You should use :utc_datetime unless you have a reason not to. The reason why Ecto’s timestamps() are naive by default is backwards compatibility. Since they are always UTC anyway, you should use :utc_datetime for them (and I believe the default will be changed to that in the future).

The difference between :utc_datetime and :naive_datetime is that the former will ensure that you can only insert UTC DateTimes and will read back UTC DateTime structs from the DB, whereas the latter will remove timezone information when writing and return NaiveDateTimes when you read from the DB.

For most cases, you should use UTC.

BTW, if you need to store the timezone too (in PostgreSQL), you have no option but to use a naive datetime field and store the offset or timezone information separately. This is because PostgreSQL does not have a data type for storing a timestamp with timezone.

There is the confusingly named timestamp with timezone but all it does is convert your input timestamp to UTC and convert it back to whatever your DB connection’s timezone is when reading (so for most cases it’s useless). It does not even store the offset/timezone. So you need to do that yourself somehow.

Also, use naive/UTC to store datetimes in the past, use timezones only when storing datetimes in the future, don’t mix them (personal experience). ^.^;

for datetime in a specific location you generally need “wall time” - as timezones and/or daylight saving time changes..

something like calecto GitHub - lau/calecto: Adapter for the Calendar library in Ecto · GitHub will get you wall time..

time is fun (not!)

Exactly, that is why future times should always have a time zone.

The “timestamp with timezone” type still seems to be useful since it makes all clients that try to read from this database interpret it as UTC time. Otherwise each client might try to interpret the timestamp in its local time, which would be a total mess. I find this article to summarize everything related to timezone in Elixir + Postgres pretty well: Time zones in PostgreSQL, Elixir and Phoenix | AmberBit Sp. z o. o.

The following argues that the exact same mess would happen when using timestamptz as well and that being the reason to not use it in ecto in the first place: utc_datetime having no affect on PostgreSQL DDL · Issue #1868 · elixir-ecto/ecto · GitHub

Interesting. The author of the article I linked to actually seems to think that this is the desired behavior:

While this will be true when you connect to PostgreSQL from your Elixir application, as soon as you connect using different client (psql, Ruby driver etc), the problems will start creeping in. SQL queries like:

SELECT * FROM events WHERE events.start_date < CURRENT_TIMESTAMP;

will have different meaning when you are in Europe/Warsaw time zone, and very much different when you are in PDT (California, USA), because of the 9 hour difference. If you have a reporting or maintenance script that you run against this database, depending on your client locale settings, you will miss more or less events from the query, or include unwanted ones in results. It would only be correct if your client’s settings were UTC.

How can you fix this issue? Use timestamptz data type in PostgreSQL, which is a shortcut to timestamp with time zone.

i.e. because the timestamp will then be guaranteed to be output in local time, the SQL queries will behave consistently across different clients.

I haven’t explored it myself yet so I’m not sure if this would indeed result in consistent behaviors for SQL queries, as the author claimed.

Wrong, timestamptz will return the timestamp in the connection’s local timezone. They explicitly won’t be getting UTC from the database, unless they have set the connection to UTC. Quoting:

When a timestamp with time zone value is output, it is always converted from UTC to the current timezone zone, and displayed as local time in that zone.

This is what I never want to happen.

Now it’s true that if you fire up psql or some script and compare the timestamp values to for example NOW(), you may end up with unexpected results if you didn’t remember to set the connection timezone. For me, using timestamptz here would be optimising for a special case and making the general case worse. It’s also reliant on magic (and frankly unintuitive) behaviour from PostgreSQL, and I feel being explicit in what you do (like you have to be when using plain timestamp) is better.

timestamp with time zone is a misnomer and IMO a misfeature that shouldn’t be in PostgreSQL.

Thanks a lot for the answer. This is definitely quite a rabbit hole to jump into and time representation is always a nightmare to deal with. At least one universal consensus seems to be to always use :utc_datetime in Ecto when there are no backwards compatibility worries. I’ll do that.

You’ve only reached the first tunnel of the rabbit hole because there’s timex and timex_ecto too.