[ARROW-17679] slow performance when reading data from GCP

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 9.0.0
• Component/s: Parquet, Python
• External issue URL: https://github.com/apache/arrow/issues/32919
• Assignee: Unassigned
• Reporter: mimoune djouallah
• Created: 12/Sep/22 05:13
• Updated: 11/Jan/23 11:52

Description

I am using pyarrow and duckdb to query some parquet files in GCP, thanks for making  the experience so smooth, but I have an issue with the performance, see code used.
import pyarrow.dataset as ds
import duckdb
import json
lineitem = ds.dataset("gs://xxxxx/lineitem")
lineitem_partition = ds.dataset("gs://xxxx/yyy",format="parquet", partitioning="hive")
lineitem_180 = ds.dataset("gs://xxxxx/lineitem_180",format="parquet", partitioning="hive")
con = duckdb.connect()
con.register("lineitem", lineitem)
con.register("lineitem_partition", lineitem_partition)
con.register("lineitem_180", lineitem_180)
def Query(request):
    SQL = request.get_json().get('name')
    df = con.execute(SQL).df()
    return json.dumps(df.to_json(orient="records")), 200, {'Content-Type': 'application/json'}
 
the issue is I am getting slow some extremely slow throughput performance, around 30 MBper second, the same files using local ssd laptop is extremely fast.
I am not sure what's the issue, I tried using pyarrow compute Query and it is the same performance