Generating a unique key in SQL Server on a per company basis - sql-server

Question

I store positions in a SQL Server 2012 database, where each position is defined by a position number and a company number.
The position numbers are unique for each company only.
For instance, my database could have the following
POSITION_NO COMPANY_NO
1 1
2 1
3 1
1 2
2 2
3 2
1 3
I need a function which takes a company number as a parameter, and returns the next sequential position number, which in the example table above would be 2 for COMPANY_NO = 3
What I use at the moment is:
CREATE PROCEDURE [DB].[GenerateKey]
#p_company_no float(53),
#return_value_argument float(53) OUTPUT
AS
BEGIN
DECLARE
#v_position_no numeric(5, 0)
SELECT #v_position_no = max(POSITION_NO) + 1
FROM DB.POSITION_TABLE with (nolock)
WHERE COMPANY_NO = #p_company_no
SET #return_value_argument = #v_position_no
RETURN
END
I am aware of the potential pitfalls of using with (nolock), but this was added in an unsuccessful attempt to prevent data-locks on my database. In fact, besides the fact that well-written code is obviously preferable, the main reason I am asking this question is to try and cut down the amount of places that could be causing the data-lock.
Is there any way my code could be improved?

Answer 1

Create an auxilliary table with sequences, with one row for every company (as you already did):
create table seq (company int, sequence int);
go
Seed the counters, one for every company (say there are two companies, 1 and 2):
insert seq values
(1, 1), (2, 1);
go
Then all you need is a way to both update and select the new value in a single statement to avoid race conditions. This is how to do it:
declare #next int;
declare #company int;
set #company = 2;
update seq
set #next = sequence = sequence + 1
where company = #company;
select #next
It would be nice to enclose this into a scalar function, but unfortunatelly no updates in functions are allowed. But you already have a stored procedure in place, so just modify the code in it.
And please tell me that the datatypes used are not really floats? Why not ints?

Answer 2

WHILE(1=1)
BEGIN
SELECT #v_position_no = max(POSITION_NO)
FROM DB.POSITION_TABLE with (nolock)
WHERE COMPANY_NO = #p_company_no
INSERT INTO DB.POSITION_TABLE
(COMPANY_NO, POSITION_NO)
SELECT TOP 1 #p_company_no, #v_position_no + 1
FROM DB.POSITION_TABLE with (nolock)
WHERE NOT EXISTS (SELECT 1
FROM DB.POSITION_TABLE with (nolock)
WHERE COMPANY_NO = #p_company_no
AND POSITION_NO = #v_position_no + 1)
IF(##ROWCOUNT > 0)
BREAK;
END
SET #return_value_argument = #v_position_no + 1
Note that this would only insert in the second statement if the POSITION_NO + 1 wasn't since added. If it was then it would try again.