Merging sql row groups - sql

Question

My applications splits a single row data into different row chunks which are always in sorted order of startdate.
Where rowpart = 0 is the start and rowpart=2 is always the end
rowpart=1 is the middle part,which can be repeated n no of times.
I need to return row in such form like startdate of rowpart=0 and enddate of rowpart=2(if present or else return enddate for rowpart )
Rowpart = 0 is the start of new row chunk
Rowpart = 2 is always the end of the chunk
Chunks can be spread across different dates.
+-----+-------------------------+-------------------------+----------+
| Id | startdate | enddate | rowpart |
+-----+-------------------------+-------------------------+----------+
| 100 | 2016-11-30 00:00:00.000 | 2016-11-30 01:00:00.000 | 0 |
| 100 | 2016-11-30 02:00:00.000 | 2016-11-30 03:00:00.000 | 1 |
| 100 | 2016-11-30 10:00:00.000 | 2016-12-01 00:00:00.000 | 0 |
| 100 | 2016-12-01 02:00:00.000 | 2016-12-01 02:30:00.000 | 1 |
| 100 | 2016-12-01 10:00:00.000 | 2016-12-01 10:30:00.000 | 1 |
| 100 | 2016-12-01 16:00:00.000 | 2016-12-01 16:30:00.000 | 2 |
| 101 | 2016-12-11 10:00:00.000 | 2016-12-11 10:30:00.000 | 0 |
+-----+-------------------------+-------------------------+----------+
So the above table should return:
+-----+-------------------------+-------------------------+
| Id | startdate | enddate |
+-----+-------------------------+-------------------------+
| 100 | 2016-11-30 00:00:00.000 | 2016-11-30 03:00:00.000 |
| 100 | 2016-12-30 10:00:00.000 | 2016-12-01 16:30:00.000 |
| 101 | 2016-12-11 10:00:00.000 | 2016-12-11 10:30:00.000 |
+-----+-------------------------+-------------------------+
Any help would be appreciated

Answer 1

This should work:
;WITH temp
AS
(
SELECT Id, startdate,enddate,rowpart,
--Find out First Record
CASE WHEN rowpart=0
THEN 1
ELSE 0
END AS is_first,
--Find out Last Record, Check if next rowpart is 0 or NULL:
CASE WHEN COALESCE(LEAD(rowpart) OVER (ORDER BY Id, startdate),0) = 0 --Check if next rowpart is 0 or NULL
THEN 1
ELSE 0
END AS is_last
FROM #tab
)
SELECT DISTINCT
Id,
CASE WHEN is_first = 1
THEN startdate
ELSE LAG(startdate) OVER (ORDER BY Id, startdate)
END AS startdate,
CASE WHEN is_last = 1
THEN enddate
ELSE LEAD(enddate) OVER (ORDER BY Id, startdate)
END AS enddate
FROM temp
WHERE is_first = 1 OR is_last = 1
ORDER BY Id, startdate
What i try to do here: Inside the CTE i mark the first and the last record for each sequence. If rowpart=0 --> it's the first record. If the next record is null or the the rowpart of the next record is 0 then we have the last record.
So when querying the CTE we can eliminate the "records in between". What remains are 1 or 2 records per sequence (the first and the last, in some cases this is the same record).
Then we replace startdate with the startdate of the first record of the sequence and enddate with the enddate of the last record of the sequence.
Eliminate duplicate values with DISTINCT and you get the desired output.
This is a dirty piece of SQL, but at least it works ;-)
If you didn't know SQL Servers LEAD and LAGfunction to access previous or following row values check this out: http://blog.sqlauthority.com/2013/09/22/sql-server-how-to-access-the-previous-row-and-next-row-value-in-select-statement/

Answer 2

Looks like a simple Group by is all you need
Try this
select Id,min(startdate),max(enddate)
From yourtable
Group by Id,cast(startdate as date)

Answer 3

Select
Id,
startdate,
enddate
from (
select Id,
startdate,
enddate,ROW_NUMBER()OVER(PARTITION BY CONVERT(DATE,startdate) ORDER BY startdate DESC )RN from #Table1
GROUP BY Id, startdate, enddate)T
WHERE T.RN = 1

Answer 4

Check This. using CTE and Joins :
with CTE as
(
select distinct *,
CASE WHEN COALESCE(LEAD(rowpart) OVER (ORDER BY Id, startdate),0) = 0
THEN 1
ELSE 0
end as RN2
from #table
)
select distinct bb.id,bb.startdate,aa.enddate from
(
select C2.*,ROW_NUMBER()OVER( ORDER BY id, startdate ) RN3
from CTE C2 where RN2= 1
) aa
join
(
select distinct *,
ROW_NUMBER()OVER( ORDER BY id, startdate ) RN3
from CTE c1 where rowpart=0
) bb on aa.RN3=bb.RN3
OutPut :

Answer 5

WITH
your_table_lead AS
(
SELECT
your_table.*,
LAG(rowpart, 1, 2) OVER (PARTITION BY id
ORDER BY startdate) AS last_rowpart,
LEAD(rowpart, 1, 0) OVER (PARTITION BY id
ORDER BY startdate) AS next_rowpart
FROM
your_table
),
filtered_sorted AS
(
SELECT
*,
ROW_NUMBER() OVER (PARTITION BY id
ORDER BY startdate) AS id_seq_num
FROM
your_table_lead
WHERE
rowpart IN (0, 2)
OR next_rowpart = 0
OR last_rowpart = 2
)
SELECT
id,
MIN(startdate),
MAX(enddate)
FROM
filtered_sorted
GROUP BY
id,
id_seq_num - CASE rowpart WHEN 2 THEN 1 ELSE rowpart END
I'm on my phone so apologies for typos, etc.
The first steps just try to filter out everything except the first and last entry of each 'group'. If the rowpart is 0 or 2, the row is included, or if the Next Row's rowpart is 0 then that row is included too (If there isn't a next row, use 0).
Then the 'trick' is to find a way to group up the 'pairs'.
If we have a sequence of 0,2,0,1,0,2,2,0 then what we want is to group them like a,a,b,b,c,c,d,e.
That can be done by turning all the 2's into 1's, the deducting the value from the ROW_NUMBER().
0,2,0,1,0,2,2,0 => 0,1,0,1,0,1,1,0
1,2,3,4,5,6,7,8 - 0,1,0,1,0,1,1,0 => 1,1,3,3,5,5,6,8
So, now we have 5 distinct 'groups', on which we can apply MIN() and MAX().