Force UTC time in Snowflake JOIN condition

user_10323675 0
Viewed 155

I have a setup where we have

  1. A landing table with a column of type TIMESTAMP_LTZ (consumption_date). Includes the timezone of +02:00
  2. A view (landing_view) that reads from the landing table
  3. A view (raw_data) that reads from a table that has a field of type TIMESTAMP_NTZ (SOURCE_TIMESTAMP), but the value itself is in UTC time.

I have to join the data from landing_view to the data from raw_data using the consumption_date and SOURCE_TIMESTAMP.

SELECT l.ID, l.consumption_date, l.RUN_TIME, r.DISPLAY_NAME, r.source_timestamp, r.value_as_double
FROM "raw_data" r
JOIN "landing_view" l
ON r.SOURCE_TIMESTAMP >= DATEADD(second,120, convert_timezone('UTC',l.consumption_date))
  and r.SOURCE_TIMESTAMP < DATEADD(second,1000, convert_timezone('UTC',l.consumption_date))

My problem is that the convert_timezone command does not seem to affect the join clause at all, insted the join is made using the local time included in the LTZ type (+02:00).

If I use the convert_timezone is a select, if works just fine, but for the JOIN it does not.

Is there a way I can tell snowflake to use UTC in the join?

snowflake-cloud-data-platform timezone-offset utc
1 Answers

This will depend on what your TIMEZONE setting is. See example below.

If TIMEZONE is UTC:

alter session set TIMEZONE = 'UTC';
select 
    -- 2AM UTC
    '2021-01-02 02:00:00'::timestamp_ntz as SOURCE_TIMESTAMP,
    -- 1AM UTC / 12PM Australia/Melbourne time / 1 hour before SOURCE_TIMESTAMP
    '2021-01-02 12:00:00 +1100'::timestamp_ltz as CONSUMPTION_DATE,
    -- so add one hour to CONSUMPTION_DATE should equal to SOURCE_TIMESTAMP
    SOURCE_TIMESTAMP = DATEADD(hour, 1, convert_timezone('UTC', CONSUMPTION_DATE)) as is_equal
;

+-------------------------------+-------------------------------+----------+
| SOURCE_TIMESTAMP              | CONSUMPTION_DATE              | IS_EQUAL |
|-------------------------------+-------------------------------+----------|
| 2021-01-02 02:00:00.000000000 | 2021-01-02 01:00:00.000 +0000 | True     |
+-------------------------------+-------------------------------+----------+

However, if you change your TIMEZONE setting to another timezone, the result will be different:

alter session set TIMEZONE = 'Australia/Melbourne';
select 
    -- 2AM UTC
    '2021-01-02 02:00:00'::timestamp_ntz as SOURCE_TIMESTAMP,
    -- 1AM UTC / 12PM Australia/Melbourne time / 1 hour before SOURCE_TIMESTAMP
    '2021-01-02 12:00:00 +1100'::timestamp_ltz as CONSUMPTION_DATE,
    -- so add one hour to CONSUMPTION_DATE should equal to SOURCE_TIMESTAMP
    SOURCE_TIMESTAMP = DATEADD(hour, 1, convert_timezone('UTC', CONSUMPTION_DATE)) as is_equal
;

+-------------------------------+-------------------------------+----------+
| SOURCE_TIMESTAMP              | CONSUMPTION_DATE              | IS_EQUAL |
|-------------------------------+-------------------------------+----------|
| 2021-01-02 02:00:00.000000000 | 2021-01-02 12:00:00.000 +1100 | False    |
+-------------------------------+-------------------------------+----------+

Since you SOURCE_TIMESTAMP stores UTC value, you should change your TIMEZONE setting to match it.

By the way, having CONVERT_TIMEZONE in the DATEADD is redundant, as it only adds extra operation, but not having any effects. See below example:

select 
    -- 1AM UTC / 9AM Australia/Perth time / 1 hour before SOURCE_TIMESTAMP
    '2021-01-02 09:00:00 +0800'::timestamp_ltz as CONSUMPTION_DATE,
    DATEADD(hour, 1, CONSUMPTION_DATE) as no_convert_tz,
    DATEADD(hour, 1, convert_timezone('UTC', CONSUMPTION_DATE)) as convert_tz,
    no_convert_tz = convert_tz
;

+-------------------------------+-------------------------------+-------------------------------+----------------------------+
| CONSUMPTION_DATE              | NO_CONVERT_TZ                 | CONVERT_TZ                    | NO_CONVERT_TZ = CONVERT_TZ |
|-------------------------------+-------------------------------+-------------------------------+----------------------------|
| 2021-01-02 12:00:00.000 +1100 | 2021-01-02 13:00:00.000 +1100 | 2021-01-02 02:00:00.000 +0000 | True                       |
+-------------------------------+-------------------------------+-------------------------------+----------------------------+

You can see that the last column returns True.

user_98793790
Related