6

I have the following table which contains values read every 15 minutes from several different devices:

ID   DeviceID   Date                    Value
----------------------------------------------
1    3          24.08.2011 00:00:00     0.51
2    3          24.08.2011 00:15:00     2.9
3    3          24.08.2011 00:30:00     0
4    3          24.08.2011 00:45:00     7.1
5    3          24.08.2011 01:00:00     1.05
6    3          24.08.2011 03:15:00     3.8

I'd like to find all the gaps in the table for each device, where there are no entries, for a given month. For the table above, the result should be something like this:

DeviceID    StartDate               EndDate
-------------------------------------------------------
3           24.08.2011 01:00:00     24.08.2011 03:15:00

The table has roughly 35000 devices and 100 million entries.

This is what I tried; it's rather slow, but returns what I need. However, besides its speed, there's another problem: it only finds missing intervals up to the last entry for a device in the given month; anything after that will be ignored, so it's possible to miss an extra interval of missing values.

SELECT
    t2.Date AS StartDate
    , t1.Date AS EndDate
FROM
    TestTable t1
    INNER JOIN TestTable t2 ON t1.DeviceID = t2.DeviceID
WHERE
    (t2.Date = (SELECT MAX(Date) FROM TestTable t3 WHERE t3.Date < t1.Date AND t3.DeviceID = t1.DeviceID)
        AND DATEDIFF(MINUTE, t2.Date, t1.Date) > 15)
    AND t1.DeviceID = @id
    AND DATEPART(YEAR, t1.Date) = @year AND DATEPART(MONTH, t1.Date) = @month
Conrad Frix
  • 51,984
  • 12
  • 96
  • 155
alex
  • 3,710
  • 32
  • 44
  • you might want to look [at this](http://www.manning.com/nielsen/SampleChapter5.pdf) its a sample chapter that covers the subject of gaps and islands from Itzik Ben-Gan – Conrad Frix Jun 06 '12 at 06:29
  • @NikolaMarkovinović thank you, but I deleted the question myself. It was a poor question, and I managed to figure it out myself (I used `ROW_NUMBER()` as well). – alex Aug 24 '12 at 11:50

1 Answers1

7

Following should work and doesn't return just a single record for a deviceid.

The gist of this is to

  • Add a rownumber to each record, ordered by Date and restarting for each DeviceID.
  • Join with self to create a result with rows consisting of the combination of two original rows. The relation between the columns of each row is the rownumber (+1) and the DeviceID.
  • Only retain those rows where the related Date is more than 15 minutes.

SQL Statement

;WITH t AS (
  SELECT  *, rn = ROW_NUMBER() OVER (PARTITION BY DeviceID ORDER BY Date)
  FROM    TestTable
)  
SELECT  t1.DeviceID, t1.Date, t2.Date
FROM    t t1
        INNER JOIN t t2 ON t2.DeviceID = t1.DeviceID AND t2.rn = t1.rn + 1
WHERE   DATEDIFF(MINUTE, t1.Date, t2.Date) > 15        

Test script

;WITH TestTable (ID, DeviceID, Date, Value) AS (
  SELECT 1, 3, '2011-08-24 00:00:00', 0.51 UNION ALL
  SELECT 2, 3, '2011-08-24 00:15:00', 2.9 UNION ALL
  SELECT 3, 3, '2011-08-24 00:30:00', 0 UNION ALL
  SELECT 4, 3, '2011-08-24 00:45:00', 7.1 UNION ALL
  SELECT 5, 3, '2011-08-24 01:00:00', 1.05 UNION ALL
  SELECT 6, 3, '2011-08-24 03:15:00', 3.8 
)
, t AS (
  SELECT  *, rn = ROW_NUMBER() OVER (PARTITION BY DeviceID ORDER BY Date)
  FROM    TestTable
)  
SELECT  t1.DeviceID, t1.Date, t2.Date
FROM    t t1
        INNER JOIN t t2 ON t2.DeviceID = t1.DeviceID AND t2.rn = t1.rn + 1
WHERE   DATEDIFF(MINUTE, t1.Date, t2.Date) > 15        
Lieven Keersmaekers
  • 57,207
  • 13
  • 112
  • 146