CODEWITHSUNDEEP
sqlsql-serversql-server-2008t-sql
Legend
Legend

Reputation: 116790

How can I merge overlapping times within a window?

I have a table like this:

CREATE TABLE #TEMP (Name VARCHAR(255), START_TIME datetime, END_TIME datetime);

INSERT INTO #TEMP VALUES('John', '2012-01-01 09:00:01', '2012-01-01 12:00:02')
INSERT INTO #TEMP VALUES('John', '2012-01-01 09:40:01', '2012-01-01 11:00:02')
INSERT INTO #TEMP VALUES('John', '2012-01-02 05:00:01', '2012-01-02 05:15:02')
INSERT INTO #TEMP VALUES('David', '2012-01-04 05:00:01', '2012-01-04 05:15:02')
INSERT INTO #TEMP VALUES('David', '2012-01-05 07:01:01', '2012-01-05 15:15:02')

SELECT *
FROM #TEMP

DROP TABLE #TEMP

And the data is:

     Name   START_TIME                 END_TIME
1    John   2012-01-01 09:00:01.000    2012-01-01 12:00:02.000
2    John   2012-01-01 09:40:01.000    2012-01-01 11:00:02.000
3    John   2012-01-02 05:00:01.000    2012-01-02 05:15:02.000
4    David  2012-01-04 05:00:01.000    2012-01-04 05:15:02.000
5    David  2012-01-05 07:01:01.000    2012-01-05 08:15:02.000

Given a number say, 6, I am trying to do a GROUP BY on this table and merge times that overlap within a window of 6 hours before and after. Therefore, in the above table, rows 1 and 2 would be merged into a single row as they contain overlapping time range:

John 2012-01-01 06:00:01.000 2012-01-01 18:00:02.000

Rows 4 and 5 will be merged because subtracting 6 hours from 07:01:01.000 falls into the window of row 4.

Is there a good way of doing this on a large table containing about a million rows?

Upvotes: 4

Views: 288

Answers (1)

dani herrera
dani herrera

Reputation: 51645

I think that the best way to do this is creating a windows table and join #temp table with this new window table:

1) Step 1, preparing window table with all possible windows gaps (contains overlaping windows):

   SELECT 
      Name,
      dateadd(hour, -6, start_time) as start_w, 
      dateadd(hour, +6, start_time) as end_w
   into #possible_windows
   FROM #TEMP

2) Create an index on temp table to improve performance

   create index pw_idx on #possible_windows ( Name, start_w)

3) Eliminate overlaping windows in a self join select. This is the reason to create the index:

   select p2.* 
   into #myWindows
   from #possible_windows p1
   right outer join #possible_windows p2
     on p1.name = p2.name and 
        p2.start_w > p1.start_W and p2.start_w <= p1.end_w
   where p1.name is null

4) Join your table with #myWindows or use it directly.

WORKING:

SELECT 
  Name,
  dateadd(hour, -6, start_time) as start_w, 
  dateadd(hour, +6, start_time) as end_w,
  ROW_NUMBER() over(partition by Name order by Name, 
                    dateadd(hour, -6, start_time) ) as rn
into #possible_windows
FROM #TEMP 

create index pw_idx on #possible_windows ( Name, start_w)

select p2.* 
from #possible_windows p1
right outer join #possible_windows p2
  on p1.name = p2.name and 
     p2.start_w > p1.start_W and p2.start_w <= p1.end_w
where p1.name is null

RESULTS:

Name  start_w       end_w         rn 
----- ------------- ------------- -- 
David 2012-01-03 23:00:012012-01-04 11:00:011  
David 2012-01-05 01:01:012012-01-05 13:01:012  
John  2012-01-01 03:00:012012-01-01 15:00:011  
John  2012-01-01 23:00:012012-01-02 11:00:013

PE: Please, go back with your performance tests!

Upvotes: 2

Related Questions