Playing with OPTION KEEPFIXED PLAN hint in SQL Server

My previous article was on Compilation& Recompilation of Stored Procedure. After reading that article, two ofmy reader sent me an email about more clarification on some terminology and thatis the base of writing this article.
Actually SQL Server Query Optimizer is really one of thesmart tools which used to find or generate best execution plan for query or storedprocedure and make the task easy for DBA. Basically it uses Cost BasedOptimizer (CBO) to do this task, even sometime feels to have specific ExecutionPlan for query so Microsoft have provided many different hints at differentlevel.
Query Optimizer (QO) used to save execution plan when itcompiles Stored Procedure first time and use the same execution plan when thesame stored procedure is being used again so that you can get rid of overheadof generating execution plan again and again. 
When index statistics changes heavily, QO used to recompilethe statement to get best plan for your stored procedure but if you are surethat the existing Plan would works fine than one should go for OPTION KEEPFIXEDPLAN for that statement so that you can save your resources and boost upperformance.
I will demonstrate this with few queries run in SQL ServerManagement Studio (SSMS) and capture the results in Profiler. 
I will select following Events in Profiler while creatingnew Trace for my server.
Stored Procedure
  • SP:Completed
  • SP:Recompile
  • SP:smtpCompleted
  • SP:smtpStarting

 

Now execute following TSQL in your SSMS.
–create onedatabase which you can delete after running this example
create database SQLHub
GO
USE SQLHub
GO
–if orderstable is already there. you can delete it than create new one with name”Orders”
IF OBJECT_ID(‘orders’, ‘U’) IS NOT NULL BEGIN
      DROP TABLE orders
END
GO
–creating table
CREATE TABLE orders (OrderID INT IDENTITY, OrderDate DATETIME, Amount MONEY, Refno INT)
GO
–creatingnonclustered index
CREATE NONCLUSTERED INDEXIDX_ORD ON Orders(amount)
GO
–inserting only1 record in our table
INSERT INTO orders VALUES(GETDATE(),1001,111)
GO
–creating SPwhich will SELECT all records from Orders table where Amount is 1001
CREATE PROC spGet_ord
AS
SELECT * FROM orders WHERE Amount=1001
GO
–execute SP
exec spGet_ord
GO
If you will see in profiler than you would get followingevent captured



Now, make change in Index statistics by inserting so manyrecords in table so that we can test the effect in profiler after executing SPagain.
–inserting50000 fack rows into table
INSERT INTO orders (OrderDate, Amount, Refno)
SELECT TOP 50000
      DATEADD(minute, ABS(a.object_id % 50000 ), CAST(‘2010-02-01’ AS DATETIME)),
      ABS(a.object_id % 10),
      CAST(ABS(a.object_id) AS VARCHAR)
FROM sys.all_objects a
CROSS JOIN sys.all_objects b
GO
–execute SPagain
exec spGet_ord
GO

Now, if you see the profiler, you will see “SP:Recompile”event as your statistics are updated so it need recompilation of statement. If youagain few more thousands record in same table and execute the same SP again,you will see recompilation again.
Now, after droping table and SP, we will make one smallchange in Stored Procedure and will use “KEEPFIXED PLAN” hint so that we canavoid recompilation.
drop proc spGet_ord
GO
drop table orders
go
–creating table
CREATE TABLE orders (OrderID INT IDENTITY, OrderDate DATETIME, Amount MONEY, Refno INT)
GO
CREATE NONCLUSTERED INDEXIDX_ORD ON Orders(amount)
GO
INSERT INTO orders VALUES(GETDATE(),1001,111)
GO
CREATE PROC spGet_ord
AS
SELECT * FROM orders WHERE Amount=1001
–adding belowstatement other than that, same code as previous SP
OPTION (KEEPFIXED PLAN);
GO
exec spGet_ord
GO
After recreating table along with fresh records, we aregoing to see what happened while executing SP in above statement.

Since our table now has only 1 record so it is time tochange statistics with bulk insert as follows again.
–inserting100000 fack rows into table (BTW, thank to Jacob Sebastian, copy in INSERTscript from one of his demo to generate big table)
INSERT INTO orders (OrderDate, Amount, Refno)
SELECT TOP 50000
      DATEADD(minute, ABS(a.object_id % 50000 ), CAST(‘2010-02-01’ AS DATETIME)),
      ABS(a.object_id % 10),
      CAST(ABS(a.object_id) AS VARCHAR)
FROM sys.all_objects a
CROSS JOIN sys.all_objects b
GO
exec spGet_ord
GO
–if you wish,you can uncomment below code and delete SQLHub database
–use master
–go
–drop database sqlhub
Here is the screen capture of Profiler which doesn’t showRecompilation.

 

You can even add few more thousands of rows and execute SPagain, you won’t get recompilation. 
Reference: Ritesh Shah
 
http://www.sqlhub.com
Note: Microsoft Books online is a default reference of all articles butexamples and explanations prepared by Ritesh Shah, founder of
http://www.SQLHub.com
Ask me any SQL Server related question at my “ASK Profile



Playing with OPTION KEEPFIXED PLAN hint in SQL Server

My previous article was on Compilation& Recompilation of Stored Procedure. After reading that article, two ofmy reader sent me an email about more clarification on some terminology and thatis the base of writing this article.
Actually SQL Server Query Optimizer is really one of thesmart tools which used to find or generate best execution plan for query or storedprocedure and make the task easy for DBA. Basically it uses Cost BasedOptimizer (CBO) to do this task, even sometime feels to have specific ExecutionPlan for query so Microsoft have provided many different hints at differentlevel.
Query Optimizer (QO) used to save execution plan when itcompiles Stored Procedure first time and use the same execution plan when thesame stored procedure is being used again so that you can get rid of overheadof generating execution plan again and again.
When index statistics changes heavily, QO used to recompilethe statement to get best plan for your stored procedure but if you are surethat the existing Plan would works fine than one should go for OPTION KEEPFIXEDPLAN for that statement so that you can save your resources and boost upperformance.
I will demonstrate this with few queries run in SQL ServerManagement Studio (SSMS) and capture the results in Profiler.
I will select following Events in Profiler while creatingnew Trace for my server.
Stored Procedure
  • SP:Completed
  • SP:Recompile
  • SP:smtpCompleted
  • SP:smtpStarting

 

Now execute following TSQL in your SSMS.
–create onedatabase which you can delete after running this example
create database SQLHub
GO
USE SQLHub
GO
–if orderstable is already there. you can delete it than create new one with name”Orders”
IF OBJECT_ID(‘orders’, ‘U’) IS NOT NULL BEGIN
      DROP TABLE orders
END
GO
–creating table
CREATE TABLE orders(OrderID INT IDENTITY, OrderDate DATETIME, Amount MONEY, Refno INT)
GO
–creatingnonclustered index
CREATE NONCLUSTERED INDEXIDX_ORD ON Orders(amount)
GO
–inserting only1 record in our table
INSERT INTO orders VALUES(GETDATE(),1001,111)
GO
–creating SPwhich will SELECT all records from Orders table where Amount is 1001
CREATE PROC spGet_ord
AS
SELECT * FROM orders WHERE Amount=1001
GO
–execute SP
exec spGet_ord
GO
If you will see in profiler than you would get followingevent captured
Now, make change in Index statistics by inserting so manyrecords in table so that we can test the effect in profiler after executing SPagain.
–inserting50000 fack rows into table
INSERT INTO orders(OrderDate, Amount, Refno)
SELECT TOP 50000
      DATEADD(minute, ABS(a.object_id % 50000 ), CAST(‘2010-02-01’ AS DATETIME)),
      ABS(a.object_id % 10),
      CAST(ABS(a.object_id) AS VARCHAR)
FROM sys.all_objects a
CROSS JOIN sys.all_objects b
GO
–execute SPagain
exec spGet_ord
GO
Now, if you see the profiler, you will see “SP:Recompile”event as your statistics are updated so it need recompilation of statement. If youagain few more thousands record in same table and execute the same SP again,you will see recompilation again.
Now, after droping table and SP, we will make one smallchange in Stored Procedure and will use “KEEPFIXED PLAN” hint so that we canavoid recompilation.
drop proc spGet_ord
GO
drop table orders
go
–creating table
CREATE TABLE orders(OrderID INT IDENTITY, OrderDate DATETIME, Amount MONEY, Refno INT)
GO
CREATE NONCLUSTERED INDEXIDX_ORD ON Orders(amount)
GO
INSERT INTO orders VALUES(GETDATE(),1001,111)
GO
CREATE PROC spGet_ord
AS
SELECT * FROM orders WHERE Amount=1001
–adding belowstatement other than that, same code as previous SP
OPTION (KEEPFIXED PLAN);
GO
exec spGet_ord
GO
After recreating table along with fresh records, we aregoing to see what happened while executing SP in above statement.
Since our table now has only 1 record so it is time tochange statistics with bulk insert as follows again.
–inserting100000 fack rows into table (BTW, thank to Jacob Sebastian, copy in INSERTscript from one of his demo to generate big table)
INSERT INTO orders(OrderDate, Amount, Refno)
SELECT TOP 50000
      DATEADD(minute, ABS(a.object_id % 50000 ), CAST(‘2010-02-01’ AS DATETIME)),
      ABS(a.object_id % 10),
      CAST(ABS(a.object_id) AS VARCHAR)
FROM sys.all_objects a
CROSS JOIN sys.all_objects b
GO
exec spGet_ord
GO
–if you wish,you can uncomment below code and delete SQLHub database
–use master
–go
–drop database sqlhub
Here is the screen capture of Profiler which doesn’t showRecompilation.

 

You can even add few more thousands of rows and execute SPagain, you won’t get recompilation.
Reference: Ritesh Shah
 
http://www.sqlhub.com
Note: Microsoft Books online is a default reference of all articles butexamples and explanations prepared by Ritesh Shah, founder of
http://www.SQLHub.com
Ask me any SQL Server related question at my “ASK Profile

Some facts about Stored Procedure Compilation & Recompilation

I generally ask one question in interview “If we have onecomplex query which JOIN five table, what would work best from .NETapplication, Call of Query or Stored Procedure?” 80% of candidate used to say “SPworks fast”. My next question always would be “WHY?” than out of those 80%, 60%would say “Because SP is a compiled code”. After that I fire my follow upquestion, I can see option of “Compile page or project in .NET but I never seethat kind of option in SSMS, how do you compile your SP?”, once I EXECUTE thisquestion, 90% candidate of those 60% candidate would like to keep silence ordivert the talk.
Anyway, intention of this article is to let you know somefacts about SP compilation & Recompilation. Since it is big topic andwouldn’t cover under on article, I may come up with some more articles on thistopic but right now, let me explain some basic facts only.
First of all, let me tell you that you don’t need to compileStored Procedure manually, when you execute it, SQL Server will compile yourStored Procedure for you and save the execution plan for future use so that itdoesn’t need to compile again and again, this is generic understanding, itdoesn’t work all time as few facts are there which cause recompilation manytime or every time. If you want to recompile your Stored Procedure manually,you should use “SP_Recompile” Stored Procedure given by SQL Server.
Now, you think if recompilation is that bad than whyMicrosoft has given facility to recompile? Well, let me have an opportunity tobreak the ice, recompilation of stored procedure is not always bad. It may bebeneficial or may be harmful, it is totally depends on the situation.
Actually compilation of Stored Procedure stores theexecution plan first time you execute your Stored Procedure and every follow upcall would use the same execution plan but recompilation of SP would be helpfulif you have new statistics or new index on the table. BTW, in SQL Server 2008+there is in-built functionality to recompile at statement level rather thanrecompiling whole stored procedure which is less resource centric. 
Following is the list of basic cause which forces StoredProcedure to recompile.
·         
  • Change in SET option within Stored Procedure
  • Execution plan is very old
  • Schema change in table, index, view or temptables which are used in Stored Procedure
  •  “Deferred object resolution”, means object wasnot available while compiling Stored Procedure  but you have created later on,may be some temp table you have created in Stored Procedure.
  •  Call of “SP_Recompile” Stored Procedure.
  • Call of RECOMPILE clause in Stored Procedure.
  • Statistics are old
How to avoid Stored Procedure recompilations?
  • Avoid using temp table or other DDL statements as long aspossible.
  • Use table variable in Stored Procedure if needed
  • Avoid changing SET option in Stored Procedure likeARITHABORT, Quoted_Identifier, ANSI_NULLS, ANSI_WARNINGS etc.
  • Avoiding recompilation by statistics change by using “KEEPFIXEDPLAN” hint.
  • Disable Auto Update statistics for your database.
Well, these are very basic understanding and each point ofthis article may consume separate dedicated article and I may come up withseries on this very soon.
Reference: Ritesh Shah
http://www.sqlhub.com
Note: Microsoft Books online is a default reference of all articles butexamples and explanations prepared by Ritesh Shah, founder of
http://www.SQLHub.com
Ask me any SQL Server related question at my “ASK Profile

Do I need backup of my database?

If you are working as a DBA or you have responsibility tomaintain your SQL Server database up & running, take my words, you have NOOPTION of full database backup.
Recently I came to know that one person is not taking backupjust because he has RAID and Mirroring setup so he is least worried aboutbackup. Believe me, no matter what redundant hardware / software and highavailability setup you have, you MUST take backup regularly because even goodand up-to-date disaster recovery system could get failed. So, you never know,when and how do you need your latest backup file.
Let us discuss why database backup is really that muchimportant.
Suppose you have high availability solutions like RAID orserver clustering but what if you hardware gets corrupt and may be completedisk array get corrupt? You have to build up your database from backup only.
Sometime, catastrophic events or natural disasters likeflooding, earth quake etc. could damage your hardware or may be complete datacenter and if you have no backup at other location, you have no place to gofor.
May be sometime Security issues comes to the picture andsomebody intentionally or unintentionally damage you data which could beaffected adversely and in that situation you might look for the latest databasebackup to restore that data.
Sometime your OS or even your SQL Server gets corrupted andyou might need to restore data from backup itself.
In short, I would HIGHLY recommend database backup as “Preventionis always better than cure”. So, keep strong backup policies with full databasebackup, differential database backup and transaction log backup, depends onyour need and business policies.
BTW, I am not against the high availability concepts likemirroring, clustering, replication or log shipping. You should implement thosetoo as per your business needs but also MUST keep strong backup policies.
Reference: Ritesh Shah
http://www.sqlhub.com
Note: Microsoft Books online is a default reference of all articles butexamples and explanations prepared by Ritesh Shah, founder of
http://www.SQLHub.com
Ask me any SQL Server related question at my “ASK Profile

Do I need backup of my database?

If you are working as a DBA or you have responsibility tomaintain your SQL Server database up & running, take my words, you have NOOPTION of full database backup.
Recently I came to know that one person is not taking backupjust because he has RAID and Mirroring setup so he is least worried aboutbackup. Believe me, no matter what redundant hardware / software and highavailability setup you have, you MUST take backup regularly because even goodand up-to-date disaster recovery system could get failed. So, you never know,when and how do you need your latest backup file.
Let us discuss why database backup is really that muchimportant.
Suppose you have high availability solutions like RAID orserver clustering but what if you hardware gets corrupt and may be completedisk array get corrupt? You have to build up your database from backup only.
Sometime, catastrophic events or natural disasters likeflooding, earth quake etc. could damage your hardware or may be complete datacenter and if you have no backup at other location, you have no place to gofor.
May be sometime Security issues comes to the picture andsomebody intentionally or unintentionally damage you data which could beaffected adversely and in that situation you might look for the latest databasebackup to restore that data.
Sometime your OS or even your SQL Server gets corrupted andyou might need to restore data from backup itself.
In short, I would HIGHLY recommend database backup as “Preventionis always better than cure”. So, keep strong backup policies with full databasebackup, differential database backup and transaction log backup, depends onyour need and business policies.
BTW, I am not against the high availability concepts likemirroring, clustering, replication or log shipping. You should implement thosetoo as per your business needs but also MUST keep strong backup policies.
Reference: Ritesh Shah
http://www.sqlhub.com
Note: Microsoft Books online is a default reference of all articles butexamples and explanations prepared by Ritesh Shah, founder of
http://www.SQLHub.com
Ask me any SQL Server related question at my “ASK Profile