Lead window function with duplicate dates

Question

Sample data here:
http://rextester.com/VNGMF66717

I have the following data:

  ID    Year    Date    
1111    2016    2016-02-28
1111    2016    2016-02-28
1111    2016    2016-03-31
1111    2016    2016-03-31
1111    2016    2016-03-31
1111    2016    2016-04-02
1111    2016    2016-05-31
1111    2016    2016-08-01
1111    2016    2016-12-11
1111    2017    2017-01-02
1111    2017    2017-01-02
1111    2017    2017-02-04
1111    2017    2017-02-04
1111    2017    2017-07-08
2222    2016    2016-02-11
2222    2016    2016-02-11
2222    2016    2016-03-28
2222    2016    2016-03-28
2222    2016    2016-03-28
2222    2016    2016-07-22
2222    2016    2016-12-31
2222    2017    2017-02-01
2222    2017    2017-02-14

I want to add a NextDate column (that resets within each ID and year) using a lead window-function directly in the SELECT statement, instead of a self join using a RANK windowfunction as i did in the provided example.

The NextDate column should be exactly like this

  ID    Year    Date        NextDate
1111    2016    2016-02-28  2016-03-31
1111    2016    2016-02-28  2016-03-31
1111    2016    2016-03-31  2016-04-02
1111    2016    2016-03-31  2016-04-02
1111    2016    2016-03-31  2016-04-02
1111    2016    2016-04-02  2016-05-31
1111    2016    2016-05-31  2016-08-01
1111    2016    2016-08-01  2016-12-11
1111    2016    2016-12-11  NULL
1111    2017    2017-01-02  2017-02-04
1111    2017    2017-01-02  2017-02-04
1111    2017    2017-02-04  2017-07-08
1111    2017    2017-02-04  2017-07-08
1111    2017    2017-07-08  NULL
2222    2016    2016-02-11  2016-03-28
2222    2016    2016-02-11  2016-03-28
2222    2016    2016-03-28  2016-07-22
2222    2016    2016-03-28  2016-07-22
2222    2016    2016-03-28  2016-07-22
2222    2016    2016-07-22  2016-12-31
2222    2016    2016-12-31  NULL
2222    2017    2017-02-01  2017-02-14
2222    2017    2017-02-14  NULL

Anyone know how to do this properly?

Answer 1

This may not be exactly what you meant, but by using two window functions with a nested select : MAX(LEAD), you get the desired result.

select
    id, yr, dt
    ,MAX(nx_dt) OVER (PARTITION BY id, dt) nxt_dt
FROM
    (
        select 
            *
            , LEAD(dt) OVER (PARTITION BY id, yr ORDER BY id, yr, dt) nx_dt
        from
        #testtable
    ) sub

/*
drop table #testtable

create table #testtable
(
    id int, yr int, dt date
)
insert into #testtable
(
    id, yr, dt
)
SELECT 1111,2016,'2016-02-28'   union all
SELECT 1111,2016,'2016-02-28'   union all
SELECT 1111,2016,'2016-03-31'   union all
SELECT 1111,2016,'2016-03-31'   union all
SELECT 1111,2016,'2016-03-31'   union all
SELECT 1111,2016,'2016-04-02'   union all
SELECT 1111,2016,'2016-05-31'   union all
SELECT 1111,2016,'2016-08-01'   union all
SELECT 1111,2016,'2016-12-11'   union all
SELECT 1111,2017,'2017-01-02'   union all
SELECT 1111,2017,'2017-01-02'   union all
SELECT 1111,2017,'2017-02-04'   union all
SELECT 1111,2017,'2017-02-04'   union all
SELECT 1111,2017,'2017-07-08'   union all
SELECT 2222,2016,'2016-02-11'   union all
SELECT 2222,2016,'2016-02-11'   union all
SELECT 2222,2016,'2016-03-28'   union all
SELECT 2222,2016,'2016-03-28'   union all
SELECT 2222,2016,'2016-03-28'   union all
SELECT 2222,2016,'2016-07-22'   union all
SELECT 2222,2016,'2016-12-31'   union all
SELECT 2222,2017,'2017-02-01'   union all
SELECT 2222,2017,'2017-02-14'
*/

Answer 2

I this you need to do this with outer apply or a subquery:

select cte.*, next_cte.date
from cte outer apply
     (select top 1 cte2.*
      from cte cte2
      where cte2.id = cte.id and cte2.year = cte.year and
            cte2.date > cte.date
      order by cte2.date desc
     ) next_cte;

Another alternative is to do the lead() on distinct values:

select cte.*, cte_next.next_date
from cte join
     (select id, year, date,
             lead(date) over (partition by id, year order by date) as next_date
      from cte
      group by id, year, date
     ) cte_next
     ON cte.id = cte_next.id and cte.year = cte_next.year and cte.date = cte_next.date;