newname: peter, oldsalary:2000, newsalary: 4000 (changed on 11/7/2012)

我正在努力想出能够产生上述结果集的sql。

我尝试过与此主题中第一个答案类似的内容：How to get difference between two rows for a column field?

然而，它并没有聚集在一起，所以想知道是否有人可以提供帮助。

Answer 1

您正在逐列查看差异。这表明使用unpivot。以下内容创建输出，其中包含列中的每个更改以及先前的值和日期：

DECLARE @t TABLE(empid INT,name SYSNAME,salary INT,createddate DATE);

INSERT @t SELECT 1, 'peter', 1000, '20121104'
UNION ALL SELECT 1, 'peter', 2000, '20121105'
UNION ALL SELECT 1, 'pete',  2000, '20121106'
UNION ALL SELECT 1, 'peter', 4000, '20121107';


with cv as (
      select empid, createddate, col, val
      from (select empid, CAST(name as varchar(8000)) as name,
                   CAST(salary as varchar(8000)) as salary, createddate
            from @t
           ) t
      unpivot (val for col in (name, salary)) as unpvt
     ),
    cvr as (
     select cv.*,
            ROW_NUMBER() over (partition by empid, col order by createddate) as seqnum_all
     from (select cv.*, ROW_NUMBER() over (partition by empid, col, thegroup order by createddate) as seqnum_group
           from (select cv.*,
                        (ROW_NUMBER() over (partition by empid, col order by createddate) -
                         ROW_NUMBER() over (partition by empid, col, val order by createddate)
                        ) as thegroup
                 from cv
                ) cv
          ) cv
     where seqnum_group = 1
    ) -- select * from cvr
select cvr.*, cvrprev.val as preval, cvrprev.createddate as prevdate
from cvr left outer join
     cvr cvrprev
     on cvr.empid = cvrprev.empid and
        cvr.col = cvrprev.col and
        cvr.seqnum_all = cvrprev.seqnum_all + 1

Answer 2

也许这些与CTE's + ROW_NUMBER加入了CASE：

WITH cte  AS 
(
  SELECT empid,
         name,
         salary, 
         rn=ROW_NUMBER()OVER(PARTITION BY empid ORDER BY createddate)
  FROM   tblemp
) 
SELECT oldname=CASE WHEN c1.Name=c2.Name THEN '' ELSE C1.Name END,
       newname=CASE WHEN c1.Name=c2.Name THEN '' ELSE C2.Name END,
       oldsalary=CASE WHEN c1.salary=c2.salary THEN NULL ELSE C1.salary END,
       newsalary=CASE WHEN c1.salary=c2.salary THEN NULL ELSE C2.salary END
FROM cte c1 INNER JOIN cte c2 
ON c1.empid=c2.empid AND c2.RN=c1.RN + 1

Sql-Fiddle Demo

Answer 3

DECLARE @t TABLE(empid INT,name SYSNAME,salary INT,createddate DATE);

INSERT @t SELECT 1, 'peter', 1000, '20121104'
UNION ALL SELECT 1, 'peter', 2000, '20121105'
UNION ALL SELECT 1, 'pete',  2000, '20121106'
UNION ALL SELECT 1, 'peter', 4000, '20121107';

;WITH x AS
(
  SELECT empid, name, salary, createddate, rn = ROW_NUMBER() OVER 
  (PARTITION BY empid ORDER BY createddate)
  FROM @t
  -- WHERE empid = 1 -- for example
)
SELECT LTRIM(
  CASE WHEN x.salary <> y.salary THEN 
    'oldsalary: ' + RTRIM(x.salary)
    + ' newsalary: ' + RTRIM(y.salary)
  ELSE '' END
  + CASE WHEN x.name <> y.name THEN 
    ' oldname: ' + x.name
    + ' newname: ' + y.name
  ELSE '' END
  + ' (changed on ' + CONVERT(CHAR(10), y.createddate, 101) + ')')
FROM x INNER JOIN x AS y
ON x.rn = y.rn - 1
AND x.empid = y.empid
AND
(
 x.salary <> y.salary 
 OR x.name <> y.name
);

除非你有一个where子句来定位特定的empid，否则输出不是很有用，除非它还包括empid。 SQLfiddle demo

Answer 4

根据你解释的内容，当这个表被更改时创建一个触发器然后用你期望的结果创建表会更容易，因为你在那一刻有旧值和新值，应该有提出你期望的结果不是问题。

显示SQL SERVER中同一表中行之间的差异

4 个答案: