Question

I just released some code into production that is randomly causing errors. I already fixed the problem by totally changing the way I was doing the query. However, it still bothers me that I don’t know what was causing the problem in the first place so was wondering if someone might know the answer. I have the following query inside of a stored procedure. I’m not looking for comments about that’s not a good practice to make queries with nested function calls and things like that :-). Just really want to find out why it doesn’t work consistently. Randomly the function in the query will return a non-numeric value and cause an error on the join. However, if I immediately rerun the query it works fine.

SELECT      cscsf.cloud_server_current_software_firewall_id,
                dbo.fn_GetCustomerFriendlyFromRuleName(cscsf.rule_name, np.policy_name) as rule_name,
                cscsf.rule_action,
                cscsf.rule_direction,
                cscsf.source_address,
                cscsf.source_mask,
                cscsf.destination_address,
                cscsf.destination_mask,
                cscsf.protocol,
                cscsf.port_or_port_range,
                cscsf.created_date_utc,
                cscsf.created_by
    FROM        CLOUD_SERVER_CURRENT_SOFTWARE_FIREWALL cscsf
    LEFT JOIN   CLOUD_SERVER cs
    ON          cscsf.cloud_server_id = cs.cloud_server_id
    LEFT JOIN   CLOUD_ACCOUNT cla
    ON          cs.cloud_account_id = cla.cloud_account_id
    LEFT JOIN   CONFIGURATION co
    ON          cla.configuration_id = co.configuration_id
    LEFT JOIN   DEDICATED_ACCOUNT da
    ON          co.dedicated_account_id = da.dedicated_account_id
    LEFT JOIN   CORE_ACCOUNT ca
    ON          da.core_account_number = ca.core_account_id
    LEFT JOIN   NETWORK_POLICY np 
    ON          np.network_policy_id = (select dbo.fn_GetIDFromRuleName(cscsf.rule_name))
    WHERE       cs.cloud_server_id = @cloud_server_id
    AND         cs.current_software_firewall_confg_guid = cscsf.config_guid
    AND         ca.core_account_id IS NOT NULL
    ORDER BY    cscsf.rule_direction, cscsf.cloud_server_current_software_firewall_id

if you notice the join

ON          np.network_policy_id = (select dbo.fn_GetIDFromRuleName(cscsf.rule_name))

calls a function.

Here is that function:

ALTER FUNCTION [dbo].[fn_GetIDFromRuleName]
(
    @rule_name              varchar(100)
)
RETURNS varchar(12)
AS
BEGIN
    DECLARE     @value      varchar(12)

        SET @value = dbo.fn_SplitGetNthRow(@rule_name, '-', 2)
        SET @value = dbo.fn_SplitGetNthRow(@value, '_', 2)
        SET @value = dbo.fn_SplitGetNthRow(@value, '-', 1)

    RETURN      @value
END

Which then calls this function:

ALTER FUNCTION [dbo].[fn_SplitGetNthRow]
(
    @sInputList     varchar(MAX),
    @sDelimiter     varchar(10) = ',',
    @sRowNumber     int = 1
)
RETURNS varchar(MAX)
AS
BEGIN
    DECLARE     @value      varchar(MAX)

    SELECT      @value = data_split.item
                        FROM
                        (
                            SELECT *, ROW_NUMBER() OVER (ORDER BY (SELECT 1)) as row_num FROM dbo.fn_Split(@sInputList, @sDelimiter)
                        ) AS data_split
                        WHERE
                        data_split.row_num = @sRowNumber

    IF          @value IS NULL
        SET     @value = ''

    RETURN      @value  
END

which finally calls this function:

ALTER FUNCTION [dbo].[fn_Split] (
    @sInputList VARCHAR(MAX),
    @sDelimiter VARCHAR(10) = ','
) RETURNS @List TABLE (item VARCHAR(MAX))
BEGIN
    DECLARE @sItem VARCHAR(MAX)
    WHILE CHARINDEX(@sDelimiter,@sInputList,0) <> 0
        BEGIN
            SELECT @sItem=RTRIM(LTRIM(SUBSTRING(@sInputList,1,CHARINDEX(@sDelimiter,@sInputList,0)-1))), @sInputList=RTRIM(LTRIM(SUBSTRING(@sInputList,CHARINDEX(@sDelimiter,@sInputList,0)+LEN(@sDelimiter),LEN(@sInputList))))
            IF LEN(@sItem) > 0
                INSERT INTO @List SELECT @sItem
        END

    IF LEN(@sInputList) > 0
        INSERT INTO @List SELECT @sInputList -- Put the last item in
    RETURN
END

Answer 1

The reason it is “randomly” returning different things has to do with how SQL Server optimizes queries, and where they get short-circuited.

One way to fix the problem is the change the return value of fn_GetIDFromRuleName:

return (case when isnumeric(@value) then @value end)

Or, change the join condition:

on np.network_policy_id = (select case when isnumeric(dbo.fn_GetIDFromRuleName(cscsf.rule_name)) = 1)
                                       then dbo.fn_GetIDFromRuleName(cscsf.rule_name) end)

The underlying problem is order of evaluation. The reason the “case” statement fixes the problem is because it checks for a numeric value before it converts and SQL Server guarantees the order of evaluation in a case statement. As a note, you could still have problems with converting numbers like “6e07” or “1.23” which are numeric, but not integers.

Why does it work sometimes? Well, clearly the query execution plan is changing, either statically or dynamically. The failing case is probably on a row that is excluded by the WHERE condition. Why does it try to do the conversion? The question is where the conversion happens.

WHere the conversion happens depends on the query plan. This may, in turn, depend on when the table cscf in question is read. If it is already in member, then it might be read and attempted to be converted as a first step in the query. Then you would get the error. In another scenario, the another table might be filtererd, and the rows removed before they are converted.

In any case, my advice is:

• NEVER have implicit conversion in queries.
• Use the case statement for explicit conversions.
• Do not rely on WHERE clauses to filter data to make conversions work. Use the case statement.