August 9, 2009 at 10:37 pm
Comments posted to this topic are about the item Twenty tips to write a good stored procedure
August 9, 2009 at 11:55 pm
Sorry Arup, this article deserves a 5. I rated it one by mistake. My apologies.
August 10, 2009 at 1:43 am
Don't you mean to advise "set nocount on"?
August 10, 2009 at 2:30 am
Sorry, but I'm going to greatly disagree with a number of these 'tips'
8. sp_executeSQL and the KEEPFIXED PLAN options - Both sp_executesql and the KEEPFIXED PLAN option avoid the recompilation of a stored procedure.
Well, maybe. There are still things that will force a recompile. Recompiles aren't always bad, sometimes they are, sometimes running the proc with an outdated and inefficient plan is far, far worse.
11. More WHERE clause hints - Avoid unnecessary conditions in the WHERE Clause. You can easily use the first example instead of the second one. The second example uses one extra OR condition which can be avoided using the first example.
SELECT emp_name FROM table_name WHERE LOWER(emp_name) = 'edu'
SELECT emp_name FROM table_name WHERE emp_name = 'EDU' OR emp_name = 'edu'
Completely the wrong way around. The first, with a function on the column, cannot use an index seek. The best it can run with is a full index scan which won't be cheap on a large table. The second can use an index seek and, in fact, will do two seeks against the table. Unless the table is very small, the second will run way faster than the first.
General rule. Never use a function on a column in the where clause unless there's absolutely no way around. It prevents SQL doing index seeks, forcing full index scans or even table scans.
Also, try to avoid IN. While checking the existence of some values, then use EXISTS instead of IN. Because IN counts the NULL values also, hence slower than EXISTS. Since EXISTS returns Boolean(Yes/No) but IN returns all values hence result set for IN is heavier than EXISTS.
SELECT * FROM employee WHERE emp_no NOT IN (SELECT emp_no from emp_detail)
SELECT * FROM employee WHERE NOT EXISTS (SELECT emp_no FROM emp_detail)
Those two queries are not equivalent, hence they won't be running the same speed.
IN checks for matches, so the first query will return all records in employee where there isn't a match in emp_detail. Exists returns true if there are any rows at all in the subquery, it doesn't care about the values returned in the select. Hence that second query will only return results if there are no rows at all in emp_detail and, if that's the case, it will return all the rows in the employee table.
The equivalent query using exists is this:
SELECT * FROM employee WHERE NOT EXISTS (SELECT 1 FROM emp_detail where employee.emp_no = emp_detail.emp_no)
That will probably run much the same speed as the IN (haven't tested)
EDIT: The queries are only equivalent when the subquery returns no NULLs. If it does, NOT IN and NOT EXISTS return different results.
12. CAST and CONVERT - Try to use CAST instead of CONVERT. CAST is ANSI-92 standard but CONVERT works in MS SQL server only. Also, Convert may be deprecated in future MS SQL releases.
Please provide a reference that states that CONVERT may be deprecated in future versions.
14. Avoid using cursors - Try to use temporary table/table variables with identity column and then iterate all the tables using WHILE loop and a looping counter, which will map with the identity column.
A while loop is no better than a cursor. It's still looping and working on rows one by one. If you're going to remove a cursor, replace it with set-based code. Properly written set-based code will be way faster than cursor or while loop for the vast majority of cases.
16.Subquery vs JOINs - But try to avoid correlated sub queries because it makes the query much slower.
Most of the time correlated subqueries run just as fast as the equivalent query with a join and most of the time have the same exec plan. The exceptions are when the correlation function is an inequality and when there's a TOP (1) ... ORDER BY in the subquery.
18. Try to use table variables instead of Temporary Tables - Temp tables can cause stored procedures to recompile. But table variables were designed specifically to guard against stored procedure recompiles during execution.
In SQL 2005 and higher, with temp table caching, use of temp tables does not always cause a recompile. Adding rows to a temp table may still cause a recompile though.
It's true that table variables don't cause recompiles, but the downside is that the optimiser estimates 1 row in the table variable. If there's a lot more, the plan that the optimise comes up with may be very bad indeed and the query run very, very slow.
Recommendation: Try both see how they behave, use the one that causes the least problems for the particular proc.
19.Index scans are much faster than table scans.
And index seeks are faster still.
But when a table returns smaller rows, then it is better to use a table scan.
I assume you mean 'smaller number of rows' rather than 'smaller rows'. Even so, that's not necessarily true. See this blog post
http://sqlblogcasts.com/blogs/simons/archive/2009/08/06/Do-you-need-to-index-very-small-table-.aspx
Gail Shaw
Microsoft Certified Master: SQL Server, MVP, M.Sc (Comp Sci)
SQL In The Wild: Discussions on DB performance with occasional diversions into recoverability
August 10, 2009 at 2:33 am
Very nice guidelines but with regards to performance point 11 doesn't make sense to me.
"11. More WHERE clause hints - Avoid unnecessary conditions in the WHERE Clause. You can easily use the first example instead of the second one. The second example uses one extra OR condition which can be avoided using the first example."
The example shows first query with an OR clause like:
--Where emp_domain = 'EDU' or emp_domain = 'edu'
against
--Where Lower(emp_domain) = 'edu'
Isn't the function in most cases slower than the OR clause? At least that's what I was taught and in most cases I tested this was the case.
[Edit] Looks Like Gail already answered this.
August 10, 2009 at 2:38 am
I disagree with point 7 - regarding the sp_ prefix, try the example below.
CREATE DATABASE Test
GO
USE [master]
GO
CREATE PROCEDURE dbo.sp_TempProc
AS
SELECT 'master'
GO
USE [Test]
GO
CREATE PROCEDURE dbo.sp_TempProc
AS
SELECT 'test'
GO
USE [Test]
GO
EXEC sp_TempProc
GO
USE [master]
GO
EXEC sp_TempProc
GO
Chris
August 10, 2009 at 2:49 am
Chris Howarth (8/10/2009)
I disagree with point 7 - regarding the sp_ prefix, try the example below.
What he said is mostly true. It was true on SQL 2000 (if I recall). On 2005, the check is first done to the MSSQLSystemResource Database (a hidden system DB). If I recall (haven't tried it) it is still possible to get resolution to master with a proc starting with sp_ but the proc in master must be marked as a system object.
Still, if you check with profiler, a call to a proc that starts with sp_ does have SQL check DBs other than the current one first.
Easy way to show that there's some odd name resolution happening.
CREATE PROCEDURE sp_help
AS
SELECT 'My Local version of sp_help'
GO
exec sp_help
Gail Shaw
Microsoft Certified Master: SQL Server, MVP, M.Sc (Comp Sci)
SQL In The Wild: Discussions on DB performance with occasional diversions into recoverability
August 10, 2009 at 3:13 am
August 10, 2009 at 3:40 am
sorry, it is better to put SET NOCOUNT ON, isn't it?
August 10, 2009 at 3:41 am
I completely agree with the fact that stored proc with the prefix "sp_" is reserved for system stored procs. However, the 2nd part is not correct.
If a stored procedure uses same name in both the user database and a system database, the stored procedure in the user database will never get executed.
To verify this I've executed the below code, which suggest irrespective of the prefix, it’s going to execute the proc which has been referred.
T-SQL Code
use master
drop proc dbo.sp_test
go
create proc dbo.sp_test
as
print 'SP from Master DB'
go
use AdventureWorks
go
drop proc dbo.sp_test
go
create proc dbo.sp_test
as
print 'SP from AdvWork DB'
go
exec dbo.sp_test
Result
SP from AdvWork DB
August 10, 2009 at 3:56 am
gfabbri (8/10/2009)
sorry, it is better to put SET NOCOUNT ON, isn't it?
Definitely - SET NOCOUNT ON suppresses the count messages. Kinda wrong way round - the code shown is the opposite what the description meant.
August 10, 2009 at 3:57 am
Hi,
I tried with these two statements but the second one is not get executing and it says "Incorrect syntax near the keyword 'exists'." What wrong with this or is this wrong statement?
SELECT * FROM employee WHERE emp_no NOT IN (SELECT emp_no from emp_detail)
SELECT * FROM employee WHERE NOT EXISTS (SELECT emp_no FROM emp_detail)
August 10, 2009 at 4:17 am
Good set of guidelines Arup...
but regarding set nocount
do you mean set nocount on?
August 10, 2009 at 4:21 am
arup_kc (8/9/2009)
Comments posted to this topic are about the item Twenty tips to write a good stored procedure
I understand that 'where lower(fieldname)' will not use the index on fieldname as the optimizer will look for an index based on the expression on the left of the =
Sam Trenchard MCT
August 10, 2009 at 4:43 am
I think you missed a very important point I try to stress to all my developers (unless I missed it in the list), and thats to review your query plan for the stored proc. It's probably one of the most important things to do when creating a sproc.
Viewing 15 posts - 1 through 15 (of 244 total)
You must be logged in to reply to this topic. Login to reply