Truncate timestamp to arbitrary intervals

Question

I want to plot my_values with some lower resolution on the time axis than I have in the database. I use something like the following SQL to get the average valu

Answer 1

You can convert to the number of seconds since epoch and go for integer division by 3*3600, 6*3600, etc.

Answer 2

Consider this demo to bring timestamps down to a resolution of 15 minutes and aggregate resulting dupes:

WITH tbl(id, ts) AS ( VALUES
    (1::int, '2012-10-04 00:00:00'::timestamp)
   ,(2, '2012-10-04 18:23:01')
   ,(3, '2012-10-04 18:30:00')
   ,(4, '2012-10-04 18:52:33')
   ,(5, '2012-10-04 18:55:01')
   ,(6, '2012-10-04 18:59:59')
   ,(7, '2012-10-05 11:01:01')
   )
SELECT to_timestamp((extract(epoch FROM ts)::bigint / 900)*900)::timestamp
                                                            AS lower_bound
     , to_timestamp(avg(extract(epoch FROM ts)))::timestamp AS avg_ts
     , count(*) AS ct
FROM   tbl
GROUP  BY 1
ORDER  BY 1;

Result:

     lower_bound     |       avg_ts        | ct
---------------------+---------------------+----
 2012-10-04 00:00:00 | 2012-10-04 00:00:00 |  1
 2012-10-04 18:15:00 | 2012-10-04 18:23:01 |  1
 2012-10-04 18:30:00 | 2012-10-04 18:30:00 |  1
 2012-10-04 18:45:00 | 2012-10-04 18:55:51 |  3
 2012-10-05 11:00:00 | 2012-10-05 11:01:01 |  1

The trick is to extract a unix epoch like @Michael already posted. Integer division lumps them together in buckets of the chosen resolution, because fractional digits are truncated.

I divide by 900, because 15 minutes = 900 seconds.

Multiply by the same number to get the resulting lower_bound. Convert the unix epoch back to a timestamp with to_timestamp().

This works great for intervals that can be represented without fractional digits in the decimal system. For even more versatility use the often overlooked function width_bucket() like I demonstrate in this recent, closely related answer. More explanation, links and an sqlfiddle demo over there.