Snowflake appending "_COL_0" when using COPY INTO with parquet - snowflake-cloud-data-platform

Question

My query runs like this:
Go to 2 different S3 buckets, do a join on the parquet files in those buckets, then copy the results to a new parquet file and copy into S3.
Im running into a strange issue where when I run a COPY INTO in Snowflake with parquet format, the parquet file created has an extra column, then converts the parquet I get from the SELECT/JOIN into a string. Also, the main select statement shows the correct object, I think its just the COPY INTO
Actual: "_COL_0": "{\"room_ids\":[],\"food_ids\":[],\"uid\":\"abc123d\"}
Expected: {"room_ids":[],"food_ids":[],"uid":"abc123d"}
Here is the query:
COPY INTO #S3_STAGE/new-data
FROM (
select *
from #S3_STAGE/food
(FILE_FORMAT => 'PARQ', pattern => '.*.parquet')
where $1:uid in
(select $1:uid from #S3_STAGE/rooms
(FILE_FORMAT => 'PARQ', pattern => '.*.parquet')
)
)
FILE_FORMAT = PARQ
A sample parquet would be like
{
{"uid": "abc123d", "foods": ["banana", "apple", "orange"]},
{"uid": "abc234d", "foods": ["banana", "apple", "grape"]},
{"uid": "abc345d", "foods": ["banana", "grape", "peach"]}
}

Answer 1

Try projecting the columns explicitly rather than using *.
Alias each column using quotes if you want the result to be lowercase
Set header=true in order to preserve column names and avoid the _COL_n
Docs:
Unloading a Relational Table to Parquet with Multiple Columns
Transforming Data During a Load
COPY INTO #S3_STAGE/new-data
FROM (
select
$1:room_ids as "room_ids",
$1:food_ids as "food_ids",
$1:uid as "uid"
from #S3_STAGE/food
(FILE_FORMAT => 'PARQ', pattern => '.*.parquet')
where $1:uid in
(select $1:uid from #S3_STAGE/rooms
(FILE_FORMAT => 'PARQ', pattern => '.*.parquet')
)
)
HEADER=true
FILE_FORMAT = PARQ