I have a data table as below
#data
---------------
Account AccountType
---------------
1 2
2 0
3 5
4 2
5 1
6 5
AccountType 2 is headers and 5 is totals. Meaning accounts of type 2 have to look after the next 1 or 0 to determin if its Dim value is 1 or 0. Totals of type 5 have to look up at nearest 1 or 0 to determin its Dim value. Accounts of type 1 or 0 have there type as Dim.
Accounts of type 2 appear as islands so its not enough to just check RowNumber + 1 and same goes for accounsts of type 5.
I have arrived at the following table using CTE's. But can't find a quick way to go from here to my final result of Account, AccountType, Dim for all accounts
T3
-------------------
StartRow EndRow AccountType Dim
-------------------
1 1 2 0
2 2 0 0
3 3 5 0
4 4 2 1
5 5 0 1
6 6 5 1
Below code is MS TSQL copy paste it all and see it run. The final join on the CTE select statement is extremly slow for even 500 rows it takes 30 sec. I have 100.000 rows i need to handle. I done a cursor based solution which do it in 10-20 sec thats workable and a fast recursive CTE solution that do it in 5 sec for 100.000 rows, but it dependent on the fragmentation of the #data table. I should add this is simplified the real problem have alot more dimension that need to be taking into account. But it will work the same for this simple problem.
Anyway is there a fast way to do this using joins or another set based solution.
SET NOCOUNT ON
IF OBJECT_ID('tempdb..#data') IS NOT NULL
DROP TABLE #data
CREATE TABLE #data
(
Account INTEGER IDENTITY(1,1),
AccountType INTEGER,
)
BEGIN -- TEST DATA
DECLARE @Counter INTEGER = 0
DECLARE @MaxDataRows INTEGER = 50 -- Change here to check performance
DECLARE @Type INTEGER
WHILE(@Counter < @MaxDataRows)
BEGIN
SET @Type = CASE
WHEN @Counter % 10 < 3 THEN 2
WHEN @Counter % 10 >= 8 THEN 5
WHEN @Counter % 10 >= 3 THEN (CASE WHEN @Counter < @MaxDataRows / 2.0 THEN 0 ELSE 1 END )
ELSE 0
END
INSERT INTO #data VALUES(@Type)
SET @Counter = @Counter + 1
END
END -- TEST DATA END
;WITH groupIds_cte AS
(
SELECT *,
ROW_NUMBER() OVER (PARTITION BY AccountType ORDER BY Account) - Account AS GroupId
FROM #data
),
islandRanges_cte AS
(
SELECT
MIN(Account) AS StartRow,
MAX(Account) AS EndRow,
AccountType
FROM groupIds_cte
GROUP BY GroupId,AccountType
),
T3 AS
(
SELECT I.*, J.AccountType AS Dim
FROM islandRanges_cte I
INNER JOIN islandRanges_cte J
ON (I.EndRow + 1 = J.StartRow AND I.AccountType = 2)
UNION ALL
SELECT I.*, J.AccountType AS Dim
FROM islandRanges_cte I
INNER JOIN islandRanges_cte J
ON (I.StartRow - 1 = J.EndRow AND I.AccountType = 5)
UNION ALL
SELECT *, AccountType AS Dim
FROM islandRanges_cte
WHERE AccountType = 0 OR AccountType = 1
),
T4 AS
(
SELECT Account, Dim
FROM (
SELECT FlattenRow AS Account, StartRow, EndRow, Dim
FROM T3 I
CROSS APPLY (VALUES(StartRow),(EndRow)) newValues (FlattenRow)
) T
)
--SELECT * FROM T3 ORDER BY StartRow
--SELECT * FROM T4 ORDER BY Account
-- Final correct result but very very slow
SELECT D.Account, D.AccountType, I.Dim FROM T3 I
INNER JOIN #data D
ON D.Account BETWEEN I.StartRow AND I.EndRow
ORDER BY Account
EDIT with some time testing
SET NOCOUNT ON
IF OBJECT_ID('tempdb..#data') IS NULL
CREATE TABLE #times
(
RecId INTEGER IDENTITY(1,1),
Batch INTEGER,
Method NVARCHAR(255),
MethodDescription NVARCHAR(255),
RunTime INTEGER
)
IF OBJECT_ID('tempdb..#batch') IS NULL
CREATE TABLE #batch
(
Batch INTEGER IDENTITY(1,1),
Bit BIT
)
INSERT INTO #batch VALUES(0)
IF OBJECT_ID('tempdb..#data') IS NOT NULL
DROP TABLE #data
CREATE TABLE #data
(
Account INTEGER
)
CREATE NONCLUSTERED INDEX data_account_index ON #data (Account)
IF OBJECT_ID('tempdb..#islands') IS NOT NULL
DROP TABLE #islands
CREATE TABLE #islands
(
AccountFrom INTEGER ,
AccountTo INTEGER,
Dim INTEGER,
)
CREATE NONCLUSTERED INDEX islands_from_index ON #islands (AccountFrom, AccountTo, Dim)
BEGIN -- TEST DATA
INSERT INTO #data
SELECT TOP 100000 ROW_NUMBER() OVER(ORDER BY t1.number) AS N
FROM master..spt_values t1
CROSS JOIN master..spt_values t2
INSERT INTO #islands
SELECT MIN(Account) AS Start, MAX(Account), Grp
FROM (SELECT *, NTILE(10) OVER (ORDER BY Account) AS Grp FROM #data) T
GROUP BY Grp ORDER BY Start
END -- TEST DATA END
--SELECT * FROM #data
--SELECT * FROM #islands
--PRINT CONVERT(varchar(20),DATEDIFF(MS,@RunDate,GETDATE()))+' ms Sub Query'
DECLARE @RunDate datetime
SET @RunDate=GETDATE()
SELECT Account, (SELECT Dim From #islands WHERE Account BETWEEN AccountFrom AND AccountTo) AS Dim
FROM #data
INSERT INTO #times VALUES ((SELECT MAX(Batch) FROM #batch) ,'subquery','',DATEDIFF(MS,@RunDate,GETDATE()))
SET @RunDate=GETDATE()
SELECT D.Account, V.Dim
FROM #data D
CROSS APPLY
(
SELECT Dim From #islands V
WHERE D.Account BETWEEN V.AccountFrom AND V.AccountTo
) V
INSERT INTO #times VALUES ((SELECT MAX(Batch) FROM #batch) ,'crossapply','',DATEDIFF(MS,@RunDate,GETDATE()))
SET @RunDate=GETDATE()
SELECT D.Account, I.Dim
FROM #data D
JOIN #islands I
ON D.Account BETWEEN I.AccountFrom AND I.AccountTo
INSERT INTO #times VALUES ((SELECT MAX(Batch) FROM #batch) ,'join','',DATEDIFF(MS,@RunDate,GETDATE()))
SET @RunDate=GETDATE()
;WITH cte AS
(
SELECT Account, AccountFrom, AccountTo, Dim, 1 AS Counting
FROM #islands
CROSS APPLY (VALUES(AccountFrom),(AccountTo)) V (Account)
UNION ALL
SELECT Account + 1 ,AccountFrom, AccountTo, Dim, Counting + 1
FROM cte
WHERE (Account + 1) > AccountFrom AND (Account + 1) < AccountTo
)
SELECT Account, Dim, Counting FROM cte OPTION(MAXRECURSION 32767)
INSERT INTO #times VALUES ((SELECT MAX(Batch) FROM #batch) ,'recursivecte','',DATEDIFF(MS,@RunDate,GETDATE()))
You can select from the #times table to see the run times :)