Snowflake offline store, feature inference doesn't work for VARCHAR types.

[tonyqiu2020]

Expected Behavior

When using Snowflake offline store, feature inference should consider VARCHAR columns in snowflake as str.

Current Behavior

When using Snowflake offline store, feature inference considers VARCHAR type columns in snowflake as object in pandas dataframe, and therefore cause UNKNOWN error.

https://github.com/feast-dev/feast/blob/master/sdk/python/feast/infra/offline_stores/snowflake_source.py#L196-L207

feast/sdk/python/feast/type_map.py

Line 532 in 98b8d8d

"object": ValueType.UNKNOWN,

Steps to reproduce

Specifications

• Version: 0.18.0+
• Platform: mac os
• Subsystem:

Possible Solution

I would suggest changing ValueType.UNKNOWN to ValueType.STRING since pandas dtype object is str or mixed. And typically we just need it as string

https://github.com/feast-dev/feast/blob/master/sdk/python/feast/type_map.py#L532

[kevjumba]

Hi Tony! Can you add a specific example of this not working? Have you encountered this because the way feature inference works is we capture the columns of a snowflake table and convert it to pandas dataframe as a single row.
The snowflake documentation states here very clearly that a VARCHAR datatype converts to str which is handled by the type map as a ValueType.STRING.
https://docs.snowflake.com/en/user-guide/python-connector-pandas.html

Also string datatypes in Snowflake have been working for Feast in the past.

[tonyqiu2020]

I think it is general any varchar or str in snowflake

create table test_table (
  name varchar (100),  -- variable string column
  preferences string, -- column used to store JSON type of data
  created_at timestamp
  )

insert into test_table
values ( 'test', 'str_test', '2020-04-04 16:57:53.653');

And if we run code like this part
https://github.com/feast-dev/feast/blob/master/sdk/python/feast/infra/offline_stores/snowflake_source.py#L196-L207

I got [('NAME', 'object'), ('PREFERENCES', 'object'), ('CREATED_AT', 'datetime64[ns]')]

[felixwang9817]

@tonyqiu2020 pretty sure #2558 resolves this! Feel free to reopen if I'm mistaken.