Postgres: grouping by date for a specific time zone - postgresql

Question

Let us say we have a Postgres table:
CREATE TABLE public.observations (
id integer NOT NULL,
utc_created_at timestamp without time zone NOT NULL
)
If we are interested in counting the observations made at a specific UTC date, we could:
SELECT COUNT(observations.id),
DATE(utc_created_at) AS date_utc_created_at
GROUP BY DATE(date_utc_created_at);
But now let us say that I would like to group observations by Pacific Standard Time dates. Is there a way without altering the schema ?
If there is no way and I have now instead:
CREATE TABLE public.observations (
id integer NOT NULL,
utc_created_at timestamp with time zone NOT NULL
)
can I make the described query ?
Would the given solution also work with EXTRACT to say group by Pacific Standard Time year/month ?

Answer 1

Since the timestamp is a timestamp without time zone in UTC, you need to:
Tell Postgres what time zone to interpret the time stamp in
Tell Postgres what time zone you want it in.
After having done that, you can reliably and reproducibly convert to a string in whatever format desired.
To see what I mean, try running:
select
utc_created_at as original_timestamp
--correct conversion to PST
, (utc_created_at at time zone'Z') at time zone 'PST' as timestamp_in_pst
, date( (utc_created_at at time zone'Z') at time zone 'PST' ) as date_from_pst
--will correctly convert it to a timestamp with timezone, but Postgres will consider your session timezone in conversions
, utc_created_at at time zone'Z' as timestamp_with_timezone_displayed_in_session_time_zone
, date(utc_created_at at time zone'Z') as date_depends_on_session
--incorrect, will interpret it as if the original timestamp is in PST
, utc_created_at at time zone'PST' as timestamp_considered_with_PST_incorrect
, date( utc_created_at at time zone 'PST') as incorrect_date
from public.observations
You know that, for example, if it's 00:01 on Jan 1 UTC, it should actually be December 31 of the previous year in PST, which you can see in this fiddle: http://sqlfiddle.com/#!17/e6d3c/10