Search code examples
sql-serveraggregate-functionssql-server-2008-r2analytic-functions

replace NULL values with latest non-NULL value in resultset series (SQL Server 2008 R2)


for SQL Server 2008 R2

I have a resultset that looks like this (note [price] is numeric, NULL below represents a NULL value, the result set is ordered by product_id and timestamp)

product timestamp          price 
------- ----------------   -----
   5678 2008-01-01 12:00   12.34
   5678 2008-01-01 12:01    NULL
   5678 2008-01-01 12:02    NULL
   5678 2008-01-01 12:03   23.45
   5678 2008-01-01 12:04    NULL

I want to transform that to a result set that (essentially) copies a non-null value from the latest preceding row, to produce a resultset that looks like this:

product timestamp          price  
------- ----------------   -----
   5678 2008-01-01 12:00   12.34
   5678 2008-01-01 12:01   12.34
   5678 2008-01-01 12:02   12.34
   5678 2008-01-01 12:03   23.45
   5678 2008-01-01 12:04   23.45

I don't find any aggregate/windowing function that will allow me to do this (again this ONLY needed for SQL Server 2008 R2.)

I was hoping to find an analytic aggregate function that do this for me, something like...

LAST_VALUE(price) OVER (PARTITION BY product_id ORDER BY timestamp)

But I don't seem to find any way to do a "cumulative latest non-null value" in the window (to bound the window to the preceding rows, rather than the entire partition)

Aside from creating a table-valued user defined function, is there any builtin that would accomplish this?


UPDATE:

Apparently, this functionality is available in the 'Denali' CTP, but not in SQL Server 2008 R2.

LAST_VALUE http://msdn.microsoft.com/en-us/library/hh231517%28v=SQL.110%29.aspx

I just expected it to be available in SQL Server 2008. It's available in Oracle (since 10gR2 at least), and I can do something similar in MySQL 5.1, using a local variable.

http://download.oracle.com/docs/cd/E14072_01/server.112/e10592/functions083.htm


Solution

  • You can try the following:

    * Updated **

    -- Test Data
    DECLARE @YourTable TABLE(Product INT, Timestamp DATETIME, Price NUMERIC(16,4))
    
    INSERT INTO @YourTable
    SELECT 5678, '20080101 12:00:00', 12.34
    UNION ALL
    SELECT 5678, '20080101 12:01:00', NULL
    UNION ALL
    SELECT 5678, '20080101 12:02:00', NULL
    UNION ALL
    SELECT 5678, '20080101 12:03:00', 23.45
    UNION ALL
    SELECT 5678, '20080101 12:04:00', NULL
    
    ;WITH CTE AS
    (
        SELECT *
        FROM @YourTable
    )
    
    -- Query
    SELECT A.Product, A.Timestamp, ISNULL(A.Price,B.Price) Price
    FROM CTE A
    OUTER APPLY (   SELECT TOP 1 *
                    FROM CTE 
                    WHERE Product = A.Product AND Timestamp < A.Timestamp
                    AND Price IS NOT NULL
                    ORDER BY Product, Timestamp DESC) B
    
    --Results
    Product Timestamp   Price
    5678    2008-01-01 12:00:00.000 12.3400
    5678    2008-01-01 12:01:00.000 12.3400
    5678    2008-01-01 12:02:00.000 12.3400
    5678    2008-01-01 12:03:00.000 23.4500
    5678    2008-01-01 12:04:00.000 23.4500