Optimise query with count and order by function - sql-server

Question

I have a problem with the optimization of this query, I have 3 tables (Products = Catalogo.GTIN, Sales Header = TEDEF.Factura and Sales Detail = TEDEF.Farmacia).
The query tries to find the Mode of the column VPRODEXENIGV_FAR. This query without the ORDER BY executes in less than 3 seconds (the table of details has about 30 million rows).
But when I add the ORDER BY clause, the query now takes more than 30 minutes to run.
I want to know how can I optimize this query or the indexes that I need to optimize this.
SELECT *
FROM Catalogo.GTIN G
CROSS APPLY
(SELECT TOP 1
COUNT(FAR.VPRODEXENIGV_FAR) [ROW],
YEAR(FAC2.VFECEMI_FAC) [AÑO],
MONTH(FAC2.VFECEMI_FAC) [MES],
FAR.VCODPROD_FAR_003,
CASE WHEN FAR.VPRODEXENIGV_FAR = 'A' THEN 1 ELSE 0 END AfectoIGV
FROM
TEDEF.Factura FAC2
INNER JOIN
TEDEF.Farmacia FAR ON FAC2.VTDOCPAGO_FAC = FAR.VTDOCPAGO_FAC
AND FAC2.VNDOCPAGO_FAC = FAR.VNDOCPAGO_FAC
WHERE
G.CODIGO = FAR.VCODPROD_FAR_003
GROUP BY
YEAR(FAC2.VFECEMI_FAC),
MONTH(FAC2.VFECEMI_FAC),
FAR.VCODPROD_FAR_003,
FAR.VPRODEXENIGV_FAR
ORDER BY
1 DESC --- <----- THE PROBLEM IS HERE
) GG

Answer 1

Ouch! You have a hugely expensive dependent subquery. It's expensive because SELECT TOP(n) ... ORDER BY col DESC does a whole lot of work to create a result set only to discard all but one row. And, it's a dependent subquery so it's run for every row of Catalogo.GTIN .
It looks like you want to count the resultset rows in the most recent month and year for each Catalogo.GTIN row. So, let's try to refactor your query to do that.
We'll start with a subquery to grab the month-start date of the latest Factura row for each catalog entry.
SELECT CODIGO,
DATEFROMPARTS(YEAR(maxd), MONTH(maxd),1) maxmes
FROM (
SELECT MAX(FAC2.VFECEMI_FAC) maxd,
G.CODIGO
FROM Catalogo.GTIN G
JOIN TDEF.Farmacia FAR
ON G.CODIGO = FAR.VCODPROD_FAR_003
JOIN TEDEF.Factura FAC2
ON FAC2.VTDOCPAGO_FAC = FAR.VTDOCPAGO_FAC
AND FAC2.VNDOCPAGO_FAC = FAR.VNDOCPAGO_FAC
GROUP BY G.CODIGO
) maxd
It's wise to test this and make sure it works correctly and performs tolerably well. If you test it in SSMS, you can use "Show Actual Execution Plan" and see if it recommends an extra index. This subquery need only be run once, rather than once per G.CODIGO row.
Then we'll use it in your larger query.
SELECT G.*,
COUNT(FAR.VPRODEXENIGV_FAR) [ROW],
YEAR(FAC2.VFECEMI_FAC) [AÑO],
MONTH(FAC2.VFECEMI_FAC) [MES],
FAR.VCODPROD_FAR_003,
CASE WHEN FAR.VPRODEXENIGV_FAR = 'A' THEN 1 ELSE 0 END AfectoIGV
FROM Catalogo.GTIN G
JOIN (
SELECT CODIGO,
DATEFROMPARTS(YEAR(maxd), MONTH(maxd),1) maxmes
FROM (
SELECT MAX(FAC2.VFECEMI_FAC) maxd,
G.CODIGO
FROM Catalogo.GTIN G
JOIN TDEF.Farmacia FAR
ON G.CODIGO = FAR.VCODPROD_FAR_003
JOIN TEDEF.Factura FAC2
ON FAC2.VTDOCPAGO_FAC = FAR.VTDOCPAGO_FAC
AND FAC2.VNDOCPAGO_FAC = FAR.VNDOCPAGO_FAC
GROUP BY G.CODIGO
) maxd
) maxmes ON G.CODIGO = maxmes.CODIGO
JOIN TEDEF.Farmacia FAR
ON G.CODIGO = FAR.VCODPROD_FAR_003
JOIN TEDEF.Factura FAC2
ON FAC2.VTDOCPAGO_FAC = FAR.VTDOCPAGO_FAC
AND FAC2.VNDOCPAGO_FAC = FAR.VNDOCPAGO_FAC
AND FAC2.VFECEMI_FAC >= maxmes.maxmes
GROUP BY maxmes.maxmes,
G.CODIGO,
FAR.VCODPROD_FAR_003,
FAR.VPRODEXENIGV_FAR
Here is the tricky bit:
DATEFROMPARTS(YEAR(maxd), MONTH(maxd),1) maxmes turns any date maxd into the first day of that month.
And, FAC2.VFECEMI_FAC >= maxmes.maxmes filters out rows before the first day of that month (for that CODIGO). It does so in a sargable way: a way that can exploit an index on FAC2.VFECEMI_FAC.
That is an alternative way to do TOP(1) ORDER BY d DESC. And faster.
It's all about sets of rows. Especially when using GROUP BY, it's performance-helpful to limit the number of rows in each set.
Obviously I cannot debug this.

Answer 2

Is me again, Finally i resolve the problem of the optimization, now the query delay is about 20 sec (with the sort instruction and with the count in a table over 30 million rows) i hope this way can help others or could be optimice more by the community.
I resolve the problem applying the sort but with the Row_Number instruction, in that way the server take my index for the sort instruction and make the magic:
WITH x
AS
(
SELECT *, ROW_NUMBER() OVER (PARTITION BY GG.COD, GG.[AÑO], GG.[MES] ORDER BY GG.[ROW] DESC) [ID]
FROM Catalogo.GTIN G
CROSS APPLY
(
SELECT COUNT(FAR.VPRODEXENIGV_FAR) [ROW]
, YEAR(FAC2.VFECEMI_FAC) [AÑO]
, MONTH(FAC2.VFECEMI_FAC) [MES]
, FAR.VCODPROD_FAR_003 [COD]
, CASE WHEN FAR.VPRODEXENIGV_FAR = 'A' THEN 1 ELSE 0 END AfectoIGV
FROM TEDEF.Factura FAC2
INNER JOIN TEDEF.Farmacia FAR
ON FAC2.VTDOCPAGO_FAC = FAR.VTDOCPAGO_FAC
AND FAC2.VNDOCPAGO_FAC = FAR.VNDOCPAGO_FAC
WHERE G.CODIGO = FAR.VCODPROD_FAR_003
GROUP BY YEAR(FAC2.VFECEMI_FAC)
, MONTH(FAC2.VFECEMI_FAC)
, FAR.VCODPROD_FAR_003
, FAR.VPRODEXENIGV_FAR
-- ORDER BY 1 DESC --- <---- this is the bad guy, please, don't do that xD
) GG
) SELECT *
FROM x WHERE ID = 1
In that way i can sort the Count instruction and calculate the Mode for the Column FAR.VPRODEXENIGV_FAR