SQL Windowing Ranks Functions - sql

Question

SELECT
*
FROM (
SELECT
Product,
SalesAmount,
ROW_NUMBER() OVER (ORDER BY SalesAmount DESC) as RowNum,
RANK() OVER (ORDER BY SalesAmount DESC) as RankOf2007,
DENSE_RANK() OVER (ORDER BY SalesAmount DESC) as DRankOf2007
FROM (
SELECT
c.EnglishProductName as Product,
SUM(a.SalesAmount) as SalesAmount,
b.CalendarYear as CalenderYear
FROM FactInternetSales a
INNER JOIN DimDate b
ON a.OrderDateKey=b.DateKey
INNER JOIN DimProduct c
ON a.ProductKey=c.ProductKey
WHERE b.CalendarYear IN (2007)
GROUP BY c.EnglishProductName,b.CalendarYear
) Sales
) Rankings
WHERE [RankOf2007] <= 5
ORDER BY [SalesAmount] DESC
I am currently sorting products based on summation of Sales Amount in descending fashion and getting rank based on the summation of sales amount of every product in 2007 and ranking product 1 if it has the highest Sales Amount in that year and so forth.
Currently my database table looks like the one mentioned in the image (apart from RankOf2008 and DRankOf2008 columns), I would like to have rankings in year 2008 for same top 5 products of 2007 (Null value if any of those top 5 products of 2007 are unsold in 2008) in the same table with side by side columns as shown in the image above.

Answer 1

May be you require something like this.
First getting ranks for all products then partition by year, that is rank of products year wise and fetching required data with help of CTE.
WITH cte
AS (
SELECT *
FROM (
SELECT Product
,SalesAmount
,CalenderYear
,ROW_NUMBER() OVER (
PARTITION BY CalenderYear ORDER BY SalesAmount DESC
) AS RowNum
,RANK() OVER (
PARTITION BY CalenderYear ORDER BY SalesAmount DESC
) AS RankOf2007
,DENSE_RANK() OVER (
PARTITION BY CalenderYear ORDER BY SalesAmount DESC
) AS DRankOf2007
FROM (
SELECT c.EnglishProductName AS Product
,SUM(a.SalesAmount) AS SalesAmount
,b.CalendarYear AS CalenderYear
FROM FactInternetSales a
INNER JOIN DimDate b ON a.OrderDateKey = b.DateKey
INNER JOIN DimProduct c ON a.ProductKey = c.ProductKey
--WHERE b.CalendarYear IN (2007)
GROUP BY c.EnglishProductName
,b.CalendarYear
) Sales
) Rankings
--WHERE [RankOf2007] <= 5
--ORDER BY [SalesAmount] DESC
)
SELECT a.*
,b.DRankOf2007 AS [DRankOf2008]
,b.RankOf2007 AS [RankOf2008]
FROM cte a
LEFT JOIN cte b ON a.Product = b.Product
AND b.CalenderYear = 2008
WHERE a.CalenderYear = 2007
AND a.[RankOf2007] <= 5

Answer 2

Use conditional aggregation in your innermost query (i.e. select both years and sum conditionally for one of the years):
select
p.productkey,
p.englishproductname as product,
ranked.salesamount2007,
ranked.salesamount2008,
ranked.rankof2007,
ranked.rankof2008
from
(
select
productkey,
salesamount2007,
salesamount2008,
rank() over (order by salesamount2007 desc) as rankof2007,
rank() over (order by salesamount2008 desc) as rankof2008
from
(
select
s.productkey,
sum(case when d.calendaryear = 2007 then s.salesamount end) as salesamount2007,
sum(case when d.calendaryear = 2008 then s.salesamount end) as salesamount2008
from factinternetsales s
inner join dimdate d on d.datekey = s.orderdatekey
where d.calendaryear in (2007, 2008)
group by s.productkey
) aggregated
) ranked
join dimproduct p on p.productkey = ranked.productkey
where ranked.rankof2007 <= 5
order by ranked.rankof2007 desc;
For the case there are no rows for a product in 2008, salesamount2008 will be null. In standard SQL we would consider this in the ORDER BY clause:
rank() over (order by salesamount2008 desc nulls last) as rankof2008
But SQL Server doesn't comply with the SQL standard here and doesn't feature NULLS FIRST/LAST in the ORDER BY clause. Fortunately, it sorts nulls last when sorting in descending order, so it implicitly does just what we want here.
By the way: we could do the aggregation and ranking in a single step, but in that case we'd have to repeat the SUM expressions. It's a matter of personal preference, whether to do this in one step (shorter query) or two steps (no repetitive expressions).