基于先前数据的SQL透视

Question

This is my first question and truthfully i haven't did any search before posting this question because i don't know what this topic would categorized as.

This is a SQL related question where i need to, i guess Pivot, the data to get a more friendlier data output.

I have a 4 columns table, respectively: ID, Username, Status, DateTime. Status is what determine the user action, like login and logout.

ID  Username  Status  DateTime  
1   A         0       2017-10-20 05:00:00  
2   A         0       2017-10-20 07:23:10  
3   B         0       2017-10-20 07:24:45  
4   A         1       2017-10-20 09:50:55  
5   A         0       2017-10-20 13:00:56  
6   B         1       2017-10-20 17:13:28  
7   B         0       2017-10-20 17:50:47  
8   A         1       2017-10-20 21:38:17  
9   A         0       2017-10-20 21:38:19  
10  B         1       2017-10-20 21:40:02

I need to filter Status0 and Status1 as login and logout, so any Status0 in between would be ignored

ID  Username  Status  DateTime  
1   A         0       2017-10-20 05:00:00  
2   A         0       2017-10-20 07:23:10  
4   A         1       2017-10-20 09:50:55 

would resulted in

Username  Status0              Status1
A         2017-10-20 05:00:00  2017-10-20 09:50:55

and the next 'A' will search for Status0's DateTime that is bigger than or equal the last Status1 (2017-10-20 09:50:55) and so on until end of data

The end data i need would be formatted as below:

Username  Status0              Status1
A         2017-10-20 05:00:00  2017-10-20 09:50:55
B         2017-10-20 07:24:45  2017-10-20 17:13:28 
A         2017-10-20 13:00:56  2017-10-20 21:38:17  
B         2017-10-20 17:50:47  2017-10-20 21:40:02
A         2017-10-20 21:38:19  null  

How can i achieve this result? My logic says that i need to recursively compare the Status1 and find the next Status0, but i'm not sure how to put it in sql query.

Any help is appreciated. Thanks before.

EDIT: I'm using SQL Server 2008.

Answer 1

Hmmm. Here is one way: look up the next status 1 (using apply ) and then aggregate:

select username,
       min(datetime) as status_0_datetime,
       status_1_datetime
from (select t.*, t2.datetime as status_1_datetime
      from t outer apply
           (select top 1 t2.*
            from t t2
            where t2.username = t.username and t2.status = 1 and
                  t2.datetime > t.datetime
            order by t2.datetime desc
           ) t2
      where t.status = 0
     ) t
group by username, status_1_datetime
order by username, min(datetime);

This will give you a row for every status 1 datetime.

Answer 2

Try the following approach. You remove repeating 0 in user status using LAG and then you use ROW_NUMBER to group appropriately your user's login/logout.

select tt.username,
    MAX(CASE WHEN status = 0 THEN tt.datetime END) status0,
    MAX(CASE WHEN status = 1 THEN tt.datetime END) status1
from
(
    select *,
        row_number() over (partition by username, status order by datetime) rn
    from
    (
      select *, lag(status) over (partition by username order by datetime) prevstatus
      from your_table
    ) t
    where not(t.status = 0 and t.prevstatus = 0)  or t.prevstatus is null -- this is to remove repeating 0
) tt
group by tt.username, tt.rn

demo

EDIT: ok, so the solution should be for SQL Server 2008 R2, therefore, there is no LAG . Then it is possible to solve it using NOT EXISTS , however, it is not very readable:

select tt.username,
    MAX(CASE WHEN status = 0 THEN tt.datetime END) status0,
    MAX(CASE WHEN status = 1 THEN tt.datetime END) status1
from
(
    select *,
        row_number() over (partition by username, status order by datetime) rn
    from
    (
      select *
      from your_table yt1
      where status = 1 or
            not exists(
              select 1
              from your_table yt2
              where yt2.status = 0 and 
                    yt2.username = yt1.username and
                    yt2.datetime = (
                      select max(yt3.datetime)
                      from your_table yt3
                      where yt3.datetime < yt1.datetime and
                            yt3.username = yt1.username
                    )
            )
    ) t
) tt
group by tt.username, tt.rn

demo

ok and the there is a last version that uses outer joins and GROUP BY instead of dependent subqueries. Therefore, it can be slightly more efficient in certain cases

select tt.username,
    MAX(CASE WHEN status = 0 THEN tt.datetime END) status0,
    MAX(CASE WHEN status = 1 THEN tt.datetime END) status1
from
(
    select *,
        row_number() over (partition by username, status order by datetime) rn
    from
    (
        select xt.*, yt.status joinstatus
        from your_table yt
        right join (
            select yt1.id, yt1.datetime, yt1.username, yt1.status, max(yt2.datetime) prevdatetime
            from your_table yt1
            left join your_table yt2 on yt1.datetime > yt2.datetime and 
                                   yt2.username = yt1.username and
                                   yt1.status = 0 
            group by yt1.id, yt1.datetime, yt1.username, yt1.status
        ) xt on yt.datetime = xt.prevdatetime and yt.username = xt.username and xt.status = yt.status
    ) t
    where t.joinstatus is null
) tt
group by tt.username, tt.rn