Unexpected behaviour of query

Question

Unexpected behaviour of query

nil.hajare

SSC Rookie

Points: 44
More actions
January 25, 2013 at 7:45 am

#273412

Hi All,
I am using below query in sql server to find out top 5 amount and date by customer.
I have used union for the same. But it gives different counts when i ran this query again and again. Sometimes it shows count as 69071 and sometimes it 69072 with same query.
Can you please help me ?
Query:
SELECT COUNT(*)
FROM EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG c
where (CONTRACT_STATUS_DESCRIPTION !='Active')
and erp_id not in ( select
a.ERP_ID
from
(select * ,ROW_NUMBER() over (Partition BY customer_number order by ANNUAL_CONTRACT_PRICE desc) as Rank
from EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG where CONTRACT_START_DATE >= GETDATE()-730
and CONTRACT_STATUS_DESCRIPTION not in ('Free','Canceled','Cancelled')
) a
where a.rank <=5
union all
select
b.ERP_ID
from
(select *,ROW_NUMBER() over (Partition BY customer_number order by CONTRACT_START_DATE desc) as Rank
from EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG where CONTRACT_START_DATE >= GETDATE()-730
and CONTRACT_STATUS_DESCRIPTION not in ('Free','Canceled','Cancelled')
) b
where b.rank <=5
)

Viewing 7 posts - 1 through 6 (of 6 total)

You must be logged in to reply to this topic. Login to reply

Mark Cowne One Orange Chip Points: 26944 More actions · Answer 1

Without sample data it's not easy to tell, however, you're using ROW_NUMBER which can be non-deterministic and can give inconsistent results.

Try using RANK or DENSE_RANK instead of ROW_NUMBER

____________________________________________________

Deja View - The strange feeling that somewhere, sometime you've optimised this query before

How to get the best help on a forum

http://www.sqlservercentral.com/articles/Best+Practices/61537

David McKinney SSChampion Points: 10472 More actions · Answer 2

I agree that the problem's going to be connected to your row_number() and the fact that the field used in the Order by section of the OVER clause i.e. ANNUAL_CONTRACT_PRICE and/or CONTRACT_START_DATE are not unique for a customer.

If you can add the other elements of the primary key into the 2 Order By's, then you should be OK. (I think!)

brendan woulfe Hall of Fame Points: 3888 More actions · Answer 3

I was thinking it's the GETDATE()-730 part. That's non-deterministic and might explain the different results as it's part of your where clause.

nil.hajare SSC Rookie Points: 44 More actions · Answer 4

My primary key for table is erp_id and service_contract_number. I have added the erp_id in order by clause. but my top records shows wrong.

e.g. Top records by customer with annual_contract_price.

select

ERP_ID ,CUSTOMER_NUMBER,ANNUAL_CONTRACT_PRICE,Rank

from

(select csa.* ,ROW_NUMBER() over (Partition BY csa.customer_number order by csa.ANNUAL_CONTRACT_PRICE desc) as Rank

from EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG csa where csa.CONTRACT_START_DATE >= GETDATE()-730

and csa.CONTRACT_STATUS_DESCRIPTION not in ('Free','Canceled','Cancelled')) a

where a.rank <=5 and a.CUSTOMER_NUMBER='110000053'

Result:

ERP_IDCUSTOMER_NUMBERANNUAL_CONTRACT_PRICERank

C10003890-axq2011-1781100000531963.941

C10003778-axq2011-178110000053720.132

C10003775-axq2011-178110000053639.393

C10003886-axq2011-178110000053591.874

C10003847-axq2011-178110000053579.685

By Adding erp_id in order by clause:

result is defferent.

select

ERP_ID ,CUSTOMER_NUMBER,ANNUAL_CONTRACT_PRICE,Rank

from

(select csa.* ,ROW_NUMBER() over (Partition BY csa.customer_number order by

csa.erp_id,csa.ANNUAL_CONTRACT_PRICE desc) as Rank

from EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG csa where csa.CONTRACT_START_DATE >= GETDATE()-730

and csa.CONTRACT_STATUS_DESCRIPTION not in ('Free','Canceled','Cancelled')) a

where a.rank <=5 and a.CUSTOMER_NUMBER='110000053'

Result :

ERP_IDCUSTOMER_NUMBERANNUAL_CONTRACT_PRICERank

C10003731-axq2011-178110000053348 1

C10003752-axq2011-178110000053559.842

C10003753-axq2011-178110000053536.133

C10003754-axq2011-178110000053545.164

C10003755-axq2011-178110000053457.455

Plz help me?

David McKinney SSChampion Points: 10472 More actions · Answer 5

Try changing

select csa.* ,ROW_NUMBER() over (Partition BY csa.customer_number order by

csa.erp_id,csa.ANNUAL_CONTRACT_PRICE desc) as Rank

to

select csa.* ,ROW_NUMBER() over (Partition BY csa.customer_number order by

csa.ANNUAL_CONTRACT_PRICE, csa.erp_id desc) as Rank

So that if you have a draw on the ANNUAL_CONTRACT_PRICE, then the erp_id will ensure consistency in the ordering.

ChrisM@Work SSC Guru Points: 186127 More actions · Answer 6

-- This looks correct:

SELECT

ERP_ID,

customer_number,

ANNUAL_CONTRACT_PRICE,

[Rank]

FROM (

SELECT

ERP_ID,

customer_number,

ANNUAL_CONTRACT_PRICE,

ROW_NUMBER() OVER (PARTITION BY customer_number ORDER BY ANNUAL_CONTRACT_PRICE DESC) AS [Rank]

FROM EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG

WHERE CONTRACT_START_DATE >= GETDATE()-730

AND CONTRACT_STATUS_DESCRIPTION NOT IN ('Free','Canceled','Cancelled')

) a

WHERE a.[rank] <=5

AND a.CUSTOMER_NUMBER = '110000053'

-- Test it:

SELECT TOP 5 *

FROM EMEA_SERVICEMAX.Servicemax.SERVICE_CONTRACTS_STG

WHERE CONTRACT_START_DATE >= GETDATE()-730

AND CONTRACT_STATUS_DESCRIPTION NOT IN ('Free','Canceled','Cancelled')

AND a.CUSTOMER_NUMBER = '110000053'

ORDER BY ANNUAL_CONTRACT_PRICE DESC

-- Suggestions:

-- Use a variable for the cutoff date and remove the time component

SELECT @CutoffDate = DATEADD(day,DATEDIFF(day,0,GETDATE()-730),0)

^{“Write the query the simplest way. If through testing it becomes clear that the performance is inadequate, consider alternative query forms.” - Gail Shaw}

For fast, accurate and documented assistance in answering your questions, please read this article.
Understanding and using APPLY, (I) and (II) Paul White
Hidden RBAR: Triangular Joins / The "Numbers" or "Tally" Table: What it is and how it replaces a loop Jeff Moden