Question

For the question, Let’s say I have a table that holds the following data :

1) Name
2) Mood
3) DateTime

I could insert records like :

Andy    Happy    '11.06.2012 - 14.06.07' -- Inserted on 11.06.2012 @ 19:12.32
Arthur  Angry    '11.06.2012 - 15.06.57' -- Inserted on 11.06.2012 @ 17:12.32
Andy    Sad      '11.06.2012 - 14.34.05' -- Inserted on 11.06.2012 @ 17:12.32
Arthur  Happy    '11.06.2012 - 13.34.05' -- Inserted on 11.06.2012 @ 14:12.32

I would like to get the “duration” information related to these moods changes !

My table holds thousands of hundred or records and I cannot afford having a process that takes too much time. What would be the best way of calculating this ?

• Trigger “After insert”, filling a “Duration” column ?
• Stored procedure that fill a previously created “Duration” column ?
• Calculated column ?
• A view (I already tried that and it takes more than 2 seconds to display, which is totally unacceptable)
• Another idea ?

Thanks for your help !

Important edit :
The mood records arrive grouped into packet and we cannot be sure that already inserted records have smaller dates! (see the above comments next to my records)

Answer 1

A possible SQL version:

WITH CTE AS(
    SELECT ROW_NUMBER()OVER(PARTITION BY [Name] ORDER BY [Time])As RowNum
    , *
    FROM @table T
)
SELECT 
    DiffSec=DATEDIFF(s,[Time],(SELECT [Time] FROM CTE c2 WHERE c2.[Name]=CTE.[Name] AND c2.RowNum=CTE.RowNum+1))
    , [Name]
    , Mood
    , [Time]
FROM CTE 
ORDER BY [Name],[RowNum]

Result:

DiffSec Name    Mood    Time
1678    Andy    Happy   2012-06-11 14:06:07.000
NULL    Andy    Sad     2012-06-11 14:34:05.000
5228    Arthur  Angry   2012-06-11 14:06:57.000
NULL    Arthur  Happy   2012-06-11 15:34:05.000

Your test data:

declare @table table(name varchar(10),mood varchar(10),time datetime);
insert into @table values('Andy','Happy',convert(datetime,'11.06.2012 14:06:07',104));
insert into @table values('Arthur','Angry',convert(datetime,'11.06.2012 14:06:57',104));
insert into @table values('Andy','Sad',convert(datetime,'11.06.2012 14:34:05',104));
insert into @table values('Arthur','Happy',convert(datetime,'11.06.2012 15:34:05',104));

---

Edit Self-Joining a CTE seems to be a very bad idea (“If you self join the CTE it will kill you“). I’ve tested my query with 500000 records in a temporary table and cancelled the query after 30 minutes.

Here’s a much faster approach(4 seconds for all) using a sub-query (with your commented schema):

SELECT T.*
    ,(SELECT DATEDIFF(s,MAX(T2.Time),T.Time)
      FROM dbo.Temp T2
      WHERE T2.HE_Id = T.HE_Id
          AND T2.Time < T.Time
    ) AS DiffSec
FROM dbo.Temp AS T
ORDER BY HE_Id ,DiffSec