Optimizing SQL Server stored procedures that use functions? - sql-server-2008

Question

I'd like some help in optimizing the following query:
SELECT DISTINCT TOP (#NumberOfResultsRequested) dbo.FilterRecentSearchesTitles(OriginalSearchTerm) AS SearchTerms
FROM UserSearches
WHERE WebsiteID = #WebsiteID
AND LEN(OriginalSearchTerm) > 20
--AND dbo.FilterRecentSearchesTitles(OriginalSearchTerm) NOT IN (SELECT KeywordUrl FROM PopularSearchesBaseline WHERE WebsiteID = #WebsiteID)
GROUP BY OriginalSearchTerm, GeoID
It runs fine without the line that is commented out. I have an index set on UserSearches.OriginalSearchTerm, WebsiteID, and PopularSearchesBaseline.KeywordUrl, but the query still runs slow with this line in there.
-- UPDATE --
The function used is as follows:
ALTER FUNCTION [dbo].[FilterRecentSearchesTitles]
(
#SearchTerm VARCHAR(512)
)
RETURNS VARCHAR(512)
AS
BEGIN
DECLARE #Ret VARCHAR(512)
SET #Ret = dbo.RegexReplace('[0-9]', '', REPLACE(#SearchTerm, '__s', ''), 1, 1)
SET #Ret = dbo.RegexReplace('\.', '', #Ret, 1, 1)
SET #Ret = dbo.RegexReplace('\s{2,}', ' ', #Ret, 1, 1)
SET #Ret = dbo.RegexReplace('\sv\s', ' ', #Ret, 1, 1)
RETURN(#Ret)
END
Using the Reglar Expression Workbench code.
However, as I mentioned - without the line that is currently commented out it runs fine.
Any other suggestions?

Answer 1

I am going to guess that dbo.FilterRecentSearchesTitles(OriginalSearchTerm) is a function. My suggestion would be to see about rewriting it into a table valued function so you can return a table that could be joined on.
Otherwise you are calling that function for each row you are trying to return which is going to cause your problems.
If you cannot rewrite the function, then why not create a stored proc that will only execute it once, similar to this:
SELECT DISTINCT TOP (#NumberOfResultsRequested) dbo.FilterRecentSearchesTitles(OriginalSearchTerm) AS SearchTerms
INTO #temp
WHERE WebsiteID = #WebsiteID
SELECT *
FROM #temp
WHERE SearchTerms NOT IN (SELECT KeywordUrl
FROM PopularSearchesBaseline
WHERE WebsiteID = #WebsiteID)
Then you get your records into a temp table after executing the function once and then you select on the temp table.

Answer 2

I might try to use a persisted computed column in this case:
ALTER TABLE UserSearches ADD FilteredOriginalSearchTerm AS dbo.FilterRecentSearchesTitles(OriginalSearchTerm) PERSISTED
You will probably have to add WITH SCHEMABINDING to your function (and the RegexReplace function) like so:
ALTER FUNCTION [dbo].[FilterRecentSearchesTitles]
(
#SearchTerm VARCHAR(512)
)
RETURNS VARCHAR(512)
WITH SCHEMABINDING -- You will need this so the function is considered deterministic
AS
BEGIN
DECLARE #Ret VARCHAR(512)
SET #Ret = dbo.RegexReplace('[0-9]', '', REPLACE(#SearchTerm, '__s', ''), 1, 1)
SET #Ret = dbo.RegexReplace('\.', '', #Ret, 1, 1)
SET #Ret = dbo.RegexReplace('\s{2,}', ' ', #Ret, 1, 1)
SET #Ret = dbo.RegexReplace('\sv\s', ' ', #Ret, 1, 1)
RETURN(#Ret)
END
This makes your query look like this:
SELECT DISTINCT TOP (#NumberOfResultsRequested) FilteredOriginalSearchTerm AS SearchTerms
FROM UserSearches
WHERE WebsiteID = #WebsiteID
AND LEN(OriginalSearchTerm) > 20
AND FilteredOriginalSearchTerm NOT IN (SELECT KeywordUrl FROM PopularSearchesBaseline WHERE WebsiteID = #WebsiteID)
GROUP BY OriginalSearchTerm, GeoID
Which could potentially be optimized for speed (if necessary) with a join instead of not in, or maybe different indexing (perhaps on the computed column, or some covering indexes). Also, DISTINCT with a GROUP BY is somewhat of a code smell to me, but it could be legit.

Answer 3

Instead of using using the function on SELECT, I modified the INSERT query to include this function. That way, I avoid calling the function for every row when I later want to retrieve the data.