Locking in Stored Procedure

dugas · 03-24-2007, 11:01 AM

Hello All,
Basically I am trying to implement assigning of unique consecutive integers from a particular row in a table in MSSQL 2000.

Example pseudocode:

<code>
SET ISOLATION LEVEL REPEATABLE READ

SELECT @VAR1 = NextAvailableNumberColumn FROM TheTable WHERE pk=@pk

UPDATE TheTable SET NextAvailableNumberColumn = @VAR1 + 1 WHERE pk=@pk

RETURN @VAR1

COMMIT
</code>

Will this ensure that each client will always get a unique number, or is it possible that this could return the same number to more than one client? Also, if this works, could someone please explain why. It seems to me that if the tranaction was ran concurrently with two instances, then the first select would issue a shared lock for both transactions, and when the update was issued there would be deadlock because the row would be share locked by the two transactions. Please advise.

Crito · 03-24-2007, 03:32 PM

The usual way of doing this is by creating a scalar function that returns max(col)+1 and assigning that as the default value. Something like:

Code:

CREATE FUNCTION [dbo].[fn1](
)
RETURNS int
AS
BEGIN
	return ((SELECT Max(NextAvailNum) FROM TheTable) + 1)
END

And the table def would look something like:

Code:

CREATE TABLE [dbo].[TheTable](
	[NextAvailNum] [int] NOT NULL CONSTRAINT [DF_Num] DEFAULT ([dbo].[fn1]()),
	[SomeAttribute] [nchar](10) NULL
) ON [PRIMARY]

dugas · 03-25-2007, 11:40 AM

Thanks for the reply Crito. If I understood correctly, the method you showed is for generating the next consecutive number on an insert. I am sorry if I did not make myself clear in my post. The WHERE clause in the update statement shows that I must keep an "autonumber" for each row in the table. Basically, each row in the table defines a particular range within which consecutive numbers must be generated in. For example: Row 1 = Company A, PO Number Range 1-1000, NextAvailablePONum = 10, Row 2 = Company B, PO Number Range 1001-2000, NextAvailablePONum = 1001. Then when the client generates a PO number for a particular company, this stored procedure gets called, and gets the next available po number from the row, then increments the column value for that row by one. The reason it was not implemented with identity columns is that there are a couple of hundred companies.