How could I replace a T-SQL cursor? - sql-server

Question

I would like to ask you how I could replace a cursor that I've inserted into my stored procedure.
Actually, we found that cursor is the only way out to manage my scenario, but as I've read this is not a best practise.
This is my scenario:I have to calculate recursively the stock row by row and set the season according to what has been calculated in the previous rows.
I can set the season when the transfer type is "purchase". The others transfers should be set with the correct season by a T-SQL query.
The table where I should calculate the season has the following template and fake data, but they reflect the real situation:
Transfer Table Example
The rows that have the "FlgSeason" set as null, are calculated as follow: in ascending order, the cursor start from the row 3 and go back the previous rows and calculate the amount of stock for each season and then update the column season with the minimum season with stock.
Here's the code I used:
CREATE TABLE [dbo].[transfers]
(
[rowId] [int] NULL,
[area] [int] NULL,
[store] [int] NULL,
[item] [int] NULL,
[date] [date] NULL,
[type] [nvarchar](50) NULL,
[qty] [int] NULL,
[season] [nvarchar](50) NULL,
[FlagSeason] [int] NULL
) ON [PRIMARY]
INSERT INTO [dbo].[transfers]
([rowId]
,[area]
,[store]
,[item]
,[date]
,[type]
,[qty]
,[season]
,[FlagSeason])
VALUES (1,1,20,300,'2015-01-01','Purchase',3,'2015-FallWinter',1)
, (2,1,20,300,'2015-01-01','Purchase',4,'2016-SpringSummer',1)
, (3,1,20,300,'2015-01-01','Sales',-1,null,null)
, (4,1,20,300,'2015-01-01','Sales',-2,null,null)
, (5,1,20,300,'2015-01-01','Sales',-1,null,null)
, (6,1,20,300,'2015-01-01','Sales',-1,null,null)
, (7,1,20,300,'2015-01-01','Purchase',4,'2016-FallWinter',1)
, (8,1,20,300,'2015-01-01','Sales',-1,null,null)
DECLARE #RowId as int
DECLARE db_cursor CURSOR FOR
Select RowID
from Transfers
where [FlagSeason] is null
order by RowID
OPEN db_cursor
FETCH NEXT FROM db_cursor INTO #RowId
WHILE ##FETCH_STATUS = 0
BEGIN
Update Transfers
set Season = (Select min (Season) as Season
from (
Select
Season
, SUM(QTY) as Qty
from Transfers
where RowID < #RowId
and [FlagSeason] = 1
group by Season
having Sum(QTY) > 0
)S
where s.QTY >= 0
)
, [FlagSeason] = 1
where rowId = #RowId
FETCH NEXT FROM db_cursor INTO #RowId
end
In this case the query would extract:
3 qty for season 2015 FW
4 for 2016 SS.
Than The update statment will set 2015-fw (the min over the two season with qty).
Then the courson go forward the row 4, and runs again the query to extract the stock updated considering the calculation at row 3. So the result should be
QTY 2 For 2015 FW
QTY 4 FOr 2016 SS
and then the update would set 2015 FW.
And so on.
The final output should be something like this:
Output
Actually, the only way-out was to implement a cursor and now it takes above 30/40 minutes to scan and update about 2,5 million rows. Do anybody know a solution without recurring to a cursor?
Thanks in advance!

Answer 1

Updated to run on 2008
IF OBJECT_ID('tempdb..#transfer') IS NOT NULL
DROP TABLE #transfer;
GO
CREATE TABLE #transfer (
RowID INT IDENTITY(1, 1) PRIMARY KEY NOT NULL,
Area INT,
Store INT,
Item INT,
Date DATE,
Type VARCHAR(50),
Qty INT,
Season VARCHAR(50),
FlagSeason INT
);
INSERT INTO #transfer ( Area,
Store,
Item,
Date,
Type,
Qty,
Season,
FlagSeason
)
VALUES (1, 20, 300, '20150101', 'Purchase', 3, '2015-SpringSummer', 1),
(1, 20, 300, '20150601', 'Purchase', 4, '2016-SpringSummer', 1),
(1, 20, 300, '20150701', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20150721', 'Sales', -2, NULL, NULL),
(1, 20, 300, '20150901', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20160101', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170101', 'Purchase', 4, '2017-SpringSummer', 1),
(1, 20, 300, '20170125', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170201', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170225', 'Sales', -1, NULL, NULL),
(1, 21, 301, '20150801', 'Purchase', 4, '2017-SpringSummer', 1),
(1, 21, 301, '20150901', 'Sales', -1, NULL, NULL),
(1, 21, 301, '20151221', 'Sales', -2, NULL, NULL),
(1, 21, 302, '20150801', 'Purchase', 1, '2016-SpringSummer', 1),
(1, 21, 302, '20150901', 'Purchase', 1, '2017-SpringSummer', 1),
(1, 21, 302, '20151101', 'Sales', -1, NULL, NULL),
(1, 21, 302, '20151221', 'Sales', -1, NULL, NULL),
(1, 20, 302, '20150801', 'Purchase', 1, '2016-SpringSummer', 1),
(1, 20, 302, '20150901', 'Purchase', 1, '2017-SpringSummer', 1),
(1, 20, 302, '20151101', 'Sales', -1, NULL, NULL),
(1, 20, 302, '20151221', 'Sales', -1, NULL, NULL);
WITH Purchases
AS (SELECT t1.RowID,
t1.Area,
t1.Store,
t1.Item,
t1.Date,
t1.Type,
t1.Qty,
t1.Season,
RunningInventory = ( SELECT SUM(t2.Qty)
FROM #transfer AS t2
WHERE t1.Type = t2.Type
AND t1.Area = t2.Area
AND t1.Store = t2.Store
AND t1.Item = t2.Item
AND t2.Date <= t1.Date
)
FROM #transfer AS t1
WHERE t1.Type = 'Purchase'
),
Sales
AS (SELECT t1.RowID,
t1.Area,
t1.Store,
t1.Item,
t1.Date,
t1.Type,
t1.Qty,
t1.Season,
RunningSales = ( SELECT SUM(ABS(t2.Qty))
FROM #transfer AS t2
WHERE t1.Type = t2.Type
AND t1.Area = t2.Area
AND t1.Store = t2.Store
AND t1.Item = t2.Item
AND t2.Date <= t1.Date
)
FROM #transfer AS t1
WHERE t1.Type = 'Sales'
)
SELECT Sales.RowID,
Sales.Area,
Sales.Store,
Sales.Item,
Sales.Date,
Sales.Type,
Sales.Qty,
Season = ( SELECT TOP 1
Purchases.Season
FROM Purchases
WHERE Purchases.Area = Sales.Area
AND Purchases.Store = Sales.Store
AND Purchases.Item = Sales.Item
AND Purchases.RunningInventory >= Sales.RunningSales
ORDER BY Purchases.Date, Purchases.Season
)
FROM Sales
UNION ALL
SELECT Purchases.RowID ,
Purchases.Area ,
Purchases.Store ,
Purchases.Item ,
Purchases.Date ,
Purchases.Type ,
Purchases.Qty ,
Purchases.Season
FROM Purchases
ORDER BY Sales.Area, Sales.Store, item, Sales.Date
*original answer below**
I don't understand the purpose of the flagseason column so I didn't include that. Essentially, this calculates a running sum for purchases and sales and then finds the season that has a purchase_to_date inventory of at least the sales_to_date outflow for each sales transaction.
IF OBJECT_ID('tempdb..#transfer') IS NOT NULL
DROP TABLE #transfer;
GO
CREATE TABLE #transfer (
RowID INT IDENTITY(1, 1) PRIMARY KEY NOT NULL,
Area INT,
Store INT,
Item INT,
Date DATE,
Type VARCHAR(50),
Qty INT,
Season VARCHAR(50),
FlagSeason INT
);
INSERT INTO #transfer ( Area,
Store,
Item,
Date,
Type,
Qty,
Season,
FlagSeason
)
VALUES (1, 20, 300, '20150101', 'Purchase', 3, '2015-FallWinter', 1),
(1, 20, 300, '20150601', 'Purchase', 4, '2016-SpringSummer', 1),
(1, 20, 300, '20150701', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20150721', 'Sales', -2, NULL, NULL),
(1, 20, 300, '20150901', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20160101', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170101', 'Purchase', 4, '2016-FallWinter', 1),
(1, 20, 300, '20170201', 'Sales', -1, NULL, NULL);
WITH Inventory
AS (SELECT *,
PurchaseToDate = SUM(CASE WHEN Type = 'Purchase' THEN Qty ELSE 0 END) OVER (ORDER BY Date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW),
SalesToDate = ABS(SUM(CASE WHEN Type = 'Sales' THEN Qty ELSE 0 END) OVER (ORDER BY Date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW))
FROM #transfer
)
SELECT Inventory.RowID,
Inventory.Area,
Inventory.Store,
Inventory.Item,
Inventory.Date,
Inventory.Type,
Inventory.Qty,
Season = CASE
WHEN Inventory.Season IS NULL
THEN ( SELECT TOP 1
PurchaseToSales.Season
FROM Inventory AS PurchaseToSales
WHERE PurchaseToSales.PurchaseToDate >= Inventory.SalesToDate
ORDER BY Inventory.Date
)
ELSE
Inventory.Season
END,
Inventory.PurchaseToDate,
Inventory.SalesToDate
FROM Inventory;
*UPDATED*******************************
You'll need an index on your data to help with the sorting in order to make this perform.
Possibly:
CREATE NONCLUSTERED INDEX IX_Transfer ON #transfer(Store, Item, Date) INCLUDE(Area,Qty,Season,Type)
You should see a index scan on the named index. It will not be a seek because the sample query does not filter any data and all of the data is included.
In addition, you need to remove Season from the Partition By clause of the SalesToDate. Resetting the sales for each season will throw your comparisons off because the rolling sales need to be compared to the rolling inventory in order for you to determine the source of sales inventory.
Two other tips for the partition clause:
Don't duplicate the fields between partition by and order by. The order of the partition fields doesn't matter since the aggregate is reset for each partition. At best, the ordered partition field will be ignored, at worst it may cause the optimizer to aggregate the fields in a particular order. This does not have any effect on the results, but can added unnecessary overhead.
Make sure your index matches the definition of the partition by/order by clause.
The index should be [partitioning fields, sequence doesn't matter] + [ordering fields, sequence needs to match order by clause].
In your scenario, the indexed columns should be on store, item, and then date. If date were before store or item, the index would not be used because the optimizer will need to first handle partitioning by store & item before sorting by date.
If you may have multiple areas in your data, the index and partition clauses would need to be
index: area, store, item, date
partition by: area, store, item order by date

Answer 2

Referring to Wes's answer, the solution proposed is almost fine. It works good but I've noticed that the assignment of the season doesn't work properly beacause, in my scenario, the stock should be calculated and updated by store and item itself. I've Updated the script adding some adjstments. Moreover, I've added some new "Fake" data to understand better my scenario and how it should work.
IF OBJECT_ID('tempdb..#transfer') IS NOT NULL
DROP TABLE #transfer;
GO
CREATE TABLE #transfer (
RowID INT IDENTITY(1, 1) PRIMARY KEY NOT NULL,
Area INT,
Store INT,
Item INT,
Date DATE,
Type VARCHAR(50),
Qty INT,
Season VARCHAR(50),
FlagSeason INT
);
INSERT INTO #transfer ( Area,
Store,
Item,
Date,
Type,
Qty,
Season,
FlagSeason
)
VALUES (1, 20, 300, '20150101', 'Purchase', 3, '2015-SpringSummer', 1),
(1, 20, 300, '20150601', 'Purchase', 4, '2016-SpringSummer', 1),
(1, 20, 300, '20150701', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20150721', 'Sales', -2, NULL, NULL),
(1, 20, 300, '20150901', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20160101', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170101', 'Purchase', 4, '2017-SpringSummer', 1),
(1, 20, 300, '20170125', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170201', 'Sales', -1, NULL, NULL),
(1, 20, 300, '20170225', 'Sales', -1, NULL, NULL),
(1, 21, 301, '20150801', 'Purchase', 4, '2017-SpringSummer', 1),
(1, 21, 301, '20150901', 'Sales', -1, NULL, NULL),
(1, 21, 301, '20151221', 'Sales', -2, NULL, NULL),
(1, 21, 302, '20150801', 'Purchase', 1, '2016-SpringSummer', 1),
(1, 21, 302, '20150901', 'Purchase', 1, '2017-SpringSummer', 1),
(1, 21, 302, '20151101', 'Sales', -1, NULL, NULL),
(1, 21, 302, '20151221', 'Sales', -1, NULL, NULL),
(1, 20, 302, '20150801', 'Purchase', 1, '2016-SpringSummer', 1),
(1, 20, 302, '20150901', 'Purchase', 1, '2017-SpringSummer', 1),
(1, 20, 302, '20151101', 'Sales', -1, NULL, NULL),
(1, 20, 302, '20151221', 'Sales', -1, NULL, NULL)
;
WITH Inventory
AS (SELECT *,
PurchaseToDate = SUM(CASE WHEN Type = 'Purchase' THEN Qty ELSE 0 END) OVER (partition by store, item ORDER BY store, item,Date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW),
SalesToDate = ABS(SUM(CASE WHEN Type = 'Sales' THEN Qty ELSE 0 END) OVER (partition by store, item,season ORDER BY store, item, Date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW))
FROM #transfer
)
SELECT Inventory.RowID,
Inventory.Area,
Inventory.Store,
Inventory.Item,
Inventory.Date,
Inventory.Type,
Inventory.Qty,
Season = CASE
WHEN Inventory.Season IS NULL
THEN ( SELECT TOP 1
PurchaseToSales.Season
FROM Inventory AS PurchaseToSales
WHERE PurchaseToSales.PurchaseToDate >= Inventory.SalesToDate
and PurchaseToSales.Item = inventory.item --//Added
and PurchaseToSales.store = inventory.store --//Added
and PurchaseToSales.Area = Inventory.area --//Added
ORDER BY Inventory.Date
)
ELSE
Inventory.Season
END,
Inventory.PurchaseToDate,
Inventory.SalesToDate
FROM Inventory
Here the output:
enter image description here
After these adjustments, it works fine, but if I switch the fake data with the real data that are within in a 6 milions row data table, the query becomes very slow (~400 rows extracted per minutes) because of the insert of these check inside the where clause of the subquery:
WHERE PurchaseToSales.PurchaseToDate >= Inventory.SalesToDate
and PurchaseToSales.Item = inventory.item --//Added
and PurchaseToSales.store = inventory.store --//Added
and PurchaseToSales.Area = Inventory.area --//Added
I've tryed to replace the subquery with the "Cross Apply" function but nothing has changed. Am I Missing somethings?
Thanks in advance