SQL Server query, get records with distinct value in 1 field and max in other field

Question

I have a table with some columns:

WON   param      |  machine   |    inputdate     |  val
---|-------------|------------|------------------|--------|
18 | PAR.SPTDM.X1|    MM01    | 20/01/2013 12:43 |  2.5   |
18 | PAR.SPTDM.Y1|    MM01    | 20/01/2013 12:43 |  3,4   |
22 | PAR.SPTDM.X1|    MM01    | 22/01/2013 16:10 |  1.2   |
22 | PAR.SPTDM.Y1|    MM01    | 22/01/2013 16:10 |  1.7   |
33 | PAR.SPTDM.X1|    MM03    | 22/01/2013 16:13 |  2.34  |
33 | PAR.SPTDM.Y1|    MM03    | 22/01/2013 16:13 |  2,21  |
27 | PAR.LAS.PWR |    MM10    | 25/01/2013 08:14 |  100.5 |
14 | PAR.LAS.UV  |    MM10    | 18/01/2013 17:27 |  134.8 |
41 | PAR.LAS.UV  |    SLA4    | 27/01/2013 09:14 |  2,1   |
62 | PAR.LAS.UV  |    SLA5    | 27/01/2013 11:15 |  14.6  |

Some of these rows get duplicated (for reasons unknown to me, it's the backend for a management system). Also the value column is a string, it needs conversion to numbers.

The database is readonly, and I want to pull in data into an Excel sheet based on a few conditions:

• group by machine and param
• filter on param: only 'PAR.SPTDM.%'
• filter on machine: only 'MM%'
• ignore val's with a comma, convert the rest to numbers

This works without any problems using the following query:

SELECT DISTINCT t1.param, t1.machine, cast(t1.val as float), t1.inputdate
FROM dbbackend t1
WHERE (t1.machine Like 'MM%')
    AND (t1.param Like 'PAR.SPTDM.%')
    AND (t1.val<>'') And (t1.val not Like '%,%')

Now I want to only retrieve the records for each machine and each parameter (of interest) with the latest inputdate. First I tried

SELECT DISTINCT t1.param, t1.machine, max(cast(t1.val as float)), max(t1.inputdate)
FROM dbbackend t1
WHERE (t1.machine Like 'MM%')
    AND (t1.param Like 'PAR.SPTDM.%')
    AND (t1.val<>'') And (t1.val not Like '%,%')
GROUP BY machine, param

But this gave me val's which weren't in the dataset, correct machine/param/inputdate, but wrong val.

Next try was

SELECT DISTINCT t1.param, t1.machine, cast(t1.val as float), t1.inputdate
FROM dbbackend t1
WHERE t1.WON IN
    (
    SELECT latestrec.WON FROM (
        SELECT DISTINCT max(t2.WON), t2.param, t2.machine, max(t2.inputdate)
        FROM dbbackend t2
        WHERE (t2.machine Like 'MM%')
            AND (t2.param Like 'PAR.SPTDM.%')
            AND (t2.val<>'') And (t2.val not Like '%,%')
            GROUP BY machine, param
        ) as latestrec
    )

But this wouldn't work, MS Query told me

Could not add the table '('.

So can someone tell me why I get wrong values with the first call, and why the second isn't working, or what is the correct method? I also suspect MS query to trip over that nested subquery :/

Answer 1

You could try this CTE with the ROW_NUMBER function:

WITH x AS (SELECT t1.param, 
                t1.machine, 
                Cast(t1.val AS FLOAT) AS val, 
                t1.inputdate, 
                RN=Row_number() 
                     OVER( 
                       partition BY t1.machine, t1.param 
                       ORDER BY t1.inputdate DESC) 
         FROM   dbbackend t1 
         WHERE  ( t1.machine LIKE 'MM%' ) 
                AND ( t1.param LIKE 'PAR.SPTDM.%' ) 
                AND ( t1.val <> '' ) 
                AND ( t1.val NOT LIKE '%,%' )) 
SELECT param, 
       machine, 
       val, 
       inputdate 
FROM   x 
WHERE  rn = 1