Question

I am looping through all my databases and aggregating the results into an aggregates database.

In my loop I call

master.dbo.xp_cmdshell osql C:\whatever.SQL 

As the loop progresses, the cmdshell takes longer and longer to execute. If I stop the loop and run a single aggregate for one database it executes quickly.

Is there anything I can add to my external SQL script to make it run faster? Maybe something to commit and free the records before the next loop? Or should I add some kind of a pause after every loop?

I want to use an external SQL file because it contains many update statements and it’s more manageable for me.

Here’s how I loop:

Update dbFoo.dbo.tblBar set Processed = 0
Go

WHILE EXISTS ( SELECT ID FROM dbFoo.dbo.tblBar WHERE Processed = 0)
    BEGIN

        SELECT @aRow = MIN(tblBar.ID) FROM dbFoo.dbo.tblBar
        SELECT @aFoo1 = Foo1 FROM dbFoo.dbo.tblBar WHERE ID = @aRow
        SELECT @aFoo2 = Foo2 FROM dbFoo.dbo.tblBar WHERE ID = @aRow
        SELECT @aFoo3 = Foo3 FROM dbFoo.dbo.tblWhatever WHERE Foo = @aFoo

        EXEC RunPreAgg @Foo1 = @aFoo1, @Foo2 = @aFoo2, @Foo3 = @aFoo3, @RetVal = @aRetVal OUTPUT
        SELECT returning = @aRetVal

        UPDATE dbFoo.dbo.tblBar SET Processed = 1 WHERE ID = @aRow
    END

Then the RunPreAgg stored procedure basically does this:

if db_id('db' + @Foo1 + '_' + @Foo2) is not null  
    BEGIN
        --This bat file creates the SQL File
        select @sql = 'master.dbo.xp_cmdshell '''+@path+'wwwRunPreAgg.bat ' + @Foo1 + ' ' + @Foo2 + ' ' + @Foo3 + ''''
        exec( @sql )

        --execute
        select @sql = 'master.dbo.xp_cmdshell ''osql -E -o '+@path+'output\tmp'+@Foo1+'_'+@Foo2+'.txt -i '+@path+'tmp' + @Foo1 + '.SQL'''
        exec( @sql )

        --This erases the SQL File
        select @sql = 'master.dbo.xp_cmdshell '''+@path+'wwwCleanup.bat ' + @Foo1 + ' ' + @Foo2 + ''''
        exec( @sql )

        Set @retval = 'Done!'
    END
ELSE
    BEGIN
        Set @retval = 'Err: No DataBase'
    END 

The variable names are changed to protect the innocent. The code works fine, I just need to optimize.

Answer 1

If it is the loops performance that is causing you trouble, you might try reducing the number of selects. Normally I dislike Cursors, but your loop might benefit from one. You can select all the values you need for the loop into memory, then loop through those values without having to run 3 or 4 selects per loop (of course if the performance hit is occurring inside the RunPreAgg SP, then this won’t help):

DECLARE cFoos CURSOR FOR 
    SELECT tblBar.ID, tblBar.Foo1, tblBar.Foo2, tblWhatever.Foo3
    FROM dbFoo.dbo.tblBar
        INNER JOIN dbFoo.dbo.tblWhatever
            ON tblWhatever.Foo = tblBar.Foo
    WHERE tblBar.Processed = 0;

OPEN cFoos;
FETCH NEXT FROM cFoos INTO @aRow, @aFoo1, @aFoo2, @aFoo3;

WHILE @@FETCH_STATUS = 0
BEGIN
    EXEC RunPreAgg @Foo1 = @aFoo1, @Foo2 = @aFoo2, @Foo3 = @aFoo3, @RetVal = @aRetVal OUTPUT
    SELECT returning = @aRetVal

    UPDATE dbFoo.dbo.tblBar SET Processed = 1 WHERE ID = @aRow

   FETCH NEXT FROM cFoos INTO @aRow, @Foo1, @Foo2, @Foo3;
END

CLOSE cFoos;
DEALLOCATE cFoos;