I've spent much too long trying to figure out how TSQL pivoting works, so I'm posting this hoping someone that understands pivots can help me.
So here's a representative source result set:
Notice that there are only 3 rows, but that's still <=4.
| Code | Date |
|-------|------------|
| 12345 | 2018-01-01 |
| 67890 | NULL |
| 13579 | 2018-01-02 |
This is target result set I want from the source result set:
| Code_1 | Date_1 | Code_2 | Date_2 | Code_3 | Date_3 | Code_4 | Date_4 |
|--------|------------|--------|--------|--------|------------|--------|--------|
| 12345 | 2018-01-01 | 67890 | NULL | 13579 | 2018-01-02 | NULL | NULL |
Here's something that does literally what you're asking by joining 2 pivot queries together:
DECLARE #tbl TABLE (code INT, dt DATETIME)
INSERT INTO #tbl VALUES (12345, '1/1/2018')
INSERT INTO #tbl VALUES (67890, null)
INSERT INTO #tbl VALUES (13579, '1/1/2018')
SELECT SQ1.Code_1, SQ2.Code_1, SQ1.Code_2, SQ2.Code_2, SQ1.Code_3, SQ2.Code_3, SQ1.Code_4, SQ2.Code_4
FROM
(
SELECT [Code_1], [Code_2], [Code_3], [Code_4]
FROM
(
SELECT 'Code_' + CAST(ROW_NUMBER() OVER (ORDER BY (SELECT 1)) AS VARCHAR(3)) AS RowID, code FROM #tbl
) source
PIVOT
(
MAX(code)
FOR RowID IN ([Code_1], [Code_2], [Code_3], [Code_4])
)sq
)SQ1
CROSS APPLY
(
SELECT [Code_1], [Code_2], [Code_3], [Code_4]
FROM
(
SELECT 'Code_' + CAST(ROW_NUMBER() OVER (ORDER BY (SELECT 1)) AS VARCHAR(3)) AS RowID, dt FROM #tbl
) source
PIVOT
(
MAX(dt)
FOR RowID IN ([Code_1], [Code_2], [Code_3], [Code_4])
)sq
)SQ2
Outputs:
Code_1 Code_1 Code_2 Code_2 Code_3 Code_3 Code_4 Code_4
12345 2018-01-01 67890 NULL 13579 2018-01-01 NULL NULL
Assuming you want a max of 4 groups (8 columns).
You may also notice that in the Order By I use (Select null). Without a proper sequence key, there is no gtd of order.
Example
Select Code_1 = max(case when Col=1 then Code end)
,Date_1 = max(case when Col=1 then Date end)
,Code_2 = max(case when Col=2 then Code end)
,Date_2 = max(case when Col=2 then Date end)
,Code_3 = max(case when Col=3 then Code end)
,Date_3 = max(case when Col=3 then Date end)
,Code_4 = max(case when Col=4 then Code end)
,Date_4 = max(case when Col=4 then Date end)
From (
Select *
,Col = Row_Number() over (Order by (Select null))
From #YourTable
) A
Returns
Code_1 Date_1 Code_2 Date_2 Code_3 Date_3 Code_4 Date_4
12345 2018-01-01 67890 NULL 13579 2018-01-01 NULL NULL
Related
I have a table with 200.000 rows in a SQL Server 2014 database looking like this:
CREATE TABLE DateRanges
(
Contract VARCHAR(8),
Sector VARCHAR(8),
StartDate DATE,
EndDate DATE
);
INSERT INTO DateRanges (Contract, Sector, StartDate, Enddate)
SELECT '111', '999', '01-01-2014', '03-31-2014'
union
SELECT '111', '999', '04-01-2014', '06-30-2014'
union
SELECT '111', '999', '07-01-2014', '09-30-2014'
union
SELECT '111', '999', '10-01-2014', '12-31-2014'
union
SELECT '111', '888', '08-01-2014', '08-31-2014'
union
SELECT '111', '777', '08-15-2014', '08-31-2014'
union
SELECT '222', '999', '01-01-2014', '03-31-2014'
union
SELECT '222', '999', '04-01-2014', '06-30-2014'
union
SELECT '222', '999', '07-01-2014', '09-30-2014'
union
SELECT '222', '999', '10-01-2014', '12-31-2014'
union
SELECT '222', '666', '11-01-2014', '11-30-2014'
UNION
SELECT '222', '555', '11-15-2014', '11-30-2014';
As you can see there can be multiple overlaps for each contract and what I would like to have is the result like this
Contract Sector StartDate EndDate
---------------------------------------------
111 999 01-01-2014 07-31-2014
111 888 08-01-2014 08-14-2014
111 777 08-15-2014 08-31-2014
111 999 09-01-2014 12-31-2014
222 999 01-01-2014 10-31-2014
222 666 11-01-2014 11-14-2014
222 555 11-15-2014 11-30-2014
222 999 12-01-2014 12-31-2014
I can not figure out how this can be done and the examples i have seen on this site quite do not fit my problem.
This answer makes use of a few different techniques. The first is a recursive-cte that creates a table with every relevant cal_date which then gets cross apply'd with unique Contract values to get every combination of both values. The second is window-functions such as lag and row_number to determine a variety of things detailed in the comments below. Lastly, and probably most importantly, gaps-and-islands to determine when one Contract/Sector combination ends and the next begins.
Answer:
--determine range of dates
declare #bgn_dt date = (select min(StartDate) from DateRanges)
, #end_dt date = (select max(EndDate) from DateRanges)
--use a recursive CTE to create a record for each day / Contract
; with dates as
(
select #bgn_dt as cal_date
union all
select dateadd(d, 1, a.cal_date) as cal_date
from dates as a
where a.cal_date < #end_dt
)
select d.cal_date
, c.Contract
into #contract_dates
from dates as d
cross apply (select distinct Contract from DateRanges) as c
option (maxrecursion 0)
--Final Select
select f.Contract
, f.Sector
, min(f.cal_date) as StartDate
, max(f.cal_date) as EndDate
from (
--Use the sum-over to obtain the Island Numbers
select dr.Contract
, dr.Sector
, dr.cal_date
, sum(dr.IslandBegin) over (partition by dr.Contract order by dr.cal_date asc) as IslandNbr
from (
--Determine if the record is the start of a new Island
select a.Contract
, a.Sector
, a.cal_date
, case when lag(a.Sector, 1, NULL) over (partition by a.Contract order by a.cal_date asc) = a.Sector then 0 else 1 end as IslandBegin
from (
--Determine which Contract/Date combinations are valid, and rank the Sectors that are in effect
select cd.cal_date
, dr.Contract
, dr.Sector
, dr.EndDate
, row_number() over (partition by dr.Contract, cd.cal_date order by dr.StartDate desc) as ConractSectorRnk
from #contract_dates as cd
left join DateRanges as dr on cd.Contract = dr.Contract
and cd.cal_date between dr.StartDate and dr.EndDate
) as a
where a.ConractSectorRnk = 1
and a.Contract is not null
) as dr
) as f
group by f.Contract
, f.Sector
, f.IslandNbr
order by f.Contract asc
, min(f.cal_date) asc
Output:
+----------+--------+------------+------------+
| Contract | Sector | StartDate | EndDate |
+----------+--------+------------+------------+
| 111 | 999 | 2014-01-01 | 2014-07-31 |
| 111 | 888 | 2014-08-01 | 2014-08-14 |
| 111 | 777 | 2014-08-15 | 2014-08-31 |
| 111 | 999 | 2014-09-01 | 2014-12-31 |
| 222 | 999 | 2014-01-01 | 2014-10-31 |
| 222 | 666 | 2014-11-01 | 2014-11-14 |
| 222 | 555 | 2014-11-15 | 2014-11-30 |
| 222 | 999 | 2014-12-01 | 2014-12-31 |
+----------+--------+------------+------------+
I currently have the following table:
+-----+-----------------------------+------------------------------+
| ID | StartDate | EndDate |
+-----+-----------------------------+------------------------------|
| 1 | 2017-07-24 08:00:00.000 | 2017-07-29 08:00:00.000 |
| 2 | 2017-07-25 08:00:00.000 | 2017-07-28 08:00:00.000 |
| 3 | 2017-07-25 08:00:00.000 | 2017-07-26 08:00:00.000 |
+-----+-----------------------------+------------------------------+
I would like to know the count of the ID's that were not Closed on each date.
So for example, I wan't to know the count of open ID's on 2017-07-26 00:00:00.000. This would be all 3 in this case.
Another example: I wan't to know the count of open ID's on 2017-07-29 00:00:00.000. Which would be result to 1. Only ID=1 is Not yet closed at that date.
I have tried using another solution here on StackOverflow, but I can't quite figure why it is giving me false results.
declare #dt date, #dtEnd date
set #dt = getdate()-7
set #dtEnd = dateadd(day, 100, #dt);
WITH CTEt1 (SupportCallID, StartDate, EndDate, Onhold)
as
(SELECT SupportCallID
,OpenDate
,MAX(CASE WHEN StatusID IN('19381771-8E81-40C5-8E36-62A7DB0A2A99', '95C7A5FB-2389-4D14-9DAE-A08BFCC3B09A', 'D5429790-3B43-4462-9E1E-2466EA29AC74') then CONVERT(DATE, LastChangeDate) end) EndDate
,OnHold
FROM [ClienteleITSM_Prod_Application].[dbo].[SupportCall]
group by SupportCallID, OpenDate, OnHold
)
SELECT dates.myDate,
(SELECT COUNT(*)
FROM CTEt1
WHERE myDate BETWEEN StartDate and EndDate
)
FROM
(select dateadd(day, number, #dt) mydate
from
(select distinct number from master.dbo.spt_values
where name is null
) n
where dateadd(day, number, #dt) < #dtEnd) dates
If you use a cte to create a table of dates that span the range of dates in your source table, you can easily left join from that to your source table and count up the rows returned:
declare #t table(ID int,StartDate datetime,EndDate datetime);
insert into #t values (1,'2017-07-24 08:00:00.000','2017-07-29 08:00:00.000'),(2,'2017-07-25 08:00:00.000','2017-07-28 08:00:00.000'),(3,'2017-07-25 08:00:00.000','2017-07-26 08:00:00.000');
declare #StartDate datetime = (select min(StartDate) from #t);
declare #EndDate datetime = (select max(EndDate) from #t);
-- Table with 10 rows in to be joined together to create a large tally table (10 * 10 * 10 * etc)
with t(t) as (select t from (values(1),(1),(1),(1),(1),(1),(1),(1),(1),(1))t(t))
-- Add the row_number of the tally table to your start date to generate all dates within your data range
,d(d) as (select top(datediff(d,#StartDate,#EndDate)+1) dateadd(d,row_number() over (order by (select null))-1,#StartDate) from t t1,t t2,t t3)
select d.d
,count(t.ID) as OpenIDs
from d
left join #t as t
on(d.d between cast(t.StartDate as date) and t.EndDate)
group by d.d
order by d.d;
Output:
+-------------------------+---------+
| d | OpenIDs |
+-------------------------+---------+
| 2017-07-24 08:00:00.000 | 1 |
| 2017-07-25 08:00:00.000 | 3 |
| 2017-07-26 08:00:00.000 | 3 |
| 2017-07-27 08:00:00.000 | 2 |
| 2017-07-28 08:00:00.000 | 2 |
| 2017-07-29 08:00:00.000 | 1 |
+-------------------------+---------+
I have a source table that looks like this:
+--------------+----------+------+------------+-----------+
| vehicleindex | parentid | year | make | model |
+--------------+----------+------+------------+-----------+
| 1 | 1 | 2007 | TOYOTA | SIENNA LE |
| 2 | 1 | 2005 | VOLKSWAGEN | JETTA GLS |
+--------------+----------+------+------------+-----------+
I'd like to select from this table such that the output looks like this:
+-------+--------+-----------+-------+------------+-----------+
| year1 | make1 | model1 | year2 | make2 | model2 |
+-------+--------+-----------+-------+------------+-----------+
| 2007 | TOYOTA | SIELLA LE | 2005 | VOLKSWAGEN | JETTA GLS |
+-------+--------+-----------+-------+------------+-----------+
How can I accomplish this on a SQL Server database with a pivot? There will always be either 1 or 2 vehicles in the source table. In the case where there's 1 vehicle, I would expect Year2, Make2 and Model2 to be NULL.
Similar to SQLZim's answer. Only difference is that the Window function Row_Number() is used just in case vehicleindex is not a consistent 1 and 2.
Select year1 = max(case when RN=1 then [year] end)
,make1 = max(case when RN=1 then make end)
,model1 = max(case when RN=1 then model end)
,year2 = max(case when RN=2 then [year] end)
,make2 = max(case when RN=2 then make end)
,model2 = max(case when RN=2 then model end)
From (
Select *
,RN = Row_Number() over (Partition By parentid Order By vehicleindex)
From YourTable
) A
Group By parentid
EDIT: Option 2 - Use PIVOT
Select *
From (
Select parentid
,item = concat(B.item,Dense_Rank() over (Partition By parentid Order By vehicleindex))
,value
From YourTable
Cross Apply ( values ('year' ,cast(Year as varchar(100)))
,('make' ,make)
,('model',model)
) B (item,value)
) A
Pivot (max(value) For [item] in ([year1],[make1],[model1],[year2],[make2],[model2]) ) p
Using conditional aggregation:
select
parentid
, year1 = max(case when vehicleindex=1 then [year] end)
, make1 = max(case when vehicleindex=1 then make end)
, model1 = max(case when vehicleindex=1 then model end)
, year2 = max(case when vehicleindex=2 then [year] end)
, make2 = max(case when vehicleindex=2 then make end)
, model2 = max(case when vehicleindex=2 then model end)
from t
group by parentid
returns:
+----------+-------+------------+-----------+-------+------------+-----------+
| parentid | year1 | make1 | model1 | year2 | make2 | model2 |
+----------+-------+------------+-----------+-------+------------+-----------+
| 1 | 2007 | TOYOTA | SIENNA LE | 2005 | VOLKSWAGEN | JETTA GLS |
| 2 | 2018 | TESLA | MODEL 3 | NULL | NULL | NULL |
+----------+-------+------------+-----------+-------+------------+-----------+
rextestder demo: http://rextester.com/ZTGXU25389
using this test data:
create table t (
vehicleindex int
, parentid int
, [year] int
, make varchar(32)
, model varchar(32)
);
insert into t values
(1,1,2007,'TOYOTA ','SIENNA LE')
, (2,1,2005,'VOLKSWAGEN','JETTA GLS')
, (1,2,2018,'TESLA','MODEL 3')
Unfortunately if you need to pivot 3 columns you need to create 3 separate pivots, then merge them in the end:
SELECT MAX(Year1) as 'Year1', MAX(Make1) as Make1, MAX(Model1) as 'Model1',
MAX(Year2) as 'Year2', MAX(Make2) as Make2, MAX(Model2) as 'Model2'
FROM
(
SELECT parentid,
[1] as 'Year1', NULL as Make1, NULL as 'model1',
[2] as 'Year2', NULL as Make2, NULL as 'model2'
FROM (
SELECT
parentid, vehicleindex, model, make , [year]
FROM #temp
) as s
PIVOT
(
max([year])
FOR vehicleindex IN ([1],[2])
) AS yearPvt
UNION
SELECT parentid,
NULL as 'Year1', [1] as Make1, NULL as 'model1',
NULL as 'Year2', [2] as Make2, NULL as 'model2'
FROM
(
SELECT
parentid, vehicleindex, model, make , [year]
FROM #temp
) as s
parentid, vehicleindex, model, make , [year]
FROM #temp
) as s
PIVOT
(
MAX([make])
FOR vehicleindex IN ([1],[2])
)AS makePvt
UNION
SELECT parentid,
NULL as 'Year1', NULL as Make1, [1] as 'model1',
NULL as 'Year2', NULL as Make2, [2] as 'model2'
FROM
(
SELECT
parentid, vehicleindex, model, make , [year]
FROM #temp
) as s
PIVOT
(
max([model])
FOR vehicleindex IN ([1],[2])
) AS modelPvt
) allPivots
Group BY parentid
Due to company policies I cannot give the actual query I am working with but heres the breakdown and general idea. We have an attendance register that records for each day if an employee was at work or not and where the employee works at. I am trying to make a summary of this to say between this and that date the employee worked 5 shifts. The problem I am sitting with is that one particular employee worked in workplace A for 2 days and was then transferred to workplace B. After a few days at workplace B the employee was then transferred back to workplace A.
My results to my attempt has showed that the employee begun working at workplace A from 1-Jan and ended at 10-Jan with only 2 working shifts. I have a group by on the working place and the begin and end dates are a min and max selection.
SELECT att.Employee, att.Workplace, dte.BeginDate, dte.EndDate, shf.WorkShift FROM
(SELECT * FROM Attendance WHERE WorkDate BETWEEN '1-Jan' AND '30-Jan') att
CROSS APPLY (SELECT COUNT(Shift) WorkShift FROM Attendance WHERE WorkDate BETWEEN '1-Jan' AND '30-Jan' AND Employee = att.Employee AND WorkPlace = att.WorkPlace AND Shift = 'Worked') shf
CROSS APPLY (SELECT MAX(WorkDate) BeginDate, MIN(WorkDate) EndDate FROM Attendance WHERE WorkDate BETWEEN '1-Jan' AND '30-Jan' AND Employee = att.Employee AND WorkPlace = att.WorkPlace) dte
So this employees records should appear like this (I am sorry for the very bad grid, I don't know how to make it look pretty, you are more than welcome to edit it to look better)
| Name | Workplace | beginDate | endDate | WorkShift |
| Jane | WorkPlaceA | 1-Jan | 2-Jan | 2 |
| Jane | WorkPlaceB | 3-Jan | 8-Jan | 5 |
| Jane | WorkPlaceA | 9-Jan | 10-Jan | 2 |
The attendance table looks something like this
| Name | Workplace | Date | Shift |
| Jane | WorkplaceA | 1-Jan | Worked |
| Jane | WorkplaceA | 2-Jan | Worked |
| Jane | WorkplaceB | 3-Jan | Worked |
| Jane | WorkplaceB | 4-Jan | Worked |
| Jane | WorkplaceB | 5-Jan | Worked |
| Jane | WorkplaceA | 6-Jan | Absent |
| Jane | WorkplaceA | 7-Jan | Absent |
| Jane | WorkplaceA | 8-Jan | Worked |
| Jane | WorkplaceB | 9-Jan | Worked |
| Jane | WorkplaceB | 10-Jan | Worked |
I believe you can accomplish this using CTE's. Here is a sample working code that shows your expected values.
;WITH CTE1 AS (
SELECT Employee, WorkPlace, TransactionDate,
ROW_NUMBER() OVER(PARTITION BY WorkPlace ORDER BY TransactionDate) AS WP,
ROW_NUMBER() OVER(ORDER BY TransactionDate) AS RN FROM Attendance WHERE Shift = 'Worked'),
CTE2 AS (SELECT Employee, WorkPlace, TransactionDate, WP, RN, WP-RN AS GB FROM CTE1),
CTE3 AS (SELECT Employee, WorkPlace, MIN(TransactionDate) AS TransactionDate, COUNT(1) AS Shifts FROM CTE2 GROUP BY Employee, WorkPlace, GB)
SELECT Employee, WorkPlace, TransactionDate AS [Start Date], DATEADD(DAY,Shifts - 1,TransactionDate) AS [End Date], Shifts FROM CTE3 ORDER BY TransactionDate ASC
I think your given output is wrong.
I think the way you are populating table is wrong.
Check my query,it can be further optmize,it do not count absent days
declare #t table(Name varchar(100),Workplace varchar(100), AttnDate date ,Shifts varchar(100))
insert into #t values
('Jane','WorkplaceA',' 1-Jan-16','Worked')
,('Jane','WorkplaceA',' 2-Jan-16','Worked')
,('Jane','WorkplaceB',' 3-Jan-16','Worked')
,('Jane','WorkplaceB',' 4-Jan-16','Worked')
,('Jane','WorkplaceB',' 5-Jan-16','Worked')
,('Jane','WorkplaceA',' 6-Jan-16','Absent')
,('Jane','WorkplaceA',' 7-Jan-16','Absent')
,('Jane','WorkplaceA',' 8-Jan-16','Worked')
,('Jane','WorkplaceB',' 9-Jan-16','Worked')
,('Jane','WorkplaceB','10-Jan-16','Worked')
DECLARE #Name VARCHAR(100) = 'Jane'
DECLARE #FromDate DATE = '01-Jan-16'
DECLARE #ToDate DATE = '31-Jan-16';
WITH CTE
AS (
SELECT *
,row_number() OVER (
ORDER BY attndate
) rn
FROM #t
WHERE NAME = #Name
AND (
AttnDate BETWEEN #FromDate
AND #ToDate
)
)
,CTE1
AS (
SELECT A.NAME
,A.workplace
,A.AttnDate
,Shifts
,rn
,1 RN1
FROM cte A
WHERE rn = 1
UNION ALL
SELECT a.NAME
,a.workplace
,a.AttnDate
,a.Shifts
,CASE
WHEN a.workplace = b.workplace
THEN b.rn
ELSE b.rn + 1
END rn
,RN1 + 1
FROM CTE A
INNER JOIN CTE1 b ON a.attndate > b.attndate
WHERE a.rn = RN1 + 1
)
,CTE2
AS (
SELECT NAME
,Workplace
,AttnDate beginDate
,(
SELECT max(AttnDate)
FROM CTE1 b
WHERE b.rn = a.rn
) endDate
,(
SELECT count(*)
FROM CTE1 b
WHERE b.rn = a.rn
AND Shifts = 'Worked'
) WorkShift
,rn
,ROW_NUMBER() OVER (
PARTITION BY rn ORDER BY rn
) rn3
FROM cte1 a
)
SELECT NAME
,workplace
,beginDate
,endDate
,WorkShift
FROM cte2
WHERE rn3 = 1
Here's the data:
[ TABLE_1 ]
id | prod1 | date1 | prod2 | date2 | prod3 | date3 |
---|--------|--------|--------|--------|--------|-------|
1 | null | null | null | null | null | null |
2 | null | null | null | null | null | null |
3 | null | null | null | null | null | null |
[ TABLE_2 ]
id | date | product |
-----|-------------|-----------|
1 | 20140101 | X |
1 | 20140102 | Y |
1 | 20140103 | Z |
2 | 20141201 | data |
2 | 20141201 | Y |
2 | 20141201 | Z |
3 | 20150101 | data2 |
3 | 20150101 | data3 |
3 | 20160101 | X |
Both tables have other columns not listed here.
date is formatted: yyyymmdd and datatype is int.
[ TABLE_2 ] doesn't have empty rows, just tried to make sample above more readable.
Here's the Goal:
I need to update [ TABLE_1 ] prod1,date1,prod2,date2,prod3,date3
with product collected from [ TABLE_2 ] with corresponding date values.
Data must be sorted so that "latest" product becomes prod1,
2nd latest product will be prod2 and 3rd is prod3.
Latest product = biggest date (int).
If dates are equal, order doesn't matter. (see id=2 and id=3).
Updated [ TABLE_1 ] should be:
id | prod1 | date1 | prod2 | date2 | prod3 | date3 |
---|--------|----------|--------|----------|--------|----------|
1 | Z | 20140103 | Y | 20140102 | X | 20140101 |
2 | data | 20141201 | Y | 20141201 | Z | 20141201 |
3 | X | 20160101 | data2 | 20150101 | data3 | 20150101 |
Ultimate goal is to get the following :
[ TABLE_3 ]
id | order1 | order2 | order3 | + Columns from [ TABLE_1 ]
---|--------------------|----------------------|------------|--------------------------
1 | 20140103:Z | 20140102:Y | 20140103:Z |
2 | 20141201:data:Y:Z | NULL | NULL |
3 | 20160101:X | 20150101:data2:data3 | NULL |
I have to admit this exceeds my knowledge and I haven't tried anything.
Should I do it with JOIN or SELECT subquery?
Should I try to make it in one SQL -clause or perhaps in 3 steps,
each prod&date -pair at the time ?
What about creating [ TABLE_3 ] ?
It has to have columns from [ TABLE_1 ].
Is it easiest to create it from [ TABLE_2 ] -data or Updated [ TABLE_1 ] ?
Any help would be highly appreciated.
Thanks in advance.
I'll post some of my own shots on comments.
After looking into it (after my comment), a stored procedure would be best, that you can call to view the data as a pivot, and do away with TABLE_1. Obviously if you need to make this dynamic, you'll need to look into dynamic pivots, it's a bit of a hack with CTEs:
CREATE PROCEDURE DBO.VIEW_AS_PIVOTED_DATA
AS
;WITH CTE AS (
SELECT ID, [DATE], 'DATE' + CAST(ROW_NUMBER() OVER(PARTITION BY ID ORDER BY [DATE] DESC) AS VARCHAR) AS [RN]
FROM TABLE_2)
, CTE2 AS (
SELECT ID, PRODUCT, 'PROD' + CAST(ROW_NUMBER() OVER(PARTITION BY ID ORDER BY [DATE] DESC) AS VARCHAR) AS [RN]
FROM TABLE_2)
, CTE3 AS (
SELECT ID, [DATE1], [DATE2], [DATE3]
FROM CTE
PIVOT(MAX([DATE]) FOR RN IN ([DATE1],[DATE2],[DATE3])) PIV)
, CTE4 AS (
SELECT ID, [PROD1], [PROD2], [PROD3]
FROM CTE2
PIVOT(MAX(PRODUCT) FOR RN IN ([PROD1],[PROD2],[PROD3])) PIV)
SELECT A.ID, [PROD1], [DATE1], [PROD2], [DATE2], [PROD3], [DATE3]
FROM CTE3 AS A
JOIN CTE4 AS B
ON A.ID=B.ID
Construction:
WITH ranked AS (
SELECT [id]
,[date]
,[product]
,row_number() over (partition by id order by date desc) rn
FROM [sistemy].[dbo].[TABLE_2]
)
SELECT id, [prod1],[date1],[prod2],[date2],[prod3],[date3]
FROM
(
SELECT id, type+cast(rn as varchar(1)) col, value
FROM ranked
CROSS APPLY
(
SELECT 'date', CAST([date] AS varchar(8))
UNION ALL
SELECT 'prod', product
) ca(type, value)
) unpivoted
PIVOT
(
max(value)
for col IN ([prod1],[date1],[prod2],[date2],[prod3],[date3])
) pivoted
You need to take a few steps to achive the aim.
Rank your products by date:
SELECT [id]
,[date]
,[product]
,row_number() over (partition by id order by date desc) rn
FROM [sistemy].[dbo].[TABLE_2]
Unpivot your date and product columns into one column. You can use UNPIVOT OR CROSS APPLY statements. I prefer CROSS APPLY
SELECT id, type+cast(rn as varchar(1)) col, value
FROM ranked
CROSS APPLY
(
SELECT 'date', CAST([date] AS varchar(8))
UNION ALL
SELECT 'prod', product
) ca(type, value)
or the same result using UNPIVOT
SELECT id, type+cast(rn as varchar(1)) col, value
FROM (
SELECT [id],
rn,
CAST([date] AS varchar(500)) date,
CAST([product] AS varchar(500)) prod
FROM ranked) t
UNPIVOT
(
value FOR type IN (date, product)
) unpvt
and at last you use PIVOTE and get a result.