converting to timestamp with time zone failed on Athena

Question

I\'m trying to create to following view:

CREATE OR REPLACE VIEW view_events AS 
(
   SELECT
     \"rank\"() OVER (PARTITION BY \"tb1\".\"innerid\" ORDER BY \"tb1         


        

Answer 1

Unfortunately Athena doesn't fully support all Presto features, it has limitations and is technically a few versions behind Presto. There's some attempt to make Athena integrate closely with the AWS Glue Metastore, which while based on Hive's metastore has some inconsistencies. I wish that Spark, Hive, Glue, Athena, Presto et al would just work with the same metastore, it would make life easier, but back to your issue:

This document about an older teradata fork of Presto mentions some issues with timestamp in presto:

Presto’s method for declaring timestamps with/with out timezone is not sql standard. In Presto, both are declared using the word TIMESTAMP, e.g. TIMESTAMP '2003-12-10 10:32:02.1212' or TIMESTAMP '2003-12-10 10:32:02.1212 UTC'. The timestamp is determined to be with or without timezone depending on whether you include a time zone at the end of the timestamp. In other systems, timestamps are explicitly declared as TIMESTAMP WITH TIME ZONE or TIMESTAMP WITHOUT TIME ZONE

The version of Presto that Athena is forked from does support both timestamp and timestamp with timezone but with that quirk as mentioned in the teradata docs which shouldn't be an issue. The real issue is that Athena does not support timestamp with timezone.

The presto docs you've linked to show that the function returns a value of that unsupported type timestamp with timezone, so you need to cast it as something else that is supported. It's an oversight that Athena allows functions and casting to a datatype that is then not supported, and hopefully that will be remedied, but for now you have to work around it.

What you need to do is use the CAST() function around that function call, which will change the type from timestamp with time zone into timestamp

Unfortunately you probably can't cast the string directly to a timestamp, although it depends on how the string is formatted. You also can't use the style of casting where you write timestamp before the string e.g. can't do timestamp '2018-01-01 15:00:00' for reasons I will explain below.

Type returned by the from_iso1601_timestamp() function

SELECT typeof("real_date") AS real_date_type
FROM
(
SELECT From_iso8601_timestamp('2018-01-01T15:00:00Z') as "real_date"
)

timestamp with time zone

This doesn't work

SELECT typeof("real_date") AS real_date_type
FROM
(
SELECT CAST('2018-01-01T15:00:00Z' AS timestamp) as "real_date"
)

SQL Error [FAILED]: INVALID_CAST_ARGUMENT: Value cannot be cast to timestamp

This style of Casting also returns timestamp with timezone :(

Note that the SELECT part of this works, and it says that it is a timestamp, but for some internal inconsistency reason you can't create a view and you'll get an error.

CREATE OR replace VIEW test 
AS 
SELECT typeof( "real_date" ) AS real_date_type
FROM
(
SELECT  timestamp '2018-01-01 15:00:00' as "real_date"
)

SQL Error [FAILED]: Could not initialize class com.facebook.presto.util.DateTimeZoneIndex

For whatever reason, creating a view requires that java class while parsing the value in the select doesn't. It's a bug that should be addressed.

This works yay

CREATE OR REPLACE VIEW test
AS
SELECT typeof("real_date") AS real_date_type
FROM
(
SELECT CAST(From_iso8601_timestamp('2018-01-01T15:00:00Z') AS timestamp) as "real_date"
)