A Check and Foreign Key Constraint Improves Query Performance

SQLServerCentral is supported by Red Gate Software Ltd.

Popular Topics

Home

Members

Calendar

Who's On

Home » Article Discussions » Article Discussions by Author » Discuss content posted by Sarvesh Singh » A Check and Foreign Key Constraint Improves...

89 posts, Page 2 of 9

««1 2 3 4 5 »»»

A Check and Foreign Key Constraint Improves Query Performance

Rate Topic

Display Mode

Topic Options

Author

Message

ss-457805

Posted Thursday, October 21, 2010 3:37 AM

Old Hand

Group: General Forum Members
Last Login: Thursday, May 16, 2013 8:14 AM
Points: 356, Visits: 1,658

Thank you satnam

blog: http://sarveshsingh.com

Twitter: @sarveshsing

Post #1008265

peter-757102

Posted Thursday, October 21, 2010 3:47 AM

SSC Veteran

Group: General Forum Members
Last Login: Yesterday @ 3:04 AM
Points: 287, Visits: 1,901

Unfortunatly it turns out that untrusted check constrains will not always become trusted after altering the contraint as is done in this article. If you find yourself in this situation, this can come due to the constraint being configured (at create time) as not for replication. This can be seen by querying sys.foreign_keys with the condition is_not_for_replication = 1.

In this situation, which can be by design, only re-creating the foreign key without not for replication will make the desired constraint trusted.

More not for replication can be read here:

http://msdn.microsoft.com/en-us/library/ms152529(SQL.90).aspx

Post #1008269

RichardDouglas

Posted Thursday, October 21, 2010 5:30 AM

SSC Eights!

Group: General Forum Members
Last Login: Yesterday @ 3:35 AM
Points: 969, Visits: 650

Great article SS

Hope this helps,
Rich

Post #1008344

ps.

Posted Thursday, October 21, 2010 5:48 AM

SSCrazy

Group: General Forum Members
Last Login: Thursday, May 02, 2013 2:32 AM
Points: 2,236, Visits: 3,620

any idea how do we check the same in sql server 2000? sysforeignkeys doesn't have is_trusted column..

Pradeep Singh

Post #1008357

cfradenburg

Posted Thursday, October 21, 2010 6:24 AM

SSCommitted

Group: General Forum Members
Last Login: Thursday, May 16, 2013 7:44 AM
Points: 1,555, Visits: 1,925

garima.arya (10/21/2010)

If someone has added a foreign key constraint, why would he write a query like this.

Interesting point about how the query optimizer works but I'm more with Garima. If there's a foreign key constraint there what's the advantage of putting the exists check in the code? The only thing I can think of is that it will make it a little easier to follow if you're not familiar with the DB but I wouldn't think that would be worth giving the optimizer the possibility of generating an execution plan that will scan both tables.

Post #1008382

cfradenburg

Posted Thursday, October 21, 2010 6:40 AM

SSCommitted

Group: General Forum Members
Last Login: Thursday, May 16, 2013 7:44 AM
Points: 1,555, Visits: 1,925

ps. (10/21/2010)

any idea how do we check the same in sql server 2000? sysforeignkeys doesn't have is_trusted column..

I'm pretty sure that SQL 2000 doesn't have that same capability. I just disabled a foreign key constraint on one of our SQL 2000 DB, confirmed that it scanned both tables then re-enabled it and without any other action it just scanned the table that I was selecting from.

Post #1008401

walt-981324

Posted Thursday, October 21, 2010 6:56 AM

Forum Newbie

Group: General Forum Members
Last Login: Thursday, April 04, 2013 3:16 PM
Points: 1, Visits: 25

Great article. Didn't understand example of 'TeamID' in the last part. Did you mean 'CustomerID'?

Post #1008417

peter-757102

Posted Thursday, October 21, 2010 7:09 AM

SSC Veteran

Group: General Forum Members
Last Login: Yesterday @ 3:04 AM
Points: 287, Visits: 1,901

cfradenburg (10/21/2010)

garima.arya (10/21/2010)

If someone has added a foreign key constraint, why would he write a query like this.

I think that trusted foreign key constraints provide information that can be used to cut out more of an execution plan then just the simple exists used in this article. It is fact that multiple trusted references to the same target table signify that the data in those foreign key fields is directly comparable to eachother. So the query optimizer can figure this out and simplify execution plans.

As for the arguments that this only involves queries written without build in knowledge by the writer, consider views that server multiple purposes. For views it is accepted that the optimizer has to cut out unused parts based on relevance to the consuming query. The writer of the consuming query might have the knowledge, but not the means to do this for the optimizer in this instance, nor does the writer of the view.

Post #1008437

peter-757102

Posted Thursday, October 21, 2010 7:48 AM

SSC Veteran

Group: General Forum Members
Last Login: Yesterday @ 3:04 AM
Points: 287, Visits: 1,901

Here is a query I wrote for SQL Server 2008 and its problably good for 2005 as well. For every untrusted foreign key it generates a small script of either two or four commands, here is what they do:

1. Find records that are in conflict with the foreign key constraint definition.
2. Enable and re-check the foreign key constraint (only works when 'not for replication' was not used).

Optional, when 'not for replication' was used during constraint creation.

3. A drop statement for the foreign key.
4. A create statement for the foreign key, this time without 'not for replication'.

set nocount on
;
with
  fkCompleteQ as 
  (
    select
      is_not_for_replication         = fk.is_not_for_replication
    , parent_schema_id               = fk.schema_id
    , parent_object_id               = fk.parent_object_id
    , referenced_object_id           = fk.referenced_object_id
    , referenced_schema_id           = rt.schema_id
    , object_id                      = fk.object_id
    , delete_referential_action_desc = replace( fk.delete_referential_action_desc, '_', ' ' )
    , update_referential_action_desc = replace( fk.update_referential_action_desc, '_', ' ' )
    , parent_fields                  = stuff( ( select ', [' + cp.name + ']' from sys.foreign_key_columns as fkc inner join sys.columns as cp on cp.object_id = fkc.parent_object_id     and cp.column_id = fkc.parent_column_id     where fkc.constraint_object_id = fk.object_id order by fkc.constraint_column_id for xml path( '' ) ), 1, 2, '' )
    , referenced_fields              = stuff( ( select ', [' + cr.name + ']' from sys.foreign_key_columns as fkc inner join sys.columns as cr on cr.object_id = fkc.referenced_object_id and cr.column_id = fkc.referenced_column_id where fkc.constraint_object_id = fk.object_id order by fkc.constraint_column_id for xml path( '' ) ), 1, 2, '' )
    , join_fields                    = stuff( ( select ' and ( p.[' + cp.name + '] is null or p.[' + cp.name + ']=r.[' + cr.name + '] )' from sys.foreign_key_columns as fkc inner join sys.columns as cp on cp.object_id = fkc.parent_object_id     and cp.column_id = fkc.parent_column_id     
                                                                                                                                                                     inner join sys.columns as cr on cr.object_id = fkc.referenced_object_id and cr.column_id = fkc.referenced_column_id where fkc.constraint_object_id = fk.object_id order by fkc.constraint_column_id for xml path( '' ) ), 1, 5, '' )
    from
      sys.foreign_keys        as fk
        inner join sys.tables as rt  on rt.object_id = fk.referenced_object_id
    where
      fk.is_not_trusted = 1 /* and fk.is_disabled = 0 */
  )
select
  result = '/* constraint: [' + schema_name( fk.parent_schema_id )+ '].[' + object_name( fk.parent_object_id ) + '].[' + object_name( fk.object_id ) + '] */' + char(10)
         + 'select p.* from [' + schema_name( fk.parent_schema_id )+ '].[' + object_name( fk.parent_object_id ) + '] as p where not exists ( select 1 from [' + schema_name( fk.referenced_schema_id )+ '].[' + object_name( fk.referenced_object_id ) + '] as r where ' + join_fields + ');' + char(10)
         + 'alter table [' + schema_name( fk.parent_schema_id )+ '].[' + object_name( fk.parent_object_id ) + '] with check check constraint [' + object_name( fk.object_id ) + '];' + char(10)
         +  case
              when is_not_for_replication = 0
              then ''
              else
                  '-- alter table [' + schema_name( fk.parent_schema_id )+ '].[' + object_name( fk.parent_object_id ) + '] drop constraint [' + object_name( fk.object_id ) + '];' + char(10)
                + '-- alter table [' + schema_name( fk.parent_schema_id )+ '].[' + object_name( fk.parent_object_id ) + '] add  constraint [' + object_name( fk.object_id ) + '] foreign key ( ' + parent_fields + ' ) references [' + schema_name( fk.referenced_schema_id )+ '].[' + object_name( fk.referenced_object_id ) + '] ( ' + referenced_fields + ' ) on DELETE ' + fk.delete_referential_action_desc + ' on UPDATE ' + fk.update_referential_action_desc +  ';' + char(10)
            end  
from
  fkCompleteQ as fk
order by
  fk.parent_schema_id
, fk.parent_object_id
, fk.referenced_schema_id
, fk.referenced_object_id
, fk.object_id
, result
;

Execute using text mode output (Ctrl-t) to get the script and copy/paste it to a new query window to examine and possibly execute parts of it. If you find parts of the resuls truncated, go to:

Tools -> Options -> Query Results -> Results to text -> Maximum number of characters displayed in each column. And set the value to a larger number (2048) should be large enough for this.

NOTE:

I modified the code a bit to take into acount optional foreign key fields.

Post #1008483

john.larsen

Posted Thursday, October 21, 2010 8:31 AM

SSC-Enthusiastic

Group: General Forum Members
Last Login: Thursday, May 16, 2013 7:31 AM
Points: 101, Visits: 70

I liked the article as it brought into focus an element of the QO that I had not considered before.

As an aside, I would like to point out that we commonly optimize exists / not exists sub-queries by using left outer joins. I've gotten into the habit now of not using 'exists' sub-queries due to the order of magnitude difference I've seen on some query executions.

In any case, it is a good article.

Post #1008534

« Prev Topic | Next Topic »

89 posts, Page 2 of 9

««1 2 3 4 5 »»»

Permissions