CODEWITHSUNDEEP
sqlsql-serversql-server-2008t-sqldatetime
Danilo Souza
Danilo Souza

Reputation: 23

Get all dates with a week interval given a start date and end date

I have an event record database, I need to find all events for each costumer and group them in one week interval from the last record of each costumer until the first.

FirstEvent                  LastEvent               Client
==
2011-01-01 06:55:21.000 |2011-04-30 21:46:09.000    |Client1
2011-01-01 06:20:00.000 |2011-02-23 12:43:55.000    |Client2
2011-01-03 08:34:33.000 |2011-04-30 09:00:39.000    |Client3
2011-01-01 02:14:45.000 |2011-04-30 15:31:41.000    |Client4
2011-01-01 08:08:12.000 |2011-02-21 09:41:28.000    |Client5
2011-02-01 11:29:28.000 |2011-04-29 09:13:25.000    |Client6

Now I have to count the records of each client and group them like this

Client     From                    To                        Records
==
Client1   |2011-04-30 21:46:09.000 |2011-04-23 21:46:09.000  |200
Client1   |2011-04-16 21:46:09.000 |2011-04-09 21:46:09.000  |400
                                 ...Until the the first date
Client1   |2011-01-08 06:55:21.000 |2011-01-01 06:55:21.000  |250
Client2   |2011-02-23 12:43:55.000 |2011-02-16 12:43:55.000  |50
                                 ...The same for each user

I hope someone can help me find a way to output this.

Upvotes: 2

Views: 2502

Answers (3)

Dibstar
Dibstar

Reputation: 2364

This should work if the number of weeks between the start and end is less than 2048, if not then I would suggest using a numbers table.

SELECT
client
,DATEADD(WK,(number * -1) -1,fromdate) as [from]
,DATEADD(WK,number * -1,fromdate) as [to]
,COUNT(client) as records
from #sample
LEFT OUTER JOIN (SELECT number FROM dbo.spt_values WHERE type = 'p') nums 
on nums.number <= DATEDIFF(WK,todate,fromdate)
GROUP BY client
,DATEADD(WK,(number * -1) -1,fromdate) 
,DATEADD(WK,number * -1,fromdate) 
ORDER BY 
client
,DATEADD(WK,(number * -1) -1,fromdate) DESC

Upvotes: 0

ANisus
ANisus

Reputation: 77935

Okay, I think this does exactly what you wish for:

SELECT CONVERT(datetime, '2011-01-01 06:55:21.000', 120) AS event, 'client1' AS Client
INTO #EVENTS
UNION ALL SELECT '2011-01-08 06:55:20.000', 'client1'
UNION ALL SELECT '2011-04-30 21:46:09.000', 'client1'


;WITH E AS (
    SELECT
        Client,
        MIN(event) AS FirstEvent
    FROM #EVENTS
    GROUP BY Client
)
,B AS (
    SELECT
        E.Client,
        E.FirstEvent,
        COUNT(*) AS Records,
        DATEDIFF(DAY, 0, c.event - E.FirstEvent) / 7 AS [weeks]     
    FROM #EVENTS c
    JOIN E ON E.Client = c.Client
    GROUP BY E.Client, E.FirstEvent, DATEDIFF(DAY, 0, c.event - E.FirstEvent) / 7
)
SELECT
    Client,
    DATEADD(WEEK, weeks, FirstEvent) AS [From],
    DATEADD(WEEK, weeks + 1, FirstEvent) AS [To],
    Records
FROM B
ORDER BY Client, weeks;

DROP TABLE #EVENTS

Though, if there are no records a certain week, it will leave out that row instead of returning a row containing the Records value of 0:

Edit:

If you wonder why I use DAY instead of WEEK for my DATEDIFF, it is to avoid problems whether Sunday or Monday is the first day of the week.

Upvotes: 1

Anthony Faull
Anthony Faull

Reputation: 17957

You can group by week using the DATEDIFF function.

SELECT e.[Client]
, [From] = MIN(e.EventDate)
, [To] = MAX(e.EventDate)
, [Records] = COUNT(*)
FROM Events AS e
GROUP BY e.[Client]
, DATEDIFF(week, '1900-01-01', e.EventDate)

Upvotes: 2

Related Questions