Given Start Time , End Time & Secs. Obtain secs in a specific Hour - sql

Question

I'm using Vertica Database. I am trying to get the total secs in a particular hour from the following example session data. Any sample SQL code would be very helpful - Thanks
start time end time session length(secs)
2010-02-21 20:30:00 2010-02-21 23:30:00 10800
2010-02-21 21:30:00 2010-02-21 22:30:00 3600
2010-02-21 21:45:00 2010-02-21 21:59:00 840
2010-02-21 22:00:00 2010-02-21 22:20:00 1200
2010-02-21 22:30:00 2010-02-21 23:30:00 3600
Desired Output
hour secs_in_that_hour
20 1800
21 6240
22 8400
23 3600

Answer 1

You would need a table containing every hour, so that you could join it in. That join would be based on the hour being within start and end time and then you can extract the time using (min(hour end,end time) - max(hour start,start time)). Then group on the hour and sum.
Since I don't know vertica, I have no complete answer to this.

Answer 2

Vertica is based on PostgresSQL, especially language-wise. The best thing you could do is look up Postgres's Date Time functions and related tutorials. I haven't found an instance where a Postgres time function does not work in Vertica.
http://www.postgresql.org/docs/8.0/interactive/functions-datetime.html
There is probably a datediff type function you can use. (Sorry, I don't have to time to look it up.)

Answer 3

See Vertica function
TIMESERIES Clause
Provides gap-filling and interpolation (GFI) computation, an important component of time series analytics computation. See Using Time Series Analytics in the Programmer's Guide for details and examples.
Syntax
TIMESERIES slice_time AS 'length_and_time_unit_expression' OVER (
... [ window_partition_clause (page 147) [ , ... ] ]
... ORDER BY time_expression )
... [ ORDER BY table_column [ , ... ] ]

Answer 4

The simplest way is to just extract epoch (number of seconds) on the interval (the difference between timestamps).
As for the overlapping sums, you'll need to first break it out by hour. Some of these hours don't exist so you'll need to generate them using a TIMESERIES clause.
The idea will be to first create your hourly time slices, then theta join to find (and fan out) for all possible matches on this. This is basically looking for any and all overlaps of the time range. Luckily, this is pretty simple as it is just anywhere the start time is before the end of the slice and the end time is greater than the start of the slice.
Then you use greatest and least to find the actual time to start and stop within the slice, subtract them out, convert interval to seconds and done.
See below for the example.
with slices as (
select slice_time slice_time_start, slice_time + interval '1 hour' slice_time_end
from (
select min(start_time) time_range from mytest
union all
select max(end_time) from mytest
) range
timeseries slice_time as '1 HOUR' over (order by range.time_range)
)
select slice_time_start "hour", extract(epoch from sum( least(end_time, slice_time_end)-greatest(slice_time_start, start_time))) secs_in_that_hour
from slices join mytest on ( start_time < slice_time_end and end_time > slice_time_start)
group by 1
order by 1
There may be some edge cases or so additional filtering needed if your data isn't so clean.