BigQuery converting to a different timezone

Question

I am storing data in unixtimestamp on google big query. However, when the user will ask for a report, she will need the filtering and grouping of data by her local timezone.

The data is stored in GMT. The user may wish to see the data in EST. The report may ask the data to be grouped by date.

I don't see the timezone conversion function here:

Does anyone know how I can do this in bigquery? i.e. how do i group by after converting the timestamp to a different timezone?

Solution 1

2016 update: Look answers below, BigQuery now provides timestamp and timezone methods.

---

You are right - BigQuery doesn't provide any timestamp conversion methods.

In this case, I suggest that you run your GROUP BY based on dimensions of the GMT/UTC timestamp field, and then convert and display the result in the local timezone in your code.

Solution 2

Standard SQL in BigQuery has built-in functions:

DATE(timestamp_expression, timezone)
TIME(timestamp, timezone)
DATETIME(timestamp_expression, timezone)

Example:

SELECT 
   original,
   DATETIME(original, "America/Los_Angeles") as adjusted
FROM sometable;

+---------------------+---------------------+
| original            | adjusted            |
+---------------------+---------------------+
| 2008-12-25 05:30:00 | 2008-12-24 21:30:00 |
+---------------------+---------------------+

You can use standard IANA timezone names or offsets.

Solution 3

As of September 2016 BigQuery has adopted standard SQL and you can now just use the "DATE(timestamp, timezone)" function to offset for a timezone. You can reference their docs here:

BigQuery DATE docs

Solution 4

To those that stumble here:

How to convert a timestamp to another timezone?

Given that TIMESTAMP values, once constructed, are stored as UTC, and that TIMESTAMP does not have a constructor (TIMESTAMP, STRING), you can convert a timestamp to another time zone by transforming it first to a DATETIME and then constructing the new TIMESTAMP from the DATETIME in the new timezone:

SELECT TIMESTAMP(DATETIME(timestamp_field, '{timezone}'))

Example:

SELECT
    input_tz,
    input,
    'America/Montreal' AS output_tz,
    TIMESTAMP(DATETIME(input,'America/Montreal')) AS output
FROM (
    SELECT 'US/Pacific' AS input_tz, TIMESTAMP(DATETIME(DATE(2021, 1, 1), TIME(16, 0, 0)), 'US/Pacific') AS input
    UNION ALL
    SELECT 'UTC' AS input_tz, TIMESTAMP(DATETIME(DATE(2021, 1, 1), TIME(16, 0, 0)), 'UTC') AS input
    UNION ALL
    SELECT 'Europe/Berlin' AS input_tz, TIMESTAMP(DATETIME(DATE(2021, 1, 1), TIME(16, 0, 0)), 'Europe/Berlin') AS input
) t

results in:

Row	input_tz	input	output_tz	output
1	US/Pacific	2021-01-02 00:00:00 UTC	America/Montreal	2021-01-01 19:00:00 UTC
2	UTC	2021-01-01 16:00:00 UTC	America/Montreal	2021-01-01 11:00:00 UTC
3	Europe/Berlin	2021-01-01 15:00:00 UTC	America/Montreal	2021-01-0110:00:00 UTC

How to strip time zone info from a DATETIME value?

DATETIME in BigQuery are time zone naive, such that they do not contain timezone info. This being said, if you have business knowledge that allows you to know the timezone of a DATETIME, you can strip that timezone offset by converting it to a TIMESTAMP with the known timezone:

SELECT TIMESTAMP(datetime_value, '{timezone}')

Given that the TIMESTAMP stores the value in UTC, you can then re-convert to DATETIME if that's your preferred method of storage, but now you'll know that your DATETIME is in UTC :)

Hopefully this can be helpful! :)

Solution 5

Your premise is right. If you group like this, then users who want EST or EDT will get incorrect date grouping:

GROUP BY UTC_USEC_TO_DAY(ts_field)

But as long as you figure out the offset that your user wants, you can still do the full calculation on the server. For example, if EST is 5 hours behind UTC then query like this:

GROUP BY UTC_USEC_TO_DAY(ts_field - (5*60*60*1000*1000000) )

Just parameterize the "5" to be the offset in hours, and you're all set. Here's a sample based on one of the sample data sets:

SELECT
  COUNT(*) as the_count,
  UTC_USEC_TO_DAY(timestamp * 1000000 - (5*60*60*1000*1000000) ) as the_day
FROM
  [publicdata:samples.wikipedia]
WHERE
  comment CONTAINS 'disaster'
  and timestamp >= 1104537600
GROUP BY
  the_day
ORDER BY
  the_day

You can remove the offset to see how some edits move to different days.

View more solutions