From 959b392bf7f2c98ee1cdb317dcadc8f89fea512e Mon Sep 17 00:00:00 2001
From: rjzamora <rzamora217@gmail.com>
Date: Thu, 4 Nov 2021 09:52:14 -0700
Subject: [PATCH] add 9599 fix into standalone branch

---
 python/cudf/cudf/_lib/parquet.pyx | 12 +++++++++++-
 python/cudf/cudf/tests/test_s3.py |  4 +++-
 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/python/cudf/cudf/_lib/parquet.pyx b/python/cudf/cudf/_lib/parquet.pyx
index 5b73b1fef10..9c24e5becfd 100644
--- a/python/cudf/cudf/_lib/parquet.pyx
+++ b/python/cudf/cudf/_lib/parquet.pyx
@@ -118,9 +118,17 @@ cpdef read_parquet(filepaths_or_buffers, columns=None, row_groups=None,
     cudf.io.parquet.read_parquet
     cudf.io.parquet.to_parquet
     """
+
+    # Convert NativeFile buffers to NativeFileDatasource,
+    # but save original buffers in case we need to use
+    # pyarrow for metadata processing
+    # (See: https://github.com/rapidsai/cudf/issues/9599)
+    pa_buffers = []
     for i, datasource in enumerate(filepaths_or_buffers):
         if isinstance(datasource, NativeFile):
+            pa_buffers.append(datasource)
             filepaths_or_buffers[i] = NativeFileDatasource(datasource)
+
     cdef cudf_io_types.source_info source = make_source_info(
         filepaths_or_buffers)
 
@@ -203,7 +211,9 @@ cpdef read_parquet(filepaths_or_buffers, columns=None, row_groups=None,
             range_index_meta = index_col[0]
             if row_groups is not None:
                 per_file_metadata = [
-                    pa.parquet.read_metadata(s) for s in filepaths_or_buffers
+                    pa.parquet.read_metadata(s) for s in (
+                        pa_buffers or filepaths_or_buffers
+                    )
                 ]
 
                 filtered_idx = []
diff --git a/python/cudf/cudf/tests/test_s3.py b/python/cudf/cudf/tests/test_s3.py
index 2d7907a43b8..ff551ec74ca 100644
--- a/python/cudf/cudf/tests/test_s3.py
+++ b/python/cudf/cudf/tests/test_s3.py
@@ -272,7 +272,8 @@ def test_read_parquet_arrow_nativefile(s3_base, s3so, pdf, columns):
     assert_eq(expect, got)
 
 
-def test_read_parquet_filters(s3_base, s3so, pdf):
+@pytest.mark.parametrize("python_file", [True, False])
+def test_read_parquet_filters(s3_base, s3so, pdf, python_file):
     fname = "test_parquet_reader_filters.parquet"
     bname = "parquet"
     buffer = BytesIO()
@@ -284,6 +285,7 @@ def test_read_parquet_filters(s3_base, s3so, pdf):
             "s3://{}/{}".format(bname, fname),
             storage_options=s3so,
             filters=filters,
+            use_python_file_object=python_file,
         )
 
     # All row-groups should be filtered out