Create membership timeseries from static table - sql

Question

I have a table (members) which shows users and any additions or deletions to a specific membership and on what date as below
User
Membership
Change
Date
1
100
A
01/01/1900
1
101
A
01/01/1990
1
100
D
01/01/2000
2
100
A
01/12/1990
2
101
A
01/01/1991
2
101
D
01/12/1991
2
100
D
01/01/1993
3
100
A
01/01/2000
I'm looking to find cases where one user is in multiple memberships in overlapping time periods, I'm using the below but it's pulling the wrong end dates when I run it
With membership As
(Select user, membership, date as Start_Date,
LEAD (date, 1, '31/12/9999') OVER (PARTITION BY membership ORDER BY
date) AS End_Date
FROM
(Select *, LAG(change,1,-1) OVER(PARTITION BY membership ORDER BY
date) AS Previous_change
From members) withprevious
Where change != previous_change),
MemberTimeSeries AS
(Select *
From membership
Where Start_Date IN
(Select a.Date
From members a
join membership b
on a.user = b.user
and a.membership = b.membership
Where a.change = 'A')),
DupeIDs AS
(Select Distinct a.user, a.membership, cast(a.start_date as date_ as
start_date, cast(a.end_date as date) as end_date
from membertimeseries a
join membertimeseries b
on a.user = b.user
and ((a.start_date >= b.start_date abd a.start_date < b.end_date)
Or (a.end_date > b.start_date and a.end_date <= b.end_date)
OR (b.start_date >= a.start_date abd b.start_date < a.end_date)
Or (b.end_date > a.start_date and b.end_date <= a.end_date)
I'm looking to see any user and membership combinations with respective start and end dates where it overlaps with another membership combination for the same user during any period of it's active membership. If there is no deletion in the table I want date to default to 31/12/9999
For example I want to see the below from my example
User
Membership
Start_Date
End_Date
1
100
01/01/1900
01/01/2000
1
101
01/01/1990
31/12/9999
2
100
01/12/1990
01/01/1993
2
101
01/01/1991
01/12/1991

Answer 1

After reviewing at your updated code, I decided it might be better to take a different approach.
You can combine the add and drop rows into ranges by first selecting the adds and then using an OUTER APPLY to look up the first following drop row with the same user and membership (if one exists).
If the above is wrapped up in a Common Table Expression (CTE), overlaps can then be identified by joining the ranges with themselves and limiting the join to those with the same user, overlapping date ranges, and excluding same row joins.
A standard test for overlapping date ranges is Start1 < End2 AND Start2 < End1. For exclusive end-dates (as appears to be the case here) < is used. If the ranges were defined with inclusive end dates <= would be used. Omitted end dates can be handled with an end-of-time default ISNULL(EndDate, '9999-12-31')'
To omit self matches, the following uses ROW_NUMBER() to assign distinct row IDs. You could alternately check for "not (all values equal)".
;WITH Ranges AS (
SELECT
A.[User], A.Membership, A.Date AS StartDate, D.Date AS EndDate,
ROW_NUMBER() OVER(ORDER BY (SELECT NULL)) AS RowId
FROM Memberships A
OUTER APPLY (
SELECT TOP 1 D.*
FROM Memberships D
WHERE D.Change = 'D'
AND D.[User] = A.[User]
AND D.Membership = A.Membership
AND D.Date > A.Date
ORDER BY D.Date
) D
WHERE A.Change = 'A'
)
SELECT DISTINCT R.[User], R.Membership, R.StartDate, R.EndDate
FROM Ranges R
JOIN Ranges R2
ON R2.[User] = R.[User]
AND R2.RowId <> R.RowId
AND R2.StartDate < ISNULL(R.EndDate, '2099-12-31')
AND R.StartDate < ISNULL(R2.EndDate, '2099-12-31')
ORDER BY R.[User], R.StartDate
In case of multiple overlaps, DISTINCT eliminates dups. An alternative would be to replace the JOIN with a WHERE EXISTS(...).
Results:
User
Membership
StartDate
EndDate
1
100
1900-01-01
2000-01-01
1
101
1990-01-01
null
2
100
1990-12-01
1993-01-01
2
101
1991-01-01
1991-12-01
5
100
2020-01-01
2020-02-01
5
102
2020-01-15
2020-02-15
5
101
2020-02-01
2020-03-01
See this db<>fiddle.