Question

Say I have the following schema:

-- Create the dbo.Transaction table
CREATE TABLE [dbo].[Transaction] (
    [TransactionId] INT NOT NULL IDENTITY,
    [AccountId] INT NOT NULL,
    [TransactionDate] DateTime2(7) NOT NULL,
    [Amount] decimal(9,3) NOT NULL
    CONSTRAINT [PK_Transaction] PRIMARY KEY ([TransactionId])
);

And the following query:

Select
    AccountId,
    TransactionDate,
    Amount,
    AverageAmount       = Avg(Amount)   Over (Partition By AccountId Order By TransactionDate ROWS BETWEEN 2 PRECEDING AND CURRENT ROW),
    TransactionCount    = Count(Amount) Over (Partition By AccountId Order By TransactionDate ROWS 2 PRECEDING),
    MinimumAmount       = Min(Amount)   Over (Partition By AccountId Order By TransactionDate ROWS 2 PRECEDING),
    MaximumAmount       = Max(Amount)   Over (Partition By AccountId Order By TransactionDate ROWS 2 PRECEDING),
    SumAmount           = Sum(Amount)   Over (Partition By AccountId Order By TransactionDate ROWS 2 PRECEDING)
From dbo.[Transaction]
Order By AccountId, TransactionDate

How would I perform this query if it were contained in a UDF or stored proc, and the sliding interval (2 in this example) wasn’t known until runtime, as passed as a parameter to the UDF / stored proc? It seems SQL 2012 doesn’t permit the use of a variable here.

Answer 1

As you mentioned, SQL Server only supports integer literals for PRECEDING and FOLLOWING in OVER clauses.

There are two options available to you: dynamic sql and re-writing the query to not use PRECEDING

Dynamic sql is the simpliest, but I’d be careful about putting it in a UDF.

set @sql = N'Select AccountId, ... ROWS ' 
          + cast(@sz as varchar(10)) + N' PRECEDING) ...'
exec sp_executesql @sql

However window functions are just fancy syntax. You can re-write the query without them:

DECLARE @sz INT
SET @sz = 2

;
WITH    q AS ( SELECT   AccountId ,
                        TransactionDate ,
                        Amount ,
                        ROW_NUMBER() OVER ( PARTITION BY AccountId 
                                            ORDER BY TransactionDate ) rw
               FROM     [Transaction]
             )
    SELECT  accountID ,
            TransactionDate ,
            Amount ,
            ( SELECT    AVG(q1.Amount) FROM  q q1
              WHERE     q1.accountid = q.accountid
                        AND q1.rw BETWEEN q.rw - @sz AND q.rw
            ) AverageAmount,
            ( SELECT    COUNT(q1.Amount) FROM  q q1
              WHERE     q1.accountid = q.accountid
                        AND q1.rw BETWEEN q.rw - @sz AND q.rw
            ) TransactionAmount
            -- etc.
            FROM q
            ORDER BY AccountID, TransactionDate

Here is another way to re-write the query as well:

DECLARE @sz INT
SET @sz = 2;
WITH    q AS ( SELECT   AccountId ,
                        TransactionDate ,
                        Amount ,
                        ROW_NUMBER() OVER ( PARTITION BY AccountId 
                                            ORDER BY TransactionDate ) rw
               FROM     [Transaction]
             )
    SELECT  q.accountID ,
            q.TransactionDate ,
            q.Amount ,
            AVG(q1.Amount) AverageAmount ,
            COUNT(q1.Amount) TransactionAmount ,
            MAX(q1.Amount) MaxAmount ,
            MIN(q1.Amount) MinAmount
                -- etc.
    FROM    q
            INNER JOIN q q1 ON q1.accountid = q.accountid
                               AND q1.rw BETWEEN q.rw - @sz AND q.rw
    GROUP BY q.accountid ,
            q.transactiondate ,
            q.amount