Question

I have a SQL CLR UDF which queries a web service. Since this can be costly, especially within the function is part of a query on multiple rows, I would like to avoid calling the web service whenever possible. In every case, the same input will yield the same output (e.g., if my input is ‘abc’ I will always get ‘xyz’ and nothing different, likewise ‘def’ will always yield ‘tuv’, etc).

I have done some tests, and it seems that SQL does not do any sort of caching on its end, so the web service will always be invoked.

Example Case: I have a table MyTable with a field MyField1. While MyTable has 500 rows, MyField1 will always only have one of 3 possible values. Example query:

SELECT MyFunction(MyField1) FROM MyTable

What happens is that the web service will be called 500 times, once for each row in the table. What I would prefer is that the web service is only called 3 times (once for each distinct value), and read from some cache for the duplicate values.

Example Code:

[SqlFunction]
public static SqlString MyFunction(SqlString input)
{
    if (input.IsNull) return SqlString.Null;

    using (var webService = new MyWebService())
    {
        string result = webService.Call(input.Value);

        return new SqlString(result);
    }
}

What I would really prefer is to keep this in a cache that is specific to the context. That is, the cache would only exist to cache results within the call of a single stored procedure, or within a single query window, etc. Is there any available mechanism to accomplish what I’m after?

Answer 1

EDIT
NOTE The updated version towards the end

Although this answer doesn’t use caching, it should minimise the number of calls to your function. using a number of CTEs to find the distinct values of myField1, then lookup the distinct values in the webservice using your function, then join these back to MyTable. The example below probably makes it clearer:

SQL Fiddle

MS SQL Server 2008 Schema Setup:

CREATE TABLE MyTable
(
    ID int PRIMARY KEY IDENTITY,
    MyField1 VARCHAR(1)
);


CREATE FUNCTION MyFunction
(
    @input As VARCHAR(1)
)
RETURNS VARCHAR(10)
AS
BEGIN
    -- This could be a CLR Function
    -- Return the result of the function
    RETURN  CASE @input WHEN 'A' THEN 'aaaaaaaaaa' WHEN 'B' THEN 'bbbbbbbbbb' ELSE 'ccccccccc' END

END;

-- DATA SET UP
DECLARE @i INT = 0
DECLARE @Field VARCHAR(1)
WHILE @i < 1000
BEGIN
    SELECT @Field = CASE @i % 3 WHEN 1 THEN 'A' WHEN 2 THEN 'B' ELSE 'C' END
    INSERT INTO MyTable (MyField1) VALUES  (@Field)
    SET @i = @i + 1
END

Query 1:

;WITH DistinctMyField1CTE
AS
(
    SELECT DISTINCT MyField1
    FROM MyTable
),
LookupValuesCTE
AS
(
    SELECT MyField1, dbo.MyFunction(MyField1) As MyOutputField
    FROM DistinctMyField1CTE
)
SELECT TOP 20 T1.Id, T1.MyField1, T2.MyOutputField
FROM MyTable T1
INNER JOIN LookupValuesCTE T2
    ON T1.MyField1 = T2.MyField1
ORDER BY T1.ID

Results:

| ID | MYFIELD1 | MYOUTPUTFIELD |
---------------------------------
|  1 |        C |     ccccccccc |
|  2 |        A |    aaaaaaaaaa |
|  3 |        B |    bbbbbbbbbb |
|  4 |        C |     ccccccccc |
|  5 |        A |    aaaaaaaaaa |
|  6 |        B |    bbbbbbbbbb |
|  7 |        C |     ccccccccc |
|  8 |        A |    aaaaaaaaaa |
|  9 |        B |    bbbbbbbbbb |
| 10 |        C |     ccccccccc |
| 11 |        A |    aaaaaaaaaa |
| 12 |        B |    bbbbbbbbbb |
| 13 |        C |     ccccccccc |
| 14 |        A |    aaaaaaaaaa |
| 15 |        B |    bbbbbbbbbb |
| 16 |        C |     ccccccccc |
| 17 |        A |    aaaaaaaaaa |
| 18 |        B |    bbbbbbbbbb |
| 19 |        C |     ccccccccc |
| 20 |        A |    aaaaaaaaaa |

EDIT:
Note when I check a SQL Profiler trace of the aboce, I see a 1000 calls to the UDF, in other words the query analyser is generating a plan which expands out the CTE’s and calls the UDF once for each row.

The following uses a table variable to ensure that the UDF is only called 3 times. I have traced this in SQL Profiler and it is much more efficient. This uses the same table and functions as above. Need SQLFiddle Attached

SQL Fiddle

MS SQL Server 2008 Schema Setup:

CREATE TABLE MyTable
(
    ID int PRIMARY KEY IDENTITY,
    MyField1 VARCHAR(1)
);


CREATE FUNCTION MyFunction
(
    @input As VARCHAR(1)
)
RETURNS VARCHAR(10)
AS
BEGIN
    -- This could be a CLR Function
    -- Return the result of the function
    RETURN  CASE @input WHEN 'A' THEN 'aaaaaaaaaa' WHEN 'B' THEN 'bbbbbbbbbb' ELSE 'ccccccccc' END

END;

-- DATA SET UP
DECLARE @i INT = 0
DECLARE @Field VARCHAR(1)
WHILE @i < 1000
BEGIN
    SELECT @Field = CASE @i % 3 WHEN 1 THEN 'A' WHEN 2 THEN 'B' ELSE 'C' END
    INSERT INTO MyTable (MyField1) VALUES  (@Field)
    SET @i = @i + 1
END

Query 1:

DECLARE @TempTable TABLE
(
    MyField1 VARCHAR(1) PRIMARY KEY,
    MyOutputField VARCHAR(10) NULL
)

INSERT INTO @TempTable (MyField1)
SELECT DISTINCT MyField1
FROM MyTable


-- UPDATE Separately otherwise the function gets called
-- for every row in MyTable
UPDATE @TempTable
    SET MyOutputField = dbo.MyFunction(MyField1)

SELECT TOP 20 T1.ID, T1.MyField1, T2.MyOutputField
FROM MyTable T1
INNER JOIN @TempTable T2
    ON T1.MyField1 = T2.MyField1

Results:

| ID | MYFIELD1 | MYOUTPUTFIELD |
---------------------------------
|  2 |        A |    aaaaaaaaaa |
|  5 |        A |    aaaaaaaaaa |
|  8 |        A |    aaaaaaaaaa |
| 11 |        A |    aaaaaaaaaa |
| 14 |        A |    aaaaaaaaaa |
| 17 |        A |    aaaaaaaaaa |
| 20 |        A |    aaaaaaaaaa |
| 23 |        A |    aaaaaaaaaa |
| 26 |        A |    aaaaaaaaaa |
| 29 |        A |    aaaaaaaaaa |
| 32 |        A |    aaaaaaaaaa |
| 35 |        A |    aaaaaaaaaa |
| 38 |        A |    aaaaaaaaaa |
| 41 |        A |    aaaaaaaaaa |
| 44 |        A |    aaaaaaaaaa |
| 47 |        A |    aaaaaaaaaa |
| 50 |        A |    aaaaaaaaaa |
| 53 |        A |    aaaaaaaaaa |
| 56 |        A |    aaaaaaaaaa |
| 59 |        A |    aaaaaaaaaa |