Presto SQL window aggregate looking back x hours/minutes/seconds

user_7373440 0
Viewed 2646

I want to do aggregate on presto sql by looking back x hours/minutes/seconds ago.

Data

id    |       timestamp       |    status
-------------------------------------------
A     |   2018-01-01 03:00:00 |     GOOD
A     |   2018-01-01 04:00:00 |     BAD
A     |   2018-01-01 05:00:00 |     GOOD
A     |   2018-01-01 09:00:00 |     BAD
A     |   2018-01-01 09:15:00 |     BAD
A     |   2018-01-01 13:00:00 |     GOOD
A     |   2018-01-01 14:00:00 |     GOOD
B     |   2018-02-01 09:00:00 |     GOOD
B     |   2018-02-01 10:00:00 |     BAD

Results: 

id    |       timestamp       |    status    | bad_status_count
----------------------------------------------------------------
A     |   2018-01-01 03:00:00 |     GOOD     |       0 
A     |   2018-01-01 04:00:00 |     BAD      |       1
A     |   2018-01-01 05:00:00 |     GOOD     |       1
A     |   2018-01-01 09:00:00 |     BAD      |       1
A     |   2018-01-01 09:15:00 |     BAD      |       2
A     |   2018-01-01 13:00:00 |     GOOD     |       0 
A     |   2018-01-01 14:00:00 |     GOOD     |       0
B     |   2018-02-01 09:00:00 |     GOOD     |       0
B     |   2018-02-01 10:00:00 |     BAD      |       1

I am counting bad status over the period of last 3 hours by business. How can I do that? I am trying something like this:

SELECT
  id,
  timestamp,
  status
  count(status) over(partition by id order by timestamp range between interval '3' hour and current_row) as bad_status_count
from table

Of course it doesnt work yet and I still have to filter out for bad status. I got this error: Error running query: line 7:1: Window frame start value type must be INTEGER or BIGINT(actual interval day to second)

presto window sql
1 Answers

I'm not 100% how to represent express this in PrestoDB, but the key idea is to convert the timestamps to hours:

select t.*,
       sum(case when status = 'Bad' then 1 else 0 end) over
           (partition by id
            order by hours
            range between -3 and current row
           ) as bad_status
from (select t.*,
             date_diff(hour, '2000-01-01', timestamp) as hours
      from t
     ) t;
user_11440350
Related