Comments on: Retrieving N rows per group

By: Rob Farley

Rob Farley — Thu, 01 Jan 2015 02:34:13 +0000

Fixed now, Kevin. Thanks for pointing out my silly error.

By: Rob Farley

Rob Farley — Wed, 31 Dec 2014 23:46:08 +0000

Yes – my focus on that last query was to demonstrate the impact of the residual predicate. I’ll tweak the post at some point soon.

By: TheSQLGuru

TheSQLGuru — Wed, 31 Dec 2014 22:30:25 +0000

Rob, I am pretty sure your final query has a very common logic flaw that is causing incorrect results. You are filtering in the WHERE clause based on the LEFT JOINed table. This restricts the rows improperly, giving 961 rows of output instead of the expected 1045. Shouldn’t the query be written thusly:
WITH Numbered AS
(
SELECT h.TransactionID, h.ProductID, h.TransactionDate, ROW_NUMBER() OVER (PARTITION BY ProductID ORDER BY TransactionDate DESC) AS RowNum
FROM Production.TransactionHistory h
)
SELECT p.Name, p.ProductID, t.TransactionID, t.TransactionDate
FROM Production.Product p
LEFT JOIN Numbered t ON t.ProductID = p.ProductID
AND t.RowNum <= 5 * p.DaysToManufacture
WHERE p.Name >= ‘M’ AND p.Name < ‘S’;
Or perhaps you intended to include only the WHERE clause rows? In this case an INNER JOIN would be more appropriate I think.

By: Rob Farley

Rob Farley — Thu, 25 Dec 2014 13:41:44 +0000

Yes – the index I’m using there matches that.

By: Uri Dimant

Uri Dimant — Thu, 25 Dec 2014 13:39:16 +0000

Hi Rob – Mery Christmas !
We want eliminate sorts by having an appropriate index. I like the concept POC introduced by Itzik Be-Gan to have an index on F (filtering) P (Partition) O (Orders) and C – Covering :-)))
Thanks

By: Rob Farley

Rob Farley — Thu, 25 Dec 2014 12:38:33 +0000

Hi Uri – Merry Christmas!
Yes, your mileage can vary depending on various factors. However, you should still be wary of Sort operators, and it’s good to be aware of the options. I generally prefer APPLY, but have been known to use both.
Rob

By: Uri Dimant

Uri Dimant — Thu, 25 Dec 2014 10:23:01 +0000

Hi Rob
If the dens is high,for example we have 10 customers with 100000000 orders then probably using cross apply which produces seek (if index exists ) will be optimal option.. On the other hand, row number will be better to have a scan if the dens is low..Just my two cents