T/SQL - Group/Multiply records - sql

Question

Source date:
CREATE TABLE #Temp (ID INT Identity(1,1) Primary Key, BeginDate datetime, EndDate datetime, GroupBy INT)
INSERT INTO #Temp
SELECT '2015-06-05 00:00:00.000','2015-06-12 00:00:00.000',7
UNION
SELECT '2015-06-05 00:00:00.000', '2015-06-08 00:00:00.000',7
UNION
SELECT '2015-10-22 00:00:00.000', '2015-10-31 00:00:00.000',7
SELECT *, DATEDIFF(DAY,BeginDate, EndDate) TotalDays FROM #Temp
DROP TABLE #Temp
ID BeginDate EndDate GroupBy TotalDays
1 6/5/15 0:00 6/8/15 0:00 7 3
2 6/5/15 0:00 6/12/15 0:00 7 7
3 10/22/15 0:00 10/31/15 0:00 7 9
Desired Output:
ID BeginDate EndDate GroupBy TotalDays GroupCnt GroupNum
1 6/5/15 0:00 6/8/15 0:00 7 3 1 1
2 6/5/15 0:00 6/12/15 0:00 7 7 1 1
3 10/22/15 0:00 10/29/15 0:00 7 9 2 1
3 10/29/15 0:00 10/31/15 0:00 7 9 2 2
Goal:
Group the records based on ID/BeginDate/EndDate.
Based on the GroupBy number (# of days) and TotalDays (days diff),
if the GroupBy => TotalDays, keep a single group record
else multiply the group records (1 record per GroupBy count) while staying within TotalDays limit.
Apologies if it's confusing but basically, in the above example, there should be one record for each group (ID/BeginDate/EndDate) for the record where days diff b/w Begin/End date = 7 or less (GroupBy).
If the days diff goes above 7 days, create another record (for every additional 7 days diff).
So since 1st two records have days diff of 7 days or less, there's only one record.
The 3rd record has days diff of 9 (7 + 2). Therefore, there should be 2 records (1st for the first 7 days and 2nd for the additional 2 days).
GroupCNT = how many records there're of the grouped records after applying the above records.
GroupNum is basically row number of the group.
GroupBy # can be different for each record. Dataset is huge so performance does matter.
One pattern I was able to figure out was related to the modulus b/w GroupBy and days diff.
When the GroupBy value is < days diff, modulus is always less than GroupBy. When the GroupBy value = days diff, modulus is always 0. And when the GroupBy value > days diff, modulus is always equals GroupBy. I'm not sure if/how to use that to group/multiply records to meet the requirement.
SELECT DISTINCT
ID
, BeginDate
, EndDate
, GroupBy
, DATEDIFF(DAY,BeginDate, EndDate) TotalDays
, CAST(GroupBy as decimal(18,6))%CAST(DATEDIFF(DAY,BeginDate, EndDate) AS decimal(18,6)) Modulus
, CASE WHEN DATEDIFF(DAY,BeginDate, EndDate) <= GroupBy THEN BeginDate END NewBeginDate
, CASE WHEN DATEDIFF(DAY,BeginDate, EndDate) <= GroupBy THEN EndDate END NewEndDate
FROM #Temp
Update:
Forgot to mention/include that the begin/enddate, when the records gets multiplied, will change accordingly. In other words, begin/end date will reflect the GroupBy - desired output shows what I mean more clearly in the 3rd and 4th record.
Also, GroupCnt/GroupNum are not as important to calculate as grouping/multiplying the records.

Answer 1

You could do something like this using a recursive CTE..
;WITH cte AS (
SELECT ID,
BeginDate,
EndDate,
GroupBy,
DATEDIFF(DAY, BeginDate, EndDate) AS TotalDays,
1 AS GroupNum
FROM #Temp
UNION ALL
SELECT ID,
BeginDate,
EndDate,
GroupBy,
TotalDays,
GroupNum + 1
FROM cte
WHERE GroupNum * GroupBy < TotalDays
)
SELECT ID,
BeginDate = CASE WHEN GroupNum = 1 THEN BeginDate
ELSE DATEADD(DAY, GroupBy * (GroupNum - 1), BeginDate)
END ,
EndDate = CASE WHEN TotalDays <= GroupBy THEN EndDate
WHEN DATEADD(DAY, GroupBy * GroupNum, BeginDate) > EndDate THEN EndDate
ELSE DATEADD(DAY, GroupBy * GroupNum, BeginDate)
END ,
GroupBy,
TotalDays,
COUNT(*) OVER (PARTITION BY ID) GroupCnt,
GroupNum
FROM cte
OPTION (MAXRECURSION 0)
the cte builds out a recordset like this.
ID BeginDate EndDate GroupBy TotalDays GroupNum
----------- ----------------------- ----------------------- ----------- ----------- -----------
1 2015-06-05 00:00:00.000 2015-06-08 00:00:00.000 7 3 1
2 2015-06-05 00:00:00.000 2015-06-12 00:00:00.000 7 7 1
3 2015-10-22 00:00:00.000 2015-10-31 00:00:00.000 7 9 1
3 2015-10-22 00:00:00.000 2015-10-31 00:00:00.000 7 9 2
then you just have to take this and use some case statements to determine what the begin and end date should be.
you should end up with
ID BeginDate EndDate GroupBy TotalDays GroupCnt GroupNum
----------- ----------------------- ----------------------- ----------- ----------- ----------- -----------
1 2015-06-05 00:00:00.000 2015-06-08 00:00:00.000 7 3 1 1
2 2015-06-05 00:00:00.000 2015-06-12 00:00:00.000 7 7 1 1
3 2015-10-22 00:00:00.000 2015-10-29 00:00:00.000 7 9 2 1
3 2015-10-29 00:00:00.000 2015-10-31 00:00:00.000 7 9 2 2
since you're using SQL 2012, you can also use the LAG and LEAD functions in your final query.
;WITH cte AS (
SELECT ID,
BeginDate,
EndDate,
GroupBy,
DATEDIFF(DAY, BeginDate, EndDate) AS TotalDays,
1 AS GroupNum
FROM #Temp
UNION ALL
SELECT ID,
BeginDate,
EndDate,
GroupBy,
TotalDays,
GroupNum + 1
FROM cte
WHERE GroupNum * GroupBy < TotalDays
)
SELECT ID,
BeginDate = COALESCE(LAG(BeginDate) OVER (PARTITION BY ID ORDER BY GroupNum) + GroupBy * (GroupNum - 1), BeginDate),
EndDate = COALESCE(LEAD(BeginDate) OVER (PARTITION BY ID ORDER BY GroupNum) + GroupBy * GroupNum, EndDate),
GroupBy,
TotalDays,
COUNT(*) OVER (PARTITION BY ID) GroupCnt,
GroupNum
FROM cte
OPTION (MAXRECURSION 0)

Answer 2

CREATE TABLE dim_number (id INT);
INSERT INTO dim_number VALUES ((0), (1), (2), (3)); -- Populate this to a large number
SELECT
#Temp.Id,
CASE WHEN dim_number.id = 0
THEN #Temp.BeginDate
ELSE DATEADD(DAY, dim_number.id * #Temp.GroupBy, #Temp.BeginDate)
END AS BeginDate,
CASE WHEN dim_number.id = parts.count
THEN #Temp.EndDate
ELSE DATEADD(DAY, (dim_number.id + 1) * #Temp.GroupBy, #Temp.BeginDate)
END AS EndDate,
#Temp.GroupBy AS GroupBy,
DATEDIFF(DAY, #Temp.BeginDate, #Temp.EndDate) AS TotalDays,
parts.count + 1 AS GroupCnt,
dim_number.id + 1 AS GroupNum
FROM
#Temp
CROSS APPLY
(SELECT DATEDIFF(DAY, #Temp.BeginDate, #Temp.EndDate) / #Temp.GroupBy AS count) AS parts
INNER JOIN
dim_number
ON dim_number.id >= 0
AND dim_number.id <= parts.count