Preformance Tuning Advice

  • I have tuned this proc from 8 min down to 5 seconds...Yipeee !!

    But now I have the bug in me to get it going faster. I have read too many articles today regarding tuning to the point where it all makes no sence. I am relatively new to T-SQL and would appreciate any advice you guys can give as to where I can improve preformance

    Thanks for taking the time to check this out

    I am calling other procs from the one below and have listed them below this main script

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ALTER PROCEDURE dbo.ReturnAverages_Filtered

    (

    @AnswerIDList VARCHAR(400)

    )

    AS

    /*

    CREATE A TEMPORY TABLE TO STORE THE CALCULATED DATA

    ONCE THE STORED PROCEDURE IS COMPLETE WE TRUNCATE

    THE LIVE TABLE AND INSERT THE DATA FROM THE TEMP TABLE

    */

    CREATE TABLE #TTFilteredAverageData

    (

    ID INT CONSTRAINT TTNationalAverageData_PK IDENTITY PRIMARY KEY NONCLUSTERED,

    QuestionID INT NOT NULL,

    AnswerID INT NOT NULL,

    SubAnswerID INT NULL,

    Average FLOAT NOT NULL

    )

    DECLARE @ProvIDINT

    DECLARE @QuestionID INT

    DECLARE @PreviousAnswerID INT

    DECLARE @QuestionTypeID INT

    DECLARE @AnswerID INT -- VARIABLE TO STORE THE ANSWER ID'S

    DECLARE @PageTypeID INT

    DECLARE @PathwayID INT

    DECLARE @AverageInjections INT

    DECLARE @WhereDoctorID VARCHAR(800)

    SET @ProvID = 0

    SET @PathwayID = 1

    set @PreviousAnswerID = 0

    SET @AverageInjections = 0

    CREATE TABLE #integers(int INT NOT NULL)

    EXEC sp_list_to_table @AnswerIDList

    CREATE TABLE #doctorIDTable(DoctorIDs INT NOT NULL)

    INSERT INTO #doctorIDTable (doctorIDs) (SELECT DISTINCT(DoctorID) as DoctorID FROM demographics inner join #integers ON demographics.AnswerID = #integers.int)

    -- SET @WhereDoctorID = 'SELECT DISTINCT(DoctorID) as DoctorID FROM demographics WHERE (AnswerID IN (' + #integers + '))'

    -- SET @DoctorIDList = SELECT distinct(DoctorID) as DoctorID FROM demographics WHERE (AnswerID IN (6,11,17))

    /* GET A RS FOR ALL THE QUESTIONS FOR PATHWAY X */

    DECLARE curr_rsQuestions CURSOR

    LOCAL FAST_FORWARD

    FOR

    SELECT Questions.ID AS questionID, Questions.QuestionTypeID AS questionTypeID, Answers.ID AS answerID, Pages.PageTypeID

    FROM QuestionTypes INNER JOIN

    Questions ON QuestionTypes.ID = Questions.QuestionTypeID INNER JOIN

    QuestionFamilies ON Questions.QuestionFamilyID = QuestionFamilies.ID INNER JOIN

    Pages ON QuestionFamilies.PageID = Pages.ID INNER JOIN

    Pathways ON Pages.PathwayID = Pathways.ID INNER JOIN

    Answers ON Questions.ID = Answers.QuestionID

    WHERE (Pathways.ID = @PathwayID) AND (Pages.PageTypeID IN (5, 6, 7, 12)) AND (Questions.QuestionTypeID <= 4)
    ORDER BY Pages.ID, Questions.QuestionOrder, Answers.AnswerOrder

    OPEN curr_rsQuestions
    FETCH NEXT FROM curr_rsQuestions INTO @QuestionID, @QuestionTypeID, @AnswerID, @PageTypeID
    WHILE (@@FETCH_STATUS = 0)
    BEGIN --START LOOPING THRU QUESTIONS
    /* GET A RS FOR ALL THE ANSWERS FOR THIS QUESTION ID */

    /* CHECK TO SEE WHAT QUESTION TYPE THIS IS AND DEPENDING ON THE QUESTION TYPE
    CALCULATE THE AVERAGES DIFFERENTLY */
    -- DECLARE VARIABLES
    DECLARE @totalU4TQ INT --VARIABLE TO STORE THE TOTAL USERS WHO ANSWERED THIS QUESTION
    DECLARE @totalU4TA INT --VARIABLE TO STORE THE TOTAL USERS WHO ANSWERED THIS ANSEWER
    DECLARE @totalTimesAnswered INT --VARIABLE TO STORE THE TOTAL TIMES THIS ANSWER WAS CHOSEN
    DECLARE @average FLOAT(2) --VARIABLE TO STORE THE AVERAGE TIMES THIS ANSWER WAS ANSWERED
    DECLARE @averageDose FLOAT(2) --VARIABLE TO STORE THE AVERAGE TIMES THIS DOSAGE WAS PRESCRIBED

    SET @average = 0

    IF @QuestionTypeID = 1 --SINGLE CHOICE
    BEGIN
    EXEC @totalU4TQ = sp_U4TQ @QuestionID, @ProvID, @PageTypeID, 1 --RETURNS THE TOTAL USERS WHO ANSWERED THIS QUESTION
    EXEC @totalU4TA = sp_U4TA @AnswerID, @ProvID, @PageTypeID, 1 --RETURNS THE TOTAL TIMES THIS ANSWER WAS CHOSEN
    IF @totalU4TA <> 0 --MAKE SURE THAT WE DON'T DEVIDE BY ZERO

    SET @average = (CAST(@totalU4TA AS FLOAT)/CAST(@totalU4TQ AS FLOAT)) * 100 --CALCULATE THE AVERAGE

    EXEC Insert_FilteredAverages @QuestionID, @AnswerID, NULL, @Average, @ProvID --INSERT THE DATA INTO #TTFilteredAverageData

    END

    ELSE

    IF @QuestionTypeID = 2 --MULTIPLE CHOICE

    BEGIN

    EXEC @totalU4TA = sp_U4TA @AnswerID, @ProvID, @PageTypeID, 1 --RETURNS THE TOTAL USERS WHO ANSWERED THIS ANSWER

    EXEC @totalTimesAnswered = sp_TimesAnswered @QuestionID, @ProvID, 1 --RETURNS THE TOTAL TIMES THIS ANSWER WAS CHOSEN

    if @AnswerID = 361

    IF @totalTimesAnswered <> 0 --MAKE SURE WE DON'T DEVIDE BY ZERO

    SET @average = (CAST(@totalU4TA AS FLOAT)/CAST(@totalTimesAnswered AS FLOAT)) * 100 --CALCULATE THE AVERAGE'

    EXEC Insert_FilteredAverages @QuestionID, @AnswerID, NULL, @Average, @ProvID --INSERT THE DATA INTO #TTFilteredAverageData

    END

    ELSE

    /*

    -- THIS QUESTION IS NOT IN USE FOR THE APPLICATION

    IF @QuestionTypeID = 3 --SUB QUESTION

    BEGIN

    END

    ELSE

    */

    IF @QuestionTypeID = 4 --ROOT

    /*

    --we already have a list of @AnswerID's 304-312 which represent the injection sites (the ROOT question)

    --qType = 5 get a rs of the sum of injection sites (basically count answerID's where = 304-312)

    --qType = 6 get a rs of the answerIDs associated with @QuestionType 6 and questionFamily of @QuestionID

    --

    */

    BEGIN

    --BRANCH

    IF @PreviousAnswerID <> @AnswerID --WE ONLY NEED TO DO THIS ONCE PER INJECTION SITE

    BEGIN

    EXEC @averageDose = ReturnAverageDose @QuestionID, @AnswerID, @ProvID, 1 --RETURNS THE AVERAGE DOSE FOR THE SITE

    EXEC Insert_FilteredAverages 43, @AnswerID, NULL, @averageDose, @ProvID --INSERT THAT AVERAGE IN THE DATABASE

    SET @PreviousAnswerID = @AnswerID --SET THE PREVIOUS ANSWERID TO CURRENT ANSWERID SO WE ONLY DO THIS ONCE PER INJECTION SITE

    END

    --LEAF

    DECLARE @SubAnswerID INT

    -- GET A RS OF ALL THE SUB ANSWERS ASSOCITATED WITH THIS QUESTION FAMILY

    DECLARE curr_rsSubAnswers CURSOR

    LOCAL FAST_FORWARD

    FOR

    SELECT Answers.ID as subAnswerID

    FROM Answers INNER JOIN

    Questions ON Answers.QuestionID = Questions.ID INNER JOIN

    QuestionTypes ON Questions.QuestionTypeID = QuestionTypes.ID INNER JOIN

    QuestionFamilies ON Questions.QuestionFamilyID = QuestionFamilies.ID INNER JOIN

    Pages ON QuestionFamilies.PageID = Pages.ID INNER JOIN

    Pathways ON Pages.PathwayID = Pathways.ID

    WHERE (Questions.QuestionTypeID = 6) AND (Pathways.ID = @PathwayID) AND (Questions.QuestionFamilyID = @QuestionID)

    ORDER BY Answers.AnswerOrder

    --START LOOPING THRU SUBANSWERS

    OPEN curr_rsSubAnswers

    FETCH NEXT FROM curr_rsSubAnswers INTO @SubAnswerID

    WHILE (@@FETCH_STATUS = 0)

    BEGIN

    SET @averageDose = 0

    DECLARE @U4TSA INT --STORES THE USERS 4 THIS SUB ANSWER

    DECLARE @TotalTimesSubAnswered INT --STORES THE TOTAL TIMES THIS SUB ANSWER WAS CHOOSEN

    EXEC @U4TSA= sp_U4TA @AnswerID, @ProvID, @PageTypeID, 1 --RETURNS THE TOTAL TIMES THIS ANSWER WAS CHOSEN

    EXEC @TotalTimesSubAnswered = sp_TotalTimesSubAnswered @AnswerID, @SubAnswerID, @ProvID, 1

    IF @TotalTimesSubAnswered <> 0 --MAKE SURE WE DON'T DEVIDE BY ZERO

    SET @averageDose = (CAST(@TotalTimesSubAnswered AS FLOAT)/CAST(@U4TSA AS FLOAT)) * 100 --CALCULATE THE AVERAGE

    EXEC Insert_FilteredAverages @QuestionID, @AnswerID, @SubAnswerID, @averageDose, @ProvID --INSERT THE DATA INTO #TTFilteredAverageData

    FETCH NEXT FROM curr_rsSubAnswers INTO @SubAnswerID

    END--END LOOPING THRU SUBANSWERS

    --CLEAN UP

    CLOSE curr_rsSubAnswers

    DEALLOCATE curr_rsSubAnswers

    END

    FETCH NEXT FROM curr_rsQuestions INTO @QuestionID, @QuestionTypeID, @AnswerID, @PageTypeID --GET THE NEXT RECORD FROM THE RS

    END --END LOOPING THRU QUESTIONS

    --CLEAN UP

    CLOSE curr_rsQuestions

    DEALLOCATE curr_rsQuestions

    --CLEAN UP AGAIN

    DROP TABLE #integers

    DROP TABLE #doctorIDTable

    SELECT #TTFilteredAverageData.QuestionID, Questions.QuestionTypeID, Questions.TextEN AS QuestionTextEn, Questions.TextFR AS QuestionTextFr,

    #TTFilteredAverageData.AnswerID, Answers.TextEN AS AnswerTextEn, Answers.TextFR AS AnswerTextFr, #TTFilteredAverageData.SubAnswerID,

    #TTFilteredAverageData.Average, Answers_1.TextEN AS SubAnswerTextEN, Answers_1.TextFR AS SubAnswerTextFR

    FROM QuestionFamilies INNER JOIN

    Questions ON QuestionFamilies.ID = Questions.QuestionFamilyID INNER JOIN

    Pages ON QuestionFamilies.PageID = Pages.ID INNER JOIN

    PageTypes ON Pages.PageTypeID = PageTypes.ID RIGHT OUTER JOIN

    #TTFilteredAverageData INNER JOIN

    Answers ON #TTFilteredAverageData.AnswerID = Answers.ID ON Questions.ID = #TTFilteredAverageData.QuestionID LEFT OUTER JOIN

    Answers Answers_1 ON #TTFilteredAverageData.SubAnswerID = Answers_1.ID

    WHERE pages.pagetypeID <> 2

    ORDER BY #TTFilteredAverageData.ID

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    PROCEDURE dbo.sp_U4TA

    (

    @AnswerID INT,

    @ProvID INT,

    @PageTypeID INT,

    @WhereDoctorID INT

    )

    AS

    DECLARE @ReturnVal INT

    DECLARE @TableName VARCHAR(15)

    DECLARE @SQL NVARCHAR(400)

    IF @PageTypeID = 2

    SET @TableName = 'Demographics'

    ELSE

    SET @TableName = 'Assessments'

    IF @WhereDoctorID = 0

    BEGIN

    IF @ProvID = 0 -- for national query

    BEGIN

    SET @SQL = N'SELECT @ReturnVal = COUNT(*) FROM ' + @TableName + N' WHERE (AnswerID = ' + CONVERT(NVARCHAR(4), @AnswerID) + N')'

    EXEC sp_executesql @SQL, N'@ReturnVal varchar(100) output', @ReturnVal output

    END

    ELSE

    BEGIN

    SET @SQL = N'SELECT @ReturnVal = COUNT(*) FROM ' + @TableName + N' WHERE (AnswerID = ' + CONVERT(NVARCHAR(4), @AnswerID) + N') AND (ProvinceID = ' + CONVERT(VARCHAR(2), @ProvID) + N')'

    EXEC sp_executesql @SQL, N'@ReturnVal varchar(100) output', @ReturnVal output

    END

    END

    ELSE

    BEGIN

    SET @SQL = N'SELECT @ReturnVal = COUNT(*) FROM ' + @TableName + ' INNER JOIN #doctorIDTable ON ' + @TableName + '.DoctorID = #doctorIDTable.doctorIDs WHERE (AnswerID = ' + CONVERT(VARCHAR(4), @AnswerID) + ')'

    EXEC sp_executesql @SQL, N'@ReturnVal varchar(100) output', @ReturnVal output

    END

    IF @ReturnVal = NULL SET @ReturnVal = 0

    RETURN @ReturnVal

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

    PROCEDURE dbo.sp_U4TQ

    (

    @QuestionID INT,

    @ProvID INT,

    @PageTypeID INT,

    @WhereDoctorID INT

    )

    AS

    DECLARE @ReturnVal as Int

    DECLARE @TableName as NVARCHAR(15)

    DECLARE @SQL AS NVARCHAR(1000)

    IF @PageTypeID = 2

    SET @TableName = 'Demographics'

    ELSE

    SET @TableName = 'Assessments'

    IF @WhereDoctorID = 0

    BEGIN

    IF @ProvID = 0 -- for national query

    BEGIN

    IF @PageTypeID = 2

    SELECT @ReturnVal = COUNT(*)

    FROM (SELECT DISTINCT DoctorID

    FROM Demographics

    WHERE (QuestionID = @QuestionID)) DERIVEDTBL

    ELSE

    SELECT @ReturnVal = COUNT(*)

    FROM (SELECT DISTINCT DoctorAssessmentNum, DoctorID

    FROM Assessments

    WHERE (QuestionID = @QuestionID)) DERIVEDTBL

    END

    ELSE

    BEGIN

    SET @SQL = N'SELECT COUNT(*) as ReturnVal FROM ' + @TableName + ' WHERE QuestionID = ' + CONVERT(VARCHAR(4), @QuestionID) + ' AND (ProvinceID = ' + CONVERT(VARCHAR(2), @ProvID) + ')'

    EXEC sp_executesql @SQL, N'@ReturnVal varchar(100) output', @ReturnVal output

    END

    END

    ELSE

    BEGIN

    SET @SQL = N'SELECT @ReturnVal = COUNT(*) FROM ' + @TableName + ' INNER JOIN #doctorIDTable ON ' + @TableName + '.DoctorID = #doctorIDTable.doctorIDs WHERE (QuestionID = ' + CONVERT(VARCHAR(4), @QuestionID) + ')'

    --SET @SQL = N'SELECT COUNT(*) as ReturnVal FROM ' + @TableName + ' WHERE QuestionID = ' + CONVERT(VARCHAR(4), @QuestionID) + N' AND (DoctorID = (' + @WhereDoctorID + N'))'

    EXEC sp_executesql @SQL, N'@ReturnVal varchar(100) output', @ReturnVal output

    END

    IF @ReturnVal = NULL SET @ReturnVal = 0

    IF @ReturnVal < 0 SET @returnVal = 0 RETURN @ReturnVal
    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    PROCEDURE dbo.sp_TimesAnswered
    (
    @QuestionID INT,
    @ProvID INT,
    @WhereDoctorID INT
    )
    AS
    DECLARE @ReturnVal as INT

    IF @WhereDoctorID = 0
    BEGIN
    IF @ProvID = 0 -- for national query
    SELECT @ReturnVal = COUNT(*)
    FROM (SELECT DISTINCT DoctorAssessmentNum, DoctorID
    FROM Assessments
    WHERE (QuestionID = @QuestionID)) DERIVEDTBL

    ELSE-- for provincial query
    SELECT @ReturnVal = COUNT(*)
    FROM (SELECT DISTINCT DoctorAssessmentNum, DoctorID
    FROM Assessments
    WHERE (QuestionID = @QuestionID) AND (ProvinceID = @ProvID)) DERIVEDTBL
    END
    ELSE
    SELECT @ReturnVal = COUNT(*)
    FROM (SELECT DISTINCT DoctorAssessmentNum, DoctorID
    FROM Assessments INNER JOIN #doctorIDTable ON Assessments.DoctorID = #doctorIDTable.doctorIDs
    WHERE (QuestionID = @QuestionID)) DERIVEDTBL

    IF @ReturnVal = NULL SET @ReturnVal = 0

    RETURN @ReturnVal

    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
    ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

  • Best Advice: STOP using CURSORS.....the 'relative' performance (with large datasets) compared to SET_BASED PROCESSING is cr*p.

    Your best strategy is to segment what you are doing into discrete sets (1 type per major question) and process each set as a unit.  ie....get all the info required to complete the insert statement in one select.  then write a seperate select statement for the next question.  there is enough difference in what you are doing/gathering to justify creating similar but different select statements.

    example

    declare cursorx for select a,b,c,d....from tableb where xxxx

    while @@fetch_status <> 0

    insert into tablea (values from cursorx)

    read next from cursorx

    loop

     

    in 99.9% of cases will perform much slower than the set-based equivalent

    insert into tablea

    select a,b,c,d from tableb where xxxx

     

     

    other pointers to look out for are:

    make sure all/most columns references in join/where conditions are covered by indices.

    look at the execution plans....minimise 'scans'

    creating the table outside of this proc....will stop the procedure being re-compiled on every run.

    you can use index_hints to force the execution plan to be better than the "system-best-guess"

  • I usually go for table variables rather than temp tables where possible.  You can create a primary key but can't create indexes on table variables though so if you need more than one index, this option is out.

    DECLARE @x TABLE (

       ID INT IDENTITY(1,1),

       Name VARCHAR(20) PRIMARY KEY CLUSTERED)

  • - Wach out for the code used in sub-SPs

    IF @ReturnVal = NULL SET @ReturnVal = 0

    use IF @ReturnVal IS NULL SET @ReturnVal = 0    or     SET @ReturnVal = ISNULL(@ReturnVal, 0)

    Also, use output parameter to return value from the SP. RETURN is supposed to be used for status code only. Even better - use function instead.

    - Create all tem tables at the very beginning of the SP. Mixing DDL and DML WILL cause unwanted SP recompilation. Do the same for variable declarations.

    - Use functions instead of dynamic SQL

    - If you can - get rid of cursors. I successfully replaced many cursors used for loop calculations with update statement using variables. Trivial example would be to enumerate rows by increment of 2. E.g.

    DECLARE @RowNum INT

    SET @RowNum = 2

    UPDATE #TempTable

    SET RowNum = @RowNum,

          @RowNum = @RowNum + 2

    (I learned this trick from a great book: "The Guru's Guide to Transact-SQL" by Ken Henderson)

     

  • Thanks for the advice Andrew it is most appreciated.

    I can see how cursors are a preformance hit but I don't think I have much choice in the matter. I need to calculate aggregate data based on questionID's and answerID's. So as I loop through the initial recordset I call sp_U4TQ and SP_U4TA to calculate an average based on question and answer ID's that I pass as arguements to the above procs.

    I don't see how I can calculate aggregate data using a set_based method?

    I will look into implimenting some of the other pointers you listed

    While reading yesterday I looked at the execution plan for my script and I tryed to figure out where my procedure was slowing down.

    I don't know how to minimise scans?

    What are index_hints?

  • Thanks Ken and Arif

    I don't need to create indexes on the #tempTable so I will impliment the use of a table instead of a temporary table

    Definitely I will update the @returnVal = Null to IS NULL as well as creating the tables at the begining of the script to avoid mixing DDL and DML

    I will try and get rid of the dynamic SQL. I used it to avoid having the same query repeated with only the table name being changed. I have never used functions in SQL how do they work?

    Thanks for the book suggestion I will check it out

  • select columnx, count(*) from tablea

    group by columnx

     

    is how you do aggregates.

    index_hints....spelling/punctuation might be wrong....but look in BOL for info...or search here for the keywords "with index, hint"

    scans....prevent by ensuring columns referenced are 'indexed'....an indexz defined on (cola,colb) in that order is no good to prevent a scan if your query is "select colb from tablea"...whereas it'll be great for a query "select cola from tablea"

  • Sorry what I meant was average data

    So what I am am doing in each itteration of the loop is passing the answerID to SP_U4TA

    SET @SQL = N'SELECT @ReturnVal = COUNT(*) FROM ' + @TableName + N' WHERE (AnswerID = ' + CONVERT(NVARCHAR(4), @AnswerID) + N')'

    And QuestionID to SP_U4TQ

    SELECT @ReturnVal = COUNT(*)

    FROM (SELECT DISTINCT DoctorAssessmentNum, DoctorID

    FROM Assessments

    WHERE (QuestionID = @QuestionID)) DERIVEDTBL

    SET @average = (CAST(@totalU4TA AS FLOAT)/CAST(@totalU4TQ AS FLOAT)) * 100

    So do you think it is possible to combine those two queries into one that returns average data for a unit of Question and Answer ID's ?? So one query with a unit of questions and answer IDs that calculates for every row the averages using the above queries. Does that make sence?

    From your first post that is what I imagined happening not count (*) from

    I will check out BOL for index_hints

    Scans....I get it now. I have run the Index Tuning Wizzard on all of my queries so I think that the scans are minimized as much as possible but I will double check to see that they are

    Thanks again everyone for all your tips thus far

    You guys are a great bunch and I am glad to have joined. I only hope that my knowledge thus far will be able to help other out

    ok gotta get back to tuning this puppy !

  • Averaging is one of the AGGREGATE -type functions.

    look up BOL for details.

    Select cola, count(colb) , avg(colb) from tablea group by cola

    is along the lines of what you want.

    note: count(*) NOT always same as count(colb)...NULL's work differently here....again BOL/searhh here for specific issues.

     

  • Mark

    Here's a little example that might help with your cursor elimination - at least for the ones that you want to loop through data and process each record individually.

    SET NOCOUNT ON

    DECLARE @x TABLE (

       ID INT IDENTITY(1,1),

       Name VARCHAR(20) PRIMARY KEY CLUSTERED)

    INSERT INTO @x SELECT 'Jake 1'

    INSERT INTO @x SELECT 'Jake 2'

    INSERT INTO @x SELECT 'Jake 3'

    INSERT INTO @x SELECT 'Jake 4'

    SELECT Name AS 'Before'

       FROM @x

    DECLARE @max-2 INT, @id INT

    SELECT @id = 1,

          @max-2 = MAX(ID)

       FROM @x

    WHILE @id <= @max-2

    BEGIN

       UPDATE @x

          SET Name = 'Elwood ' + CONVERT(VARCHAR(2),ID)

          WHERE ID = @id

       SELECT @id = @id + 1

    END

    SET NOCOUNT OFF

    SELECT Name AS 'After'

       FROM @x

    As far as eliminating dynamic SQL, try your best but if you're looping through table names and processing each table individually, you may have to resign yourself to using dynamic SQL and moving on.  I'd recommend using sp_executesql if possible.  Here's a nifty trick to pass 'processed' info back from sp_executesql:

    SET NOCOUNT ON

    DECLARE @Tables TABLE (

       RowID INT   IDENTITY(1,1),

       TableName   VARCHAR(50),

       Row_Count   INT NULL)

    DECLARE @max-2   INT,

       @row INT,

       @exec_str   NVARCHAR(2000),

       @tablename  VARCHAR(50),

       @retcode    INT,

       @Row_Count  INT

    INSERT INTO @Tables (TableName)

       SELECT Table_Name

       FROM Information_Schema.Tables

    SELECT @max-2 = MAX(RowID),

          @row = 1

       FROM @Tables

    WHILE @row <= @max-2

    BEGIN

       SELECT @tablename = TableName

          FROM @Tables

          WHERE RowID = @row

       SELECT @exec_str = "SELECT @Row_Count=COUNT(*) FROM " + @TableName

       EXEC @retcode = sp_ExecuteSQL @exec_str, N'@row_count INT OUTPUT', @Row_Count OUTPUT

       UPDATE @Tables

          SET Row_Count = @Row_Count

          WHERE RowID = @row

       SELECT @row = @row + 1

    END

    SELECT * FROM @Tables

    SET NOCOUNT OFF

    Word wrap may screw up this code, but I think you'll get the point.

    Cheers,

    Ken

Viewing 10 posts - 1 through 9 (of 9 total)

You must be logged in to reply to this topic. Login to reply