SQL Server 2012+ : Merge elements by time periods - sql

Question

I have been struggling for a while now with that problem and I need some help.
I have the following query :
CREATE TABLE Example(
Start NVARCHAR(8),
Endd NVARCHAR(8),
Col1 NVARCHAR(2),
Col2 NVARCHAR(2));
INSERT into Example (Start,Endd,Col1,Col2)
VALUES ('20130801','20140316','02','01'),
('20140317','20140319','04','02'),
('20140320','20140320','04','02'),
('20140321','20140421','02','Z8'),
('20140422','20140429','02','Z9'),
('20140430','20140902','04','02'),
('20140903','20150201','04','02'),
('20150202','20150223','04','02'),
('20150224','20150527','04','02'),
('20150528','99991231','04','02')
;
select MIN(Start)AS Start,MAX(Endd) AS Endd,Col1,Col2 from
(
SELECT top (100000000) Start, Endd,Col1, Col2,dense_rank() over(partition by Col1, Col2 order by Start,Endd) as rank
,LEAD (Col1) OVER (order by Start,Endd DESC) as l1
,LEAD (Col2) OVER (order by Start,Endd DESC) as l2
,LAG (Col1) OVER (order by Start,Endd DESC) as l11
,LAG (Col2) OVER (order by Start,Endd DESC) as l22
FROM Example sp
order by Start,Endd
)rq
GROUP BY Col1,Col2,case when (rq.l1=Col1 and rq.l2=Col2) or (rq.l11=Col1 and rq.l22=Col2) then 0 else rank end
order by Start,Endd;
My goal is to merge those data to have the following result:
However as you can see in the query result, when i have the same values for Col1 and Col2 on different time periods, the merge is not done correctly. It basically tries to merge them all in one, which create issues in the value for the new period.
Would someone be able to help me?

Answer 1

You were getting close in your query and you may have found a solution by now. This is a classic Islands and Gaps problem. I am giving the longer version with no use of LEAD AND LAG. You can replace perhaps 45% of the code below by using those windowing functions with perhaps a dense rank.
DECLARE #Example TABLE(
Start NVARCHAR(8),
Endd NVARCHAR(8),
Col1 NVARCHAR(2),
Col2 NVARCHAR(2));
INSERT into #Example (Start,Endd,Col1,Col2)
VALUES ('20130801','20140316','02','01'),
('20140317','20140319','04','02'),
('20140320','20140320','04','02'),
('20140321','20140421','02','Z8'),
('20140422','20140429','02','Z9'),
('20140430','20140902','04','02'),
('20140903','20150201','04','02'),
('20150202','20150223','04','02'),
('20150224','20150527','04','02'),
('20150528','99991231','04','02')
SELECT
TableID=MAX(TableID),Col1=MAX(Col1),Col2=MAX(Col2),Start=MIN(Start),Endd=MAX(Endd)
FROM
(
SELECT
TableID,Col1,Col2,Start,Endd,ChangeID=MAX(ChangeOnlyTableID)
FROM
(
SELECT
AllRecords.TableID,AllRecords.Col1,AllRecords.Col2,AllRecords.Start,AllRecords.Endd,ChangeOnlyTableID=ChangesOnly.TableID
FROM
(
SELECT * FROM
(
SELECT
This.Start,This.Endd,This.TableID,This.Col1,This.Col2,
Changed=CASE WHEN (Next.Col1=This.Col1 AND Next.Col2=This.Col2) THEN 0 ELSE 1 END
FROM
(
SELECT TableID=ROW_NUMBER() OVER(ORDER BY Start,Endd,Col1,Col2),Start,Endd,Col1,Col2 FROM #Example
)AS This
LEFT OUTER JOIN
(
SELECT TableID=ROW_NUMBER() OVER(ORDER BY Start,Endd,Col1,Col2),Start,Endd,Col1,Col2 FROM #Example
)
AS Next ON This.TableID=Next.TableID+1
)
AS ChangeMarkers
WHERE Changed=1
)
AS AllRecords
INNER JOIN
(
SELECT * FROM
(
SELECT
This.Start,This.Endd,This.TableID,This.Col1,This.Col2,
Changed=CASE WHEN (Next.Col1=This.Col1 AND Next.Col2=This.Col2) THEN 0 ELSE 1 END
FROM
(
SELECT TableID=ROW_NUMBER() OVER(ORDER BY Start,Endd,Col1,Col2),Start,Endd,Col1,Col2 FROM #Example
) AS This
LEFT OUTER JOIN
(
SELECT TableID=ROW_NUMBER() OVER(ORDER BY Start,Endd,Col1,Col2),Start,Endd,Col1,Col2 FROM #Example
) AS Next ON This.TableID=Next.TableID+1
)
AS ChangeMarkers
WHERE Changed=1
)
AS ChangesOnly ON ChangesOnly.Col1=AllRecords.Col1 AND ChangesOnly.Col2=AllRecords.Col2 AND ChangesOnly.TableID<=AllRecords.TableID
)AS JoinedResults
GROUP BY
TableID,Col1,Col2,Start,Endd
)
AS Final
GROUP BY
Col1,Col2,ChangeID
ORDER BY
MAX(TableID)
You may choose to shorten this somewhat with a few CTE's to produce a query such as:
;WITH TableWithIDs AS
(
SELECT TableID=ROW_NUMBER() OVER(ORDER BY Start,Endd,Col1,Col2),Start,Endd,Col1,Col2 FROM #Example
)
,ChangeMarkers AS
(
SELECT
This.Start,This.Endd,This.TableID,This.Col1,This.Col2,
Changed=CASE WHEN (Next.Col1=This.Col1 AND Next.Col2=This.Col2) THEN 0 ELSE 1 END
FROM
TableWithIDs AS This
LEFT OUTER JOIN TableWithIDs AS Next ON This.TableID=Next.TableID+1
)
,ChangesOnly AS
(
SELECT * FROM ChangeMarkers WHERE Changed=1
)
,
JoinedResults AS
(
SELECT
AllRecords.TableID,AllRecords.Col1,AllRecords.Col2,AllRecords.Start,AllRecords.Endd,ChangeOnlyTableID=ChangesOnly.TableID
FROM
ChangeMarkers AllRecords
INNER JOIN ChangesOnly ON ChangesOnly.Col1=AllRecords.Col1 AND ChangesOnly.Col2=AllRecords.Col2 AND ChangesOnly.TableID<=AllRecords.TableID
)
SELECT
TableID=MAX(TableID),Col1=MAX(Col1),Col2=MAX(Col2),Start=MIN(Start),Endd=MAX(Endd)
FROM
(
SELECT
TableID,Col1,Col2,Start,Endd,ChangeID=MAX(ChangeOnlyTableID)
FROM
JoinedResults
GROUP BY
TableID,Col1,Col2,Start,Endd
)
AS Final
GROUP BY
Col1,Col2,ChangeID
ORDER BY
MAX(TableID)
There are also some clever hacks that can be applied further using virtual keys however I went the most direct but more verbose route. You should be able to improve on this using a DENSE_RANK() with LEAD() OR LAG()