CODEWITHSUNDEEP
sqlsql-servert-sqlsql-server-2008if-statement
Jed
Jed

Reputation: 10887

T-SQL IF condition is not working as expected

I have a stored procedure that will INSERT a record into a table. I have created a unique key constraint on the table to avoid duplicate records.

Regardless of the fact that there is a unique key constraint defined, I am using an IF() condition (prior to the INSERT statement) to check for a duplicate record that may already exist.

However, the conditional statement that I am using to check for a duplicate record is seemingly having no affect on whether or not the INSERT is executed. - i.e. when a duplicate record in submitted to the sproc, a "Violation of UNIQUE KEY constraint..." exception is thrown.

Here is a sample of my sproc:

BEGIN       
    if
        (SELECT Count(f1)
        FROM table
        WHERE f1 = @f1
        AND f2= @f2) 
    <= 0
    BEGIN
        INSERT INTO table
        (f1,f2)
        VALUES
        (@f1, @f2)

        RETURN @@IDENTITY
    END
END

Is there something wrong with my syntax? Or, maybe, I'm going about this the wrong way?

Upvotes: 1

Views: 687

Answers (5)

OMG Ponies
OMG Ponies

Reputation: 332581

This is more concurrency safe:

INSERT INTO table
SELECT @f1, @f2, @f3
  FROM TABLE WITH (UPDLOCK, HOLDLOCK)
 WHERE NOT EXISTS(SELECT NULL
                    FROM table
                   WHERE f1 = @f1
                     AND f2 = @f2
                     AND f3 = @f3)

Or, if you want to keep the decision logic (but less concurrency safe), use NOT EXISTS:

IF NOT EXISTS(SELECT NULL
                FROM table
               WHERE f1 = @f1
                 AND f2 = @f2
                 AND f3 = @f3) 
BEGIN

    INSERT INTO table
      (f1,f2, f3)
    VALUES
      (@f1, @f2, @f3)

    RETURN @@SCOPE_IDENTITY

END

Upvotes: 2

FatherStorm
FatherStorm

Reputation: 7183

SET NOCOUNT ON  
SELECT * FROM table  
WHERE f1 = @f1
        AND f2= @f2  

IF @@ROWCOUNT = 0  
BEGIN  
 INSERT INTO table  
        (f1,f2)  
        VALUES  
        (@f1, @f2)  


    RETURN @@IDENTITY  
END

Upvotes: 0

Jed
Jed

Reputation: 10887

Thanks for your input OMG/Martin, but it turns out that the root of my problem was due to a parameter value that was null.

Specifically:

--sproc params
@f1 int,
@f2 nvarchar(30),
@f3 datetime = null --<<a null value will screw up the IF() condition

IF(SELECT COUNT(f1)
FROM table
WHERE f1 = @f1
AND f2 = @f2
AND f3 = @f3)
<1

BEGIN
  INSERT INTO....
END

Notice that the default value of the @f3 param is null. So, in the case where the caller of the sproc doesn't pass an @f3 parameter, the IF() condition will return 0 (zero) even if there is a matching @f1, @f2, and null @f3.

For example: Say the table already has a record of

f1   f2   f3
--------------
45   foo  NULL

Now a caller fires off the sproc by sending it:

@f1=45

@f2=foo

(Note the caller does not specify @f3)

When the IF() condition if fired, it will return 0 (zero). Strange, eh? Intuitively, I would think that the condition would return 1 since the params are an exact match as the existing values in the table.

To make matters a bit more confusing (for me, anyway), is that even though the IF() succeeds (returns zero), I receive a "Violation of UNIQUE KEY constraint.." exception. Actually, I'm not surprised that the exception is thrown since I consider the param values to be a perfect match to the existing record in the table - the confusing part, to me, is why does the IF() condition contradict the violation exception.

By the way, OMGPonies, I tried using your suggestion (EXISTS), and the same symptoms occur. Apparently, the NULL factor makes things funky.

Upvotes: 1

LukeH
LukeH

Reputation: 269358

If you're using SQL Server 2008 or newer then you should use the MERGE statement:

MERGE INTO table AS t
USING (VALUES (@f1, @f2)) AS s (f1, f2)
    ON s.f1 = t.f1
        AND s.f2 = t.f2
WHEN NOT MATCHED BY t THEN
    INSERT (f1, f2) VALUES (@f1, @f2)

Upvotes: 3

GSerg
GSerg

Reputation: 78155

Count(field) counts values that are not null.
This may be different to count(*), which counts rows.

But I would discourage using count here.
Have a look at this question.

Upvotes: 3

Related Questions