apache · Dandandan · Jul 1, 2026 · Jul 2, 2026 · Jul 2, 2026
diff --git a/datafusion/common/src/config.rs b/datafusion/common/src/config.rs
@@ -1106,6 +1106,16 @@ config_namespace! {
         /// the parquet file
         pub pruning: bool, default = true
 
+        /// (reading) If true, a scan restricted to a byte range of a parquet
+        /// file reads the proportional slice of rows from row groups that
+        /// partially overlap the range, so multiple partitions can decode
+        /// different rows of the same row group in parallel ("morsel"
+        /// splitting). This parallelizes reading of files with fewer row
+        /// groups than partitions, e.g. files with a single large row group.
+        /// If false, each row group is read entirely by the single partition
+        /// whose range contains the offset of its first data page.
+        pub split_row_groups_by_range: bool, default = true
+
         /// (reading) If true, the parquet reader skip the optional embedded metadata that may be in
         /// the file Schema. This setting can help avoid schema conflicts when querying
         /// multiple parquet files with schemas containing compatible types but different metadata

diff --git a/datafusion/common/src/file_options/parquet_writer.rs b/datafusion/common/src/file_options/parquet_writer.rs
@@ -238,6 +238,7 @@ impl ParquetOptions {
             pushdown_filters: _,
             reorder_filters: _,
             force_filter_selections: _, // not used for writer props
+            split_row_groups_by_range: _, // not used for writer props
             allow_single_file_parallelism: _,
             maximum_parallel_row_group_writers: _,
             maximum_buffered_record_batches_per_stream: _,
@@ -494,6 +495,7 @@ mod tests {
             pushdown_filters: defaults.pushdown_filters,
             reorder_filters: defaults.reorder_filters,
             force_filter_selections: defaults.force_filter_selections,
+            split_row_groups_by_range: defaults.split_row_groups_by_range,
             allow_single_file_parallelism: defaults.allow_single_file_parallelism,
             maximum_parallel_row_group_writers: defaults
                 .maximum_parallel_row_group_writers,
@@ -613,6 +615,8 @@ mod tests {
                 pushdown_filters: global_options_defaults.pushdown_filters,
                 reorder_filters: global_options_defaults.reorder_filters,
                 force_filter_selections: global_options_defaults.force_filter_selections,
+                split_row_groups_by_range: global_options_defaults
+                    .split_row_groups_by_range,
                 allow_single_file_parallelism: global_options_defaults
                     .allow_single_file_parallelism,
                 maximum_parallel_row_group_writers: global_options_defaults

diff --git a/datafusion/datasource-parquet/src/access_plan.rs b/datafusion/datasource-parquet/src/access_plan.rs
@@ -360,6 +360,13 @@ impl ParquetAccessPlan {
         self.row_groups[idx].should_scan()
     }
 
+    /// Return true if any row group is accessed via a [`RowGroupAccess::Selection`]
+    pub fn has_selection(&self) -> bool {
+        self.row_groups
+            .iter()
+            .any(|rg| matches!(rg, RowGroupAccess::Selection(_)))
+    }
+
     /// Marks the i-th row group as fully matched.
     ///
     /// Fully matched row groups are still read according to their

diff --git a/datafusion/datasource-parquet/src/opener/mod.rs b/datafusion/datasource-parquet/src/opener/mod.rs