How to Count Consecutive Enteries for Users in SQL Server - sql

Question

Given the table with the following columns:
UserId int,
DateEntered DateTime
And the data:
1 | 2016-02-24
1 | 2016-02-23
1 | 2016-02-22
1 | 2016-02-20
2 | 2016-02-24
2 | 2016-02-14
3 | 2016-02-23
3 | 2016-02-22
3 | 2016-02-21
2 | 2016-01-30
2 | 2016-01-29
2 | 2016-01-28
2 | 2016-01-27
2 | 2016-01-26
2 | 2016-01-25
I would like to return the latest streak of entries for each user and actually for a specific user from today.
Case 1
Today = 2016-02-24
userid = 1
Return value = 3 // User missed Day 21 so streak is from 22-24
Case 2
Today = 2016-02-24
userid = 2
Return value = 1 // Even though user has a longer streak from 1/25 – 1/30, it is not his latest streak
Case 3
Today = 2016-02-24
userid = 3
Return value = 0 // User missed today. Therefore, he has no consecutive days counting today
Any ideas on how this can be done in T-SQL?
Update 1:
Based on the response, I've modified the example query given as follows: Yet, the value returned in the second column is always only either 1 or 0, even though the data shows that there are more consecutive days present.
select
a.UserId,
sum(case when dayseq = '2016-02-01' then 1 else 0 end)
from
(select
t.*,
dateadd(day, 1 - row_number() over (partition by UserId order by DateCreated), DateCreated) as dayseq
from
fa.User_Journal t) a
where
DateCreated <= '2016-02-01'
group by
a.UserId;
Update 2
The following query illustrates the problem further. The solution provided below almost resolves this.
In this query, I illustrate what "should" happen given the #EndDate values. By un-commenting the desires assignment to #EndDate, you can see that the query does not return the desired result according to the cases provided.
Any assistance would be greatly appreciated.
DECLARE #Temp TABLE
(
UserId nvarchar(128),
DateCreated Date
)
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-01-19');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-01-24');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-01-28');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-01-29');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-02-01');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-02-02');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-02-03');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-02-07');
INSERT INTO #TEMP (UserId, DateCreated) values ('uid123', '2016-01-19');
DECLARE #EndDate Date
SET #EndDate = '2016-02-03' -- Should return 5, as they are 5 consecutive days since #EndDate
--SET #EndDate = '2016-02-02' -- Should return 4, as they are 4 consecutive days since #EndDate
--SET #EndDate = '2016-02-19' -- Should return 1, as they are 4 consecutive days since #EndDate
--SET #EndDate = '2016-02-18' -- Should return 0, as they are 0 consecutive days since #EndDate
SELECT a.UserId,
SUM(CASE WHEN dayseq <= #EndDate then 1 else 0 end) -1 as FitStreak
from (select t.*,
dateadd(day,
1 - row_number() over (partition by UserId order by DateCreated),
DateCreated) as dayseq
from #Temp t
) a
where DateCreated <= #EndDate
group by a.UserId;

Answer 1

I think the following does what you want:
select t.UserId,
sum(case when dayseq = '2014-02-24' then 1 else 0 end)
from (select t.*,
dateadd(day,
row_number() over (partition by UserId order by DateEntered desc) - 1,
DateEntered) as dayseq
from t
) t
where DateEntered <= '2014-02-24'
group by t.UserId;
This subtracts a sequential number from each date, starting with 0. It then compares the result to the "current date" -- and voila, the value is the "current date" for the initial sequence.
The logic in the case statement can also be in the where. However, the results will then filter out users with no days.
Note two important assumptions:
There is no time component on the date
There are no duplicate dates
Both of these can be easily handled, but the resulting query is a bit more complicated.
EDIT:
With a time component, truncate the value:
select t.UserId,
sum(case when dayseq = '2014-02-24' then 1 else 0 end)
from (select t.*,
dateadd(day,
row_number() over (partition by UserId order by DateEntered desc) - 1,
cast(DateEntered as date)) as dayseq
from t
) t
where DateEntered < dateadd(day, 1, '2014-02-24')
group by t.UserId;
Here is a SQL Fiddle illustrating the fixed code.