diff --git a/datafusion/core/src/datasource/file_format/parquet.rs b/datafusion/core/src/datasource/file_format/parquet.rs
index 57211fcf45f0..e5fb7a95f12f 100644
--- a/datafusion/core/src/datasource/file_format/parquet.rs
+++ b/datafusion/core/src/datasource/file_format/parquet.rs
@@ -1193,8 +1193,10 @@ mod tests {
     use super::super::test_util::scan_format;
     use crate::datasource::listing::{ListingTableUrl, PartitionedFile};
     use crate::physical_plan::collect;
+    use crate::test_util::bounded_stream;
     use std::fmt::{Display, Formatter};
     use std::sync::atomic::{AtomicUsize, Ordering};
+    use std::time::Duration;
 
     use super::*;
 
@@ -2212,4 +2214,105 @@ mod tests {
 
         Ok(())
     }
+
+    #[tokio::test]
+    async fn parquet_sink_write_memory_reservation() -> Result<()> {
+        async fn test_memory_reservation(global: ParquetOptions) -> Result<()> {
+            let field_a = Field::new("a", DataType::Utf8, false);
+            let field_b = Field::new("b", DataType::Utf8, false);
+            let schema = Arc::new(Schema::new(vec![field_a, field_b]));
+            let object_store_url = ObjectStoreUrl::local_filesystem();
+
+            let file_sink_config = FileSinkConfig {
+                object_store_url: object_store_url.clone(),
+                file_groups: vec![PartitionedFile::new("/tmp".to_string(), 1)],
+                table_paths: vec![ListingTableUrl::parse("file:///")?],
+                output_schema: schema.clone(),
+                table_partition_cols: vec![],
+                overwrite: true,
+                keep_partition_by_columns: false,
+            };
+            let parquet_sink = Arc::new(ParquetSink::new(
+                file_sink_config,
+                TableParquetOptions {
+                    key_value_metadata: std::collections::HashMap::from([
+                        ("my-data".to_string(), Some("stuff".to_string())),
+                        ("my-data-bool-key".to_string(), None),
+                    ]),
+                    global,
+                    ..Default::default()
+                },
+            ));
+
+            // create data
+            let col_a: ArrayRef = Arc::new(StringArray::from(vec!["foo", "bar"]));
+            let col_b: ArrayRef = Arc::new(StringArray::from(vec!["baz", "baz"]));
+            let batch =
+                RecordBatch::try_from_iter(vec![("a", col_a), ("b", col_b)]).unwrap();
+
+            // create task context
+            let task_context = build_ctx(object_store_url.as_ref());
+            assert_eq!(
+                task_context.memory_pool().reserved(),
+                0,
+                "no bytes are reserved yet"
+            );
+
+            let mut write_task = parquet_sink.write_all(
+                Box::pin(RecordBatchStreamAdapter::new(
+                    schema,
+                    bounded_stream(batch, 1000),
+                )),
+                &task_context,
+            );
+
+            // incrementally poll and check for memory reservation
+            let mut reserved_bytes = 0;
+            while futures::poll!(&mut write_task).is_pending() {
+                reserved_bytes += task_context.memory_pool().reserved();
+                tokio::time::sleep(Duration::from_micros(1)).await;
+            }
+            assert!(
+                reserved_bytes > 0,
+                "should have bytes reserved during write"
+            );
+            assert_eq!(
+                task_context.memory_pool().reserved(),
+                0,
+                "no leaking byte reservation"
+            );
+
+            Ok(())
+        }
+
+        let write_opts = ParquetOptions {
+            allow_single_file_parallelism: false,
+            ..Default::default()
+        };
+        test_memory_reservation(write_opts)
+            .await
+            .expect("should track for non-parallel writes");
+
+        let row_parallel_write_opts = ParquetOptions {
+            allow_single_file_parallelism: true,
+            maximum_parallel_row_group_writers: 10,
+            maximum_buffered_record_batches_per_stream: 1,
+            ..Default::default()
+        };
+        test_memory_reservation(row_parallel_write_opts)
+            .await
+            .expect("should track for row-parallel writes");
+
+        let col_parallel_write_opts = ParquetOptions {
+            allow_single_file_parallelism: true,
+            maximum_parallel_row_group_writers: 1,
+            maximum_buffered_record_batches_per_stream: 2,
+            ..Default::default()
+        };
+        test_memory_reservation(col_parallel_write_opts)
+            .await
+            .expect("should track for column-parallel writes");
+
+        Ok(())
+    }
 }