Postgres: query schedule given day, time_of_day, and time zone name - postgresql

Question

I'm writing a "send me messages at this time" app. I'm storing recurrence information in this manner:
Schedules
----------
days_of_week: [3, 4, 5]
hours_of_day: [8, 13, 22]
time_zone: "Pacific Time (US & Canada)"
Works fine in displaying, but I need to write a frequent cron job that grabs all schedules for "right now (utc)". So, if the the cron job is running at 09:00 UTC Monday, I need to grab all schedules where
Monday is in days_of_week (where days_of_week #> ARRAY[1])
hours_of_day is at 09:00 UTC. This is given hours_of_day is stored as an array of integers, but we are also storing the user's time_zone.
So the user may say: "deliver me a message at 9am Monday" (which we store as [9]), but that means 9am in their time zone.
Questions:
Any way to query all schedules given these parameters?
If not, is there a better way to structure the data to ensure easier querying through Postgres? The schema is flexible.
Thanks in advance!

Answer 1

Postgres has superb facilities for working with timezones, and I've written something very similar to what you're asking about here using the AT TIME ZONE construct. In addition to your fields, I use a last_scheduled_at to flag when a schedule was last "executed"--i.e., when the last successful cron job ran for that schedule to avoid double-scheduling, and a deleted_at for logical deletion of schedules.
My schema for schedules was similar, except that I had only a single hour. I stored days in an array, like you, and the timezone as text. The fields in my schedules table are dows, hour, and timezone.
This was the query:
SELECT
s.*
FROM
schedules s
WHERE
ARRAY[extract(dow from timestamptz (now() at time zone timezone))] && dows
AND hour = extract(hour from timestamptz (now() at time zone timezone))
AND (s.last_scheduled_at IS NULL
OR s.last_scheduled_at < (now() - interval '12 hours'))
AND s.deleted_at IS NULL
LIMIT
1000
I use && (overlaps) rather than #> (contains), but either works. You'll probably also want the limit so you can process the work in batches (keep running this and you're done for hour X if you get zero results; make sure you're done well before the hour is up). You'll also probably want to pass the timestamp as a parameter to this query--I've inlined it here as now() to simplify things, but passing the time as a parameter makes testing this a lot easier.
Note also that Postgres can be picky with time zone names and abbreviations and its behavior with daylight saving time can be counterintuitive: e.g., Pacific Standard Time and Pacific Daylight Time are treated as two distinct time zones (for the purposes of AT TIME ZONE):
maciek=# select now() at time zone 'pst';
timezone
----------------------------
2015-10-09 23:14:51.856813
(1 row)
maciek=# select now() at time zone 'pdt';
timezone
----------------------------
2015-10-10 00:14:54.402524
(1 row)
That is, Daylight Saving Time is always there, whether you are currently observing it or not. If you're letting people enter the time zone directly, it's good to either reject these or automatically coerce these to 'America/Los_Angeles' (or whatever time zone they happen to map to), which will handle these conversions for you automatically according to the time zone rules your Postgres version has (make sure you update to point releases promptly if accuracy is critical here for areas that have frequent time zone changes). The list of time zone names used by Postgres can be found in the Olson database. The Postgres tables pg_timezone_names
and pg_timezone_abbrevs may also be of interest.