Summing Records within a Moving Date Range, Date Distances - sql

Question

I have complex calculation requirement for a user logging system. I need to locate the most frequently active users based on their number of logins within a 180 day window. Once two login dates are 181 days apart, they do not count towards a total but could count towards a total when grouped with other dates.
For example here is Jim's login history:
Jim 2018-01-01
Jim 2018-04-01
Jim 2018-05-01
Jim 2018-06-01
Jim 2018-07-01
Jim 2018-08-01
Jim 2018-09-01
Jim 2018-12-01
Using 6 months, instead of 180 days, for simplicity, and only looking 6 months in one direction, Jim had the following totals:
Logins: 5 (2018-01-01 + 6 months)
Logins: 6 (2018-04-01 + 6 months)
Logins: 5 (2018-05-01 + 6 months)
Logins: 5 (2018-06-01 + 6 months)
Logins: 4 (2018-07-01 + 6 months)
Logins: 3 (2018-08-01 + 6 months)
Logins: 2 (2018-09-01 + 6 months)
Logins: 1 (2018-12-01 + 6 months)
So my system would report back 6 because it only wants the maximum total.
Other than brute force calculation, I'm lost on how to construct this system. Yes I can denormalize data to any degree, speed is most important.

Answer 1

Try this:
declare #tbl table(name char(3), dt date);
insert into #tbl values
('Jim', '2018-01-01'),
('Jim', '2018-04-01'),
('Jim', '2018-05-01'),
('Jim', '2018-06-01'),
('Jim', '2018-07-01'),
('Jim', '2018-08-01'),
('Jim', '2018-09-01'),
('Jim', '2018-12-01');
;with cte as (
select name, dt, DATEADD(day, 181, dt) upperDt from #tbl
), cte2 as (
select name,
(select COUNT(*) from cte where dt between c.dt and c.upperDt and name = c.name) cnt
from cte c
)
select name, MAX(cnt) [max]
from cte2
group by name

Answer 2

Try this, using a Common Table Expression to Calculate the EndDate Window and CROSS APPLY to calculate the total number of logins
DECLARE #t TABLE (UserName NVARCHAR(10), LoginDate DATETIME)
INSERT INTO #t
(UserName,LoginDate) VALUES
('Jim','2018-01-01'),
('Jim','2018-04-01'),
('Jim','2018-05-01'),
('Jim','2018-06-01'),
('Jim','2018-07-01'),
('Jim','2018-08-01'),
('Jim','2018-09-01'),
('Jim','2018-12-01')
; WITH CteDateRange
AS(
SELECT
T.UserName
,T.LoginDate
--,EndDateRange = DATEADD(DAY, 181, LoginDate)
,EndDateRange = DATEADD(MONTH, 6, LoginDate)
FROM #t T
)
SELECT
DR.UserName
,DR.LoginDate
,DR.EndDateRange
,T.Total
FROM CteDateRange DR
CROSS APPLY ( SELECT Total = COUNT(D.LoginDate)
FROM CteDateRange D
WHERE D.LoginDate >= DR.LoginDate
AND D.LoginDate <= DR.EndDateRange
AND D.UserName = DR.UserName
) T
Output
UserName LoginDate EndDateRange Total
Jim 2018-01-01 00:00:00.000 2018-07-01 00:00:00.000 5
Jim 2018-04-01 00:00:00.000 2018-10-01 00:00:00.000 6
Jim 2018-05-01 00:00:00.000 2018-11-01 00:00:00.000 5
Jim 2018-06-01 00:00:00.000 2018-12-01 00:00:00.000 5
Jim 2018-07-01 00:00:00.000 2019-01-01 00:00:00.000 4
Jim 2018-08-01 00:00:00.000 2019-02-01 00:00:00.000 3
Jim 2018-09-01 00:00:00.000 2019-03-01 00:00:00.000 2
Jim 2018-12-01 00:00:00.000 2019-06-01 00:00:00.000 1

Answer 3

One basic solution uses a join:
select l.*
from (select l.name, count(*) as cnt,
row_number() over (partition by name order by count(*) desc) as seqnum
from logins l join
logins l2
on l.name = l2.name and
l2.date >= l.date and l2.date < dateadd(day, 181, l.date)
group by l.name
) l
where seqnum = 1;
This might have acceptable performance with an index on logins(name, date).