Change in way read_glue_table addresses tables

Describe the bug Previously, I have used the following code without issue:

item_ids = wr.s3.read_parquet(path=f'{processed_data_bucket}/distinct_path')
keys = list(wr.s3.read_parquet_table(database=gluedatabase, table='path_tokens').itemid.unique())

Now when I try to run the same code, I can see there has been some changes in the way that awswrangler refers to parquet file locations, as I now get:

item_ids = wr.s3.read_parquet(path=f'{processed_data_bucket}/distinct_path')
...
InvalidArgumentValue: '<redacted s3 path that does not start with s3://>/distinct_path' is not a valid path. It MUST start with 's3://'

which can obviously be fixed by changing the specified path, however I am not sure how to resolve the read_parquet_table issue:

keys = list(wr.s3.read_parquet_table(database=gluedatabase, table='path_tokens').itemid.unique())
...
InvalidArgumentValue: '<redacted s3 path that does not start with s3://>/path_tokens' is not a valid path. It MUST start with 's3://'

digging into the call to read_parquet_table, I see that the location is resolved via

res = client_glue.get_table(DatabaseName=gluedatabase, Name="path_tokens")
res['Table']['StorageDescriptor']['Location']

>> <same path from above that does not include s3://>

Because of this, I figured it must have been a configuration issue in the data catalogue, but I checked and all the tables have the full location path specified including s3, as does the database, so I cannot see how to work around this issue

wr.version = 1.9.0 boto3.version = 1.14.53

[gkennos]

Looks correct to me?

[igorborgest]

A last tip about:

table_list = wr.s3.list_objects(f's3://{processed_data_bucket}')
table_list_for_updating = list(set([t.split(processed_data_bucket)[1].split('/')[1] for t in table_list]))

It can be done using wr.s3.list_directories() - It is faster cause will not fetch all filenames to the client side.