Question

Ok. First let me apologize profusely if this question has been covered. I did look but none of the solutions addressed the particulars of my problem.

I have a table of over 160 million rows of data tracking employee/server conditions over time. I want to create a subset of this data and remove the repetition that occurs throughout, BUT keeps the sequence of changes as they occur. The reduction for most employees would be from 700 rows (and growing) to 1.

Here’s a simplified example of what I’m trying to get to:

Given:

RowID  Employee  Server  Timestamp
-----  --------  ------  ---------
5      E000001   Serv-B  May01
4      E000001   Serv-A  Apr01
3      E000001   Serv-B  Mar01
2      E000001   Serv-A  Feb01
1      E000001   Serv-A  Jan01

Doing a "Min(Timestamp) Group By Employee, Server" would yield:
Employee Server  Timestamp
-------- ------  ---------
E000001  Serv-B  Mar01
E000001  Serv-A  Jan01
.
What I need is:
Employee Server  Timestamp
-------- ------  ---------
E000001  Serv-B  May01
E000001  Serv-A  Apr01
E000001  Serv-B  Mar01
E000001  Serv-A  Jan01

The table and the process that feeds it do not belong to our group so I can’t affect a solution there and I’d rather not be stuck with a copy of the entire thing. I can’t realistically do a cursor/RBAR approach given the size of the table. If backed into a corner, I can write an application program to do this but I was wondering if any of the gods from SQoLympus had any wisdom for doing this in a stored procedure. Thanks in advance!

Edit: It’s SQL Server 2008 – Sorry for not mentioning it.

Answer 1

If SQL Server (assuming I have understood your requirements correctly)

/*Set up test table*/
DECLARE @T TABLE (
  RowID       INT,
  Employee    CHAR(7),
  [Server]    CHAR(6),
  [timestamp] DATETIME );

INSERT INTO @T
SELECT 5,'E000001','Serv-B',  '20010501' UNION ALL
SELECT 4,'E000001','Serv-A',  '20010401' UNION ALL
SELECT 3,'E000001','Serv-B',  '20010301' UNION ALL
SELECT 2,'E000001','Serv-A',  '20010201' UNION ALL
SELECT 1,'E000001','Serv-A',  '20010101';

WITH cte
     As (SELECT ROW_NUMBER() OVER (PARTITION BY Employee ORDER BY RowID) -
                ROW_NUMBER() OVER (PARTITION BY Employee, Server
                                       ORDER BY RowID) AS Grp,
                *
         FROM   @T),
     cte2
     AS (SELECT *,
                ROW_NUMBER() OVER (PARTITION BY Employee, Grp ORDER BY RowID) AS
                Rn
         FROM   cte)

/* Edit: Actually - You want a SELECT not a DELETE I think?
DELETE FROM cte2 WHERE  Rn > 1*/

SELECT   RowID, Employee, [Server], [timestamp]
FROM cte2
WHERE  Rn = 1