Revert "Fix dataset-serialize benchmark by setting pre_buffer=False (#…

…152)"

This reverts commit d412d2f.

benchmarks/dataset_serialize_benchmark.py

@@ -199,7 +199,7 @@ def run(self, source, case=None, **kwargs):
 
                 yield self.benchmark(
                     f=self._get_benchmark_function(
-                        case, source.name, source_ds, source.format_str, dirpath
+                        case, source.name, source_ds, dirpath
                     ),
                     extra_tags=tags,
                     options=kwargs,
@@ -220,12 +220,7 @@ def run(self, source, case=None, **kwargs):
         self._report_dirsize_and_wipe(OUTPUT_DIR_PREFIX)
 
     def _get_benchmark_function(
-        self,
-        case,
-        source_name: str,
-        source_ds: ds.Dataset,
-        source_fmt: str,
-        dirpath: str,
+        self, case, source_name: str, source_ds: ds.Dataset, dirpath: str
     ):
         (selectivity, serialization_format) = case
 
@@ -239,13 +234,6 @@ def _get_benchmark_function(
         except KeyError:
             pass
 
-        # Need this or arrow#38438 will cause a segfault. TODO: remove once fixed.
-        data_read_kwargs = {}
-        if source_fmt == "parquet":
-            data_read_kwargs["fragment_scan_options"] = ds.ParquetFragmentScanOptions(
-                pre_buffer=False
-            )
-
         if n_rows_only:
             # Pragmatic method for reading only a subset of the data set. A
             # different method for sub selection of rows could use a
@@ -254,10 +242,10 @@ def _get_benchmark_function(
             # Note that `head()` returns a `Table` object, i.e. loads data
             # into memory.
             log.info("read %s rows of dataset %s into memory", n_rows_only, source_name)
-            data = source_ds.head(n_rows_only, **data_read_kwargs)
+            data = source_ds.head(n_rows_only)
         else:
             log.info("read complete dataset %s into memory", source_name)
-            data = source_ds.to_table(**data_read_kwargs)
+            data = source_ds.to_table()
 
         log.info("read source dataset into memory in %.4f s", time.monotonic() - t0)