Dates and times in RethinkDB

RethinkDB has native support for millisecond-precision times with time zones. Some highlights:

  • Times are integrated with the official drivers, which will automatically convert to and from your language’s native time type.
  • Queries are timezone-aware, so you can ask questions like “Did this event happen on a Monday in the time zone where it was recorded?”
  • Times work as indexes, so you can efficiently retrieve events based on when they occurred.
  • Time operations are pure ReQL, which means that even complicated date-time queries can be distributed efficiently across the cluster.

Dates and Times Illustration

A quick example

First, let’s create a table and insert some events. We’ll insert the first event using a native time object, and the second with the epoch_time constructor:


    [{'id' => 0, 'timestamp' =>},
     {'id' => 1, 'timestamp' => r.epoch_time(1376436769.923)}]

Now, let’s get those back:

> r.table('events').run(conn).to_a
[{"timestamp"=>2013-08-13 16:32:48 -0700, "id"=>0},
 {"timestamp"=>2013-08-13 23:32:49 +0000, "id"=>1}]

You’ll notice that both times we inserted are returned as native Ruby Time objects. They’re in different time zones because creates a time object in the local time zone, but r.epoch_time creates a UTC time (it doesn’t know or care what time zone the client is in). If we had instead inserted, they’d both be in UTC when we retrieved them.

We can now filter based on these times:

> r.table('events').filter{|row| row['timestamp'].hours() > 20}.run(conn)
[{"timestamp"=>2013-08-13 23:32:49 +0000, "id"=>1}]
> r.table('events').filter{|row|
    row['timestamp'].in_timezone('-02:00').hours() > 20
[{"timestamp"=>2013-08-13 16:32:48 -0700, "id"=>0},
 {"timestamp"=>2013-08-13 23:32:49 +0000, "id"=>1}]

Or create a secondary index on them:

> r.table('events').index_create('timestamp').run(conn)
> r.table('events').between(r.epoch_time(1376436769.913),
                            :index => 'timestamp').run(conn)
[{"timestamp"=>2013-08-13 23:32:49 +0000, "id"=>1}]

Technical details

Times are stored on the server as seconds since epoch (UTC) with millisecond precision plus a time zone. Currently the only available time zones are minute-precision time offsets from UTC, but we may add support for DST-aware time zones in the future. Time zones are strings as specified by ISO 8601.

Times are considered equal when their epoch (UTC) time values are equal, regardless of what time zone they’re in. This is true for both comparisons and indexed operations. Times are compared in floating point with millisecond precision.

Most date operations are only defined on years in the range [1400, 10000] (but note that times in the year 10000 cannot be printed as ISO 8601 dates).

Leap-seconds aren’t well-supported right now: 2012-06-30T23:59:60 and 2012-07-01T00:00:00 parse to the same time.

Inserting times

You can insert times by simply passing a native Ruby Time object. If the local time object contains a time zone, the inserted time will have that time zone; if it doesn’t, the inserted time will be in UTC. (Check this if you’re using a third-party driver.)

> r.table('events').insert({'id' => 2, 'timestamp' =>}).run(conn)
{"unchanged"=>0, "skipped"=>0, "replaced"=>0, "inserted"=>1, "errors"=>0, "deleted"=>0}

You can also use (which the server interprets as the time the query was received in UTC), or construct a time using r.time, r.epoch_time, or r.iso8601.

> r.time(2013, r.august, 9, 18, 53, 15.012, '-07:00').to_iso8601().run(conn)
> r.epoch_time(1376074395.012).to_iso8601().run(conn)
> r.iso8601("2013-08-09T18:53:15.012-07:00").to_iso8601().run(conn)

Times may be used as the primary key for a table. Two times are considered equal if they have the same number of milliseconds since epoch (UTC), regardless of time zone.

> r.table('t').insert({'id' => r.iso8601("2013-08-09T11:58:00.1111-07:00")}).run(conn)
{"unchanged"=>0, "skipped"=>0, "replaced"=>0, "inserted"=>1, "errors"=>0, "deleted"=>0}
> r.table('t').insert({'id' => r.iso8601("2013-08-09T10:58:00.1112-08:00")}).run(conn)
{"unchanged"=>0, "skipped"=>0, "replaced"=>0, "inserted"=>0,
 "first_error"=>"Duplicate primary key `id`: ...", "errors"=>1, "deleted"=>0}

You may also insert a time by inserting a literal pseudotype object. This is useful if, for instance, you exported a row using :time_format => 'raw' (see Retrieving Times below).

Note: Avoid using keys matching the regular expression ^\$reql_.+\$$ in your objects. RethinkDB considers those to be reserved keywords.

> r.expr({'$reql_type$' => 'TIME',
          'epoch_time' => 1376075362.662,
          'timezone' => '+00:00'}).to_iso8601().run(conn)

Retrieving times

By default, times are converted into native time objects when they are retrieved from the server. This may be overridden by passing the optarg time_format to run. The only options right now are native, the default, and raw.

2013-08-09 19:09:11 UTC
2013-08-09 12:14:56 -0700
>, :time_format => 'raw')
{'timezone'=>'+00:00', "epoch_time"=>1376075362.662, "$reql_type$"=>"TIME"}
>'-07:00').run(conn, :time_format => 'raw')
{"timezone"=>"-07:00", "epoch_time"=>1376075702.485, "$reql_type$"=>"TIME"}

You can also transform a time object on the server using either to_epoch_time or to_iso8601.


Working with times

There are only three useful things you can do with a time: modify it, compare it to another time, or retrieve a portion of it.

Modifying times

You can put a time into a new time zone:

> r.expr(
> r.expr('-06:00').to_iso8601().run(conn)

You can also add or subtract a duration (in seconds):

> (r.time(2015, 1, 1, 'Z') + 86400).run(conn)
2015-01-02 00:00:00 +0000

If you subtract two times, you get a duration:

> (r.time(2015, 1, 2, 'Z') - r.time(2015, 1, 1, 'Z')).run(conn)

Comparing times

All of the normal comparison operators are defined on times:

> (r.epoch_time(1376081287.982) <

Times are only compared with millisecond precision:

> r.epoch_time(1376081287.9821).eq(r.epoch_time(1376081287.9822)).run(conn)

There’s also the during command, which can check whether a time is in a particular range of times.

Retrieving portions of times

If you have a time, you can retrieve a particular portion (like the month, or the hours) relative to the current time zone. (See the full list at the API reference.)

> r.expr(
2013-08-09 13:53:00 -0700
> r.expr(
> r.expr(
> r.expr('-06:00').hours().run(conn)

We use the ISO 8601 definition of a week, which starts with Monday, represented as 1.

> r.expr(
5 # Friday

We define r.monday...r.sunday and r.january...r.december for convenience:

> r.expr(

We also let you slice the time into the date and the current time of day (a time and a duration, respectively):


Putting it all together

By combining these operations, you can write surprisingly useful queries in pure ReQL. For example, let’s say you have a table of sales your company has made, and you want to figure out how much of the gross comes from people who were working overtime:

r.table('sales').filter {|sale|
  # Weekends are overtime.
  sale['time'].day_of_week().eq(r.saturday) |
  sale['time'].day_of_week().eq(r.sunday) |
  # Weekdays outside 9-5 are overtime.
  (sale['time'].hours() < 9) |
  (sale['time'].hours() >= 17)

If your timestamps are stored with time zones, this query will work even if you have sales from different offices in different countries (assuming they all work 9-5 local time).

Since this query is pure ReQL, the entire query will be distributed efficiently over the cluster without any computation being done on the client.

Further, because it’s ReQL, the query’s individual pieces are easily composable. If you decide you want those numbers on a per-month basis, you can just throw a group in there:

r.table('sales').filter {|sale|
  # Weekends are overtime.
  sale['time'].day_of_week().eq(r.saturday) |
  sale['time'].day_of_week().eq(r.sunday) |
  # Weekdays outside 9-5 are overtime.
  (sale['time'].hours() < 9) |
  (sale['time'].hours() >= 17)
}.group{|sale| sale['time'].month()}.sum('dollars').run(conn)