SQL Prevent multiple inserts at the same time for non-duplicates - sql

Question

I am trying to prevent multiple inserts at the exact same time so I can prevent duplicate inserts. I have two tables:
Table B, this table has 4 columns id, timeToken, tokenOrder and taken.
Table A which I will be inserting into and that has id, createDate and timeToken.
What I am trying to do is prevent the timeToken in Table A not to have duplicate values in the case where multiple inserts are happening at the exact same time. I have the following code:
DECLARE #ReturnValue nvarchar
SELECT Top 1 #ReturnValue=timeToken FROM TableB WHERE taken = 0 Order By tokenOrder
Update TableB SET taken = 1 WHERE timeToken = #ReturnValue
INSERT INTO TableA Values(#ReturnValue, GETDATE())
Now that I think about it, is it possible to have my timeToken table in TableA auto increment with the timeToken from TableB?
Table B sample data:
id timeToken tokenOrder taken
1 1:00am 1 0
2 2:00am 2 0
3 3:00am 3 1
4 4:00am 4 0
5 5:00am 5 0
This is what I am expecting Table A to look like after 4 calls all at the exact same time that would cause duplicates (id starting at 5 - this could be because I have deleted old records).
Table A sample data:
id createDate timeToken
5 2014-11-22 12:45:34.243 1:00am
6 2014-11-22 12:45:34.243 2:00am
7 2014-11-22 12:45:34.243 4:00am
8 2014-11-22 12:45:34.243 5:00am

Answer 1

Try to rewrite like this, this should ensure that you do not get the row with taken=0 in TableB updated twice.
BEGIN TRANSACTION
DECLARE #taken table(
id int NOT NULL,
timeToken nvarchar(max));
Update TOP (1) TableB
SET taken = 1
OUTPUT UPDATED.id, UPDATED.timeToken
INTO #taken
WHERE timeToken =
(SELECT Top 1 timeToken FROM TableB WHERE taken = 0 Order By tokenOrder)
INSERT INTO TableA
SELECT id, GETDATE(), timeToken
FROM #taken
COMMIT TRANSACTION
See SQL Server isolation levels - read commited. READ COMMITTED is the default isolation level for the Microsoft SQL Server Database Engine.
In the example I copy id from TableB to TableA, but it is not probably required.

Answer 2

I think you can solve this problem in two steps:
Step1: Buffer all requests as soon as they arrive.
Step2: Periodically assign free tokens to the buffered requests.
Preparation
A sequence object will help resolve any order ambiguity:
CREATE SEQUENCE dbo.Taken_Seq
START WITH 1
INCREMENT BY 1 ;
GO
An auxiliary table will play the role of the buffer:
CREATE TABLE buffer (
requester uniqueidentifier, createdate datetime, seq_value bigint, id int);
I will also use a GUID to refer to the different processes asking for a token (requesters):
ALTER TABLE TableA add Requester uniqueidentifier;
Solution outline
As soon as a request comes (identified by a GUID) buffer it consuming the next sequence value, like this (here I use newid() to get a GUID, your application should have already assigned one to your request):
declare #seq bigint;
SELECT #seq = NEXT VALUE FOR dbo.Taken_Seq;
insert buffer values (newid(), getdate(), #seq, null);
Suppose now that three such requests arrive simultaneously, as in:
declare #seq bigint;
SELECT #seq = NEXT VALUE FOR dbo.Taken_Seq;
insert buffer values (newid(), getdate(), #seq, null);
SELECT #seq = NEXT VALUE FOR dbo.Taken_Seq;
insert buffer values (newid(), getdate(), #seq, null);
SELECT #seq = NEXT VALUE FOR dbo.Taken_Seq;
insert buffer values (newid(), getdate(), #seq, null);
The contents of the buffer table will then look like this:
requester createdate seq_value id
------------------------------------ ----------------------- -------------------- -----------
109B560C-155C-40BD-A13A-59D21EBEB1F8 2017-04-05 11:17:35.127 31 NULL
FAC00C2E-14AA-4502-AB5C-DDD756914653 2017-04-05 11:17:35.127 32 NULL
E95889C3-E291-4A1C-A7E8-0B8CC53D4D7B 2017-04-05 11:17:35.127 33 NULL
Next we can match each buffered request to a token. This will be done by assinging an id value to each request in our buffered table:
; with a as
(select rn =row_number() over (order by seq_value), *
from buffer
where id is null),
b as
(
select rn=row_number() over (order by tokenOrder), *
from TableB
where taken = 0
)
update buffer set buffer.id = b.id
from buffer
join a on buffer.requester = a.requester
join b on a.rn = b.rn
This is now how our buffer table looks like:
requester createdate seq_value id
------------------------------------ ----------------------- -------------------- -----------
109B560C-155C-40BD-A13A-59D21EBEB1F8 2017-04-05 11:17:35.127 31 1
FAC00C2E-14AA-4502-AB5C-DDD756914653 2017-04-05 11:17:35.127 32 2
E95889C3-E291-4A1C-A7E8-0B8CC53D4D7B 2017-04-05 11:17:35.127 33 3
Join the buffer table with TableB to find the tokens:
select buffer.requester, tableB.* from buffer join tableB on buffer.id= tableB.id
Mark the tokens as taken:
update TableB set taken = 1 from buffer where buffer.id = TableB.id
Finally, insert into TableA:
insert TableA (requester, createdate, timeToken)
select buffer.requester, buffer.createdate, TableB.timeToken
from buffer join TableB on buffer.id = TableB.id
Note:
Obviously some of these steps must be contained within a single transaction