September 3, 2013 at 12:00 am
hi there,
hope it's the right forum 😉
I have an example where I am not sure about setting my primary key and my clustered index for best performance and used storage size.
I have a table containing a summary of parcels that were distributed for customers.
A customer can buy an additional service for a parcel.
Here's the (simplified!) table with some sample data.
create table #ShippingNote(
DeliveryDay date not null,
DebitorNumber char(5) not null,
DestinationPostalCode varchar(10) not null,
AdditionalService varchar(20) null,
ParcelCount int not null
);
go
insert into #ShippingNote values
( '20130801', '12345', '6200', 'PRIORITY', 5 )
,( '20130801', '12345', '6200', NULL, 7 )
,( '20130801', '12345', '6200', 'DANGEROUS_GOOD', 10 )
,( '20130801', '12345', '8020', 'PRIORITY', 3 )
,( '20130801', '12345', '8020', 'FRAGILE', 12 )
,( '20130801', '98765', '6200', 'PRIORITY', 9 )
,( '20130801', '98765', '6200', NULL, 3 )
,( '20130801', '98765', '6200', 'DANGEROUS_GOOD', 7 )
,( '20130801', '98765', '8020', 'PRIORITY', 11 )
,( '20130801', '98765', '8020', 'FRAGILE', 5 )
,( '20130802', '12345', '6200', 'PRIORITY', 3 )
,( '20130802', '12345', '6200', NULL, 1 )
,( '20130802', '12345', '6200', 'DANGEROUS_GOOD', 15 )
,( '20130802', '12345', '8020', 'PRIORITY', 3 )
,( '20130802', '12345', '8020', 'FRAGILE', 1 )
,( '20130802', '98765', '6200', 'PRIORITY', 5 )
,( '20130802', '98765', '6200', NULL, 20 )
,( '20130802', '98765', '6200', 'DANGEROUS_GOOD', 1 )
,( '20130802', '98765', '8020', 'PRIORITY', 9 )
,( '20130802', '98765', '8020', 'FRAGILE', 8 )
Now I want a report, how many parcels we distributed per customer, per DestionationPostalCode and per AdditionalService. The report can - within a stored procedure - be filtered on a date range, the DebitorNumber and the PostalCode.
Here's the query:
declare
@FromDeliveryDay date,
@ToDeliveryDay date,
@DebitorNumber char(5),
@DestinationPostalCode varchar(10);
set @FromDeliveryDay = '20130801';
set @ToDeliveryDay = '20130802';
set @DebitorNumber = '12345';
set @DestinationPostalCode = '6200'
select
s.DebitorNumber,
s.DestinationPostalCode,
s.AdditionalService,
SumParcels = sum( s.ParcelCount )
from #ShippingNote s
where s.DeliveryDay between @FromDeliveryDay and @ToDeliveryDay and
s.DebitorNumber = @DebitorNumber and
s.DestinationPostalCode = @DestinationPostalCode
group by
s.DebitorNumber,
s.DestinationPostalCode,
s.AdditionalService
order by
s.DebitorNumber,
s.DestinationPostalCode,
s.AdditionalService
This query is the only query on this table. Yes, the table and the query are made just for this one report because it is used that much often.
For my understanding the perfect clustered index for performance would be the following:
create UNIQUE clustered index #cix_ShippingNote
on #ShippingNote ( DeliveryDay, DebitorNumber, DestinationPostalCode, AdditionalService )
The query results in an index seek and all the ordering of the data for grouping is already done.
Now the question:
I need to replicate this table to another server. This means I have to define a primary key.
In theory the perfect primary key would be the clustered index, like this:
alter table #cix_ShippingNote add constraint ShippingNote_PK
primary key clustered( DeliveryDay, DebitorNumber, DestinationPostalCode, AdditionalService )
Unfortunately, the column AdditionalService is nullable. This way I can't create the PK.
For this I added an identity column to the table and used it as a primary key. I left the clustered index as shown above.
But now, as the primary key is nonclustered, it contains the full clustered key. This means I double the size of the whole table. I don't think that makes sense.
I thought about making the PK on the identity column, clustered, and add an additional unique nonclustered index on the rest of the columns. This way the query should still do fine. It's only the "overhead" of the PK column.
Is there a better way?
Thank's a lot, WolfgangE
September 3, 2013 at 12:55 am
Primary key is not decided by a query. It's part of the design process of the table. It's a column or set of columns that uniquely (with no nulls) identifies the row.
Clustered index should organise the table primarily, be usable for queries as a secondary goal only (yes, there are people who hold to the opposite of that, some get very 'passionate' about their point of view)
Indexing basics
http://www.sqlservercentral.com/articles/Indexing/68439/
http://www.sqlservercentral.com/articles/Indexing/68563/
http://www.sqlservercentral.com/articles/Indexing/68636/
No, I haven't read through the query in detail, waaay too early for complex questions
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
September 3, 2013 at 3:11 am
I do have a natural PK. The only difficulty is: one of these columns is nullable. So I cannot create the PK constraint.
The question is: do I add a artificial PK, like an identity column, which is not used in any way, or are there better solutions, like replacing the NULL-values by an empty string or something?
September 3, 2013 at 3:16 am
I'd add an artificial primary key and put a unique constraint on the unique set of columns.
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
September 19, 2013 at 2:48 am
WolfgangE (9/3/2013)
I thought about making the PK on the identity column, clustered, and add an additional unique nonclustered index on the rest of the columns.
In this case There is no use to make PK clustred index as you are not goung to use it for Query performance better. Keep the clusterd in index on your searchable column (as mentioend in your index ) and make the PK (int) non clustered index. because in any case your query will get benefit if clustered index is on searchable columns .
-------Bhuvnesh----------
I work only to learn Sql Server...though my company pays me for getting their stuff done;-)
Viewing 5 posts - 1 through 4 (of 4 total)
You must be logged in to reply to this topic. Login to reply