diff --git a/datafusion/physical-plan/src/joins/sort_merge_join/materializing_stream.rs b/datafusion/physical-plan/src/joins/sort_merge_join/materializing_stream.rs
index 4840b56f55fff..069e94d0a9fd6 100644
--- a/datafusion/physical-plan/src/joins/sort_merge_join/materializing_stream.rs
+++ b/datafusion/physical-plan/src/joins/sort_merge_join/materializing_stream.rs
@@ -235,6 +235,18 @@ pub(super) struct BufferedBatch {
     pub null_joined: Vec<usize>,
     /// Size estimation used for reserving / releasing memory
     pub size_estimation: usize,
+    /// Memory footprint of `join_arrays` cached at construction time.
+    /// Used during spill to track the residual memory that remains after
+    /// the main batch is written to disk.
+    pub join_arrays_mem: usize,
+    /// Actual amount tracked in the memory reservation for this batch.
+    ///
+    /// - `InMemory`: equals `size_estimation` (full batch + join_arrays + metadata)
+    /// - `Spilled`: equals `join_arrays_mem` (join key arrays stay in memory)
+    ///
+    /// Invariant: `free_reservation()` shrinks by exactly this amount, so we never
+    /// shrink by more than we grew.
+    pub reserved_amount: usize,
     /// Tracks filter outcomes for buffered rows in full outer joins.
     /// Indexed by absolute row position within the batch. See [`FilterState`].
     pub join_filter_status: Vec<FilterState>,
@@ -258,11 +270,13 @@ impl BufferedBatch {
         // + worst case null_joined (as vector capacity * element size)
         // + Range size
         // + size of this estimation
+        let join_arrays_mem: usize = join_arrays
+            .iter()
+            .map(|arr| arr.get_array_memory_size())
+            .sum();
+
         let size_estimation = batch.get_array_memory_size()
-            + join_arrays
-                .iter()
-                .map(|arr| arr.get_array_memory_size())
-                .sum::<usize>()
+            + join_arrays_mem
             + batch.num_rows().next_power_of_two() * size_of::<usize>()
             + size_of::<Range<usize>>()
             + size_of::<usize>();
@@ -274,6 +288,8 @@ impl BufferedBatch {
             join_arrays,
             null_joined: vec![],
             size_estimation,
+            join_arrays_mem,
+            reserved_amount: 0,
             join_filter_status: vec![FilterState::Unvisited; num_rows],
             num_rows,
         }
@@ -947,18 +963,16 @@ impl MaterializingSortMergeJoinStream {
         }
     }
 
-    fn free_reservation(&mut self, buffered_batch: &BufferedBatch) -> Result<()> {
-        // Shrink memory usage for in-memory batches only
-        if let BufferedBatchState::InMemory(_) = buffered_batch.batch {
-            self.reservation
-                .try_shrink(buffered_batch.size_estimation)?;
+    fn free_reservation(&mut self, buffered_batch: &BufferedBatch) {
+        if buffered_batch.reserved_amount > 0 {
+            self.reservation.shrink(buffered_batch.reserved_amount);
         }
-        Ok(())
     }
 
     fn allocate_reservation(&mut self, mut buffered_batch: BufferedBatch) -> Result<()> {
         match self.reservation.try_grow(buffered_batch.size_estimation) {
             Ok(_) => {
+                buffered_batch.reserved_amount = buffered_batch.size_estimation;
                 self.join_metrics
                     .peak_mem_used()
                     .set_max(self.reservation.size());
@@ -978,6 +992,22 @@ impl MaterializingSortMergeJoinStream {
                             .unwrap(); // Operation only return None if no batches are spilled, here we ensure that at least one batch is spilled
 
                         buffered_batch.batch = BufferedBatchState::Spilled(spill_file);
+
+                        // Join key arrays remain in memory after the batch is
+                        // spilled — the comparator needs them for key boundary
+                        // detection. Force-grow the reservation so the pool
+                        // reflects actual memory usage even if this pushes
+                        // pool.reserved() above the configured limit. This is
+                        // safe because the memory is physically consumed and
+                        // not tracking it would let other operators over-allocate
+                        // against a stale pool view.
+                        let join_arrays_mem = buffered_batch.join_arrays_mem;
+                        self.reservation.grow(join_arrays_mem);
+                        buffered_batch.reserved_amount = join_arrays_mem;
+                        self.join_metrics
+                            .peak_mem_used()
+                            .set_max(self.reservation.size());
+
                         Ok(())
                     }
                     _ => internal_err!("Buffered batch has empty body"),
@@ -1006,7 +1036,7 @@ impl MaterializingSortMergeJoinStream {
                                 self.buffered_data.batches.pop_front()
                             {
                                 self.produce_buffered_not_matched(&mut buffered_batch)?;
-                                self.free_reservation(&buffered_batch)?;
+                                self.free_reservation(&buffered_batch);
                                 head_changed = true;
                             }
                         } else {
diff --git a/datafusion/physical-plan/src/joins/sort_merge_join/tests.rs b/datafusion/physical-plan/src/joins/sort_merge_join/tests.rs
index 5d70530528728..bc34c351c5e21 100644
--- a/datafusion/physical-plan/src/joins/sort_merge_join/tests.rs
+++ b/datafusion/physical-plan/src/joins/sort_merge_join/tests.rs
@@ -2487,6 +2487,223 @@ async fn overallocation_multi_batch_spill() -> Result<()> {
     Ok(())
 }
 
+/// Verifies that `peak_mem_used` reflects join_arrays memory on the spill path.
+///
+/// Uses a memory limit smaller than a single batch's `size_estimation` so that
+/// every batch spills — the `Ok` arm of `allocate_reservation` is never hit.
+/// Before the fix, `peak_mem_used` would stay 0 because `set_max` was only
+/// called in the `Ok` arm. After the fix, the spill path calls
+/// `grow(join_arrays_mem)` + `set_max`, so `peak_mem_used > 0`.
+#[tokio::test]
+async fn spill_join_arrays_memory_accounting() -> Result<()> {
+    use arrow::array::Array;
+
+    let left_batch = build_table_i32(
+        ("a1", &vec![0, 1]),
+        ("b1", &vec![1, 1]),
+        ("c1", &vec![4, 5]),
+    );
+    let size_estimation = left_batch.get_array_memory_size()
+        + Int32Array::from(vec![1, 1]).get_array_memory_size()
+        + 2usize.next_power_of_two() * size_of::<usize>()
+        + size_of::<std::ops::Range<usize>>()
+        + size_of::<usize>();
+    let join_arrays_mem = Int32Array::from(vec![1, 1]).get_array_memory_size();
+
+    // Memory limit: too small for a full batch, large enough for join_arrays.
+    // Every batch hits the Err arm → spills → grow(join_arrays_mem).
+    let memory_limit = (size_estimation + join_arrays_mem) / 2;
+    assert!(
+        memory_limit < size_estimation && memory_limit > join_arrays_mem,
+        "limit {memory_limit} must be between join_arrays_mem {join_arrays_mem} \
+         and size_estimation {size_estimation}"
+    );
+
+    let left_batches: Vec<RecordBatch> = (0..4)
+        .map(|i| {
+            build_table_i32(
+                ("a1", &vec![i * 2, i * 2 + 1]),
+                ("b1", &vec![1, 1]),
+                ("c1", &vec![100 + i, 101 + i]),
+            )
+        })
+        .collect();
+    let left = build_table_from_batches(left_batches);
+
+    let right_batches: Vec<RecordBatch> = (0..2)
+        .map(|i| {
+            build_table_i32(
+                ("a2", &vec![i * 2, i * 2 + 1]),
+                ("b2", &vec![1, 1]),
+                ("c2", &vec![200 + i, 201 + i]),
+            )
+        })
+        .collect();
+    let right = build_table_from_batches(right_batches);
+
+    let on = vec![(
+        Arc::new(Column::new_with_schema("b1", &left.schema())?) as _,
+        Arc::new(Column::new_with_schema("b2", &right.schema())?) as _,
+    )];
+    let sort_options = vec![SortOptions::default(); on.len()];
+
+    let runtime = RuntimeEnvBuilder::new()
+        .with_memory_limit(memory_limit, 1.0)
+        .with_disk_manager_builder(
+            DiskManagerBuilder::default().with_mode(DiskManagerMode::OsTmpDirectory),
+        )
+        .build_arc()?;
+
+    let session_config = SessionConfig::default().with_batch_size(50);
+    let task_ctx = Arc::new(
+        TaskContext::default()
+            .with_session_config(session_config)
+            .with_runtime(Arc::clone(&runtime)),
+    );
+
+    let join = join_with_options(
+        Arc::clone(&left),
+        Arc::clone(&right),
+        on.clone(),
+        Inner,
+        sort_options,
+        NullEquality::NullEqualsNothing,
+    )?;
+
+    let stream = join.execute(0, task_ctx)?;
+    let result = common::collect(stream).await.unwrap();
+
+    assert!(!result.is_empty(), "Expected non-empty join result");
+
+    let metrics = join.metrics().unwrap();
+    assert!(
+        metrics.spill_count().unwrap() > 0,
+        "Expected spilling to occur"
+    );
+
+    // Before the fix, peak_mem_used was 0 here because set_max was only
+    // called in the Ok arm of allocate_reservation, which is never reached
+    // when every batch spills. After the fix, the spill path calls
+    // grow(join_arrays_mem) + set_max unconditionally.
+    let peak_mem = metrics
+        .sum_by_name("peak_mem_used")
+        .map(|m| m.as_usize())
+        .unwrap_or(0);
+    assert!(
+        peak_mem >= join_arrays_mem,
+        "peak_mem_used ({peak_mem}) should be >= join_arrays_mem ({join_arrays_mem})"
+    );
+
+    // All memory must be released (grow/shrink balanced, no underflow)
+    assert_eq!(
+        runtime.memory_pool.reserved(),
+        0,
+        "All memory should be released after join completes"
+    );
+
+    Ok(())
+}
+
+/// Test the no-headroom scenario: pool is so tight that even
+/// join_arrays_mem exceeds the pool limit. With force-grow, the
+/// reservation still tracks the join_arrays unconditionally so the
+/// pool reflects actual memory usage.
+#[tokio::test]
+async fn spill_join_arrays_no_headroom() -> Result<()> {
+    use arrow::array::Array;
+
+    let join_arrays_mem = Int32Array::from(vec![1, 1]).get_array_memory_size();
+
+    // Pool smaller than join_arrays_mem: try_grow(size_estimation) fails → spill.
+    // Force-grow(join_arrays_mem) succeeds unconditionally → reserved_amount > 0.
+    let memory_limit = join_arrays_mem / 2;
+    assert!(
+        memory_limit < join_arrays_mem,
+        "limit {memory_limit} must be smaller than join_arrays_mem {join_arrays_mem}"
+    );
+
+    let left_batches: Vec<RecordBatch> = (0..4)
+        .map(|i| {
+            build_table_i32(
+                ("a1", &vec![i * 2, i * 2 + 1]),
+                ("b1", &vec![1, 1]),
+                ("c1", &vec![100 + i, 101 + i]),
+            )
+        })
+        .collect();
+    let left = build_table_from_batches(left_batches);
+
+    let right_batches: Vec<RecordBatch> = (0..2)
+        .map(|i| {
+            build_table_i32(
+                ("a2", &vec![i * 2, i * 2 + 1]),
+                ("b2", &vec![1, 1]),
+                ("c2", &vec![200 + i, 201 + i]),
+            )
+        })
+        .collect();
+    let right = build_table_from_batches(right_batches);
+
+    let on = vec![(
+        Arc::new(Column::new_with_schema("b1", &left.schema())?) as _,
+        Arc::new(Column::new_with_schema("b2", &right.schema())?) as _,
+    )];
+    let sort_options = vec![SortOptions::default(); on.len()];
+
+    let runtime = RuntimeEnvBuilder::new()
+        .with_memory_limit(memory_limit, 1.0)
+        .with_disk_manager_builder(
+            DiskManagerBuilder::default().with_mode(DiskManagerMode::OsTmpDirectory),
+        )
+        .build_arc()?;
+
+    let session_config = SessionConfig::default().with_batch_size(50);
+    let task_ctx = Arc::new(
+        TaskContext::default()
+            .with_session_config(session_config)
+            .with_runtime(Arc::clone(&runtime)),
+    );
+
+    let join = join_with_options(
+        Arc::clone(&left),
+        Arc::clone(&right),
+        on.clone(),
+        Inner,
+        sort_options,
+        NullEquality::NullEqualsNothing,
+    )?;
+
+    let stream = join.execute(0, task_ctx)?;
+    let result = common::collect(stream).await.unwrap();
+
+    assert!(!result.is_empty(), "Expected non-empty join result");
+
+    let metrics = join.metrics().unwrap();
+    assert!(
+        metrics.spill_count().unwrap() > 0,
+        "Expected spilling to occur"
+    );
+
+    // Force-grow means peak_mem_used is always tracked, even when pool is tight.
+    let peak_mem = metrics
+        .sum_by_name("peak_mem_used")
+        .map(|m| m.as_usize())
+        .unwrap_or(0);
+    assert!(
+        peak_mem >= join_arrays_mem,
+        "peak_mem_used ({peak_mem}) should be >= join_arrays_mem ({join_arrays_mem})"
+    );
+
+    // Pool should be fully released (grow/shrink balanced)
+    assert_eq!(
+        runtime.memory_pool.reserved(),
+        0,
+        "All memory should be released after join completes"
+    );
+
+    Ok(())
+}
+
 /// Build a c1 < c2 filter on the third column of each side.
 fn build_c1_lt_c2_filter(left_schema: &Schema, right_schema: &Schema) -> JoinFilter {
     JoinFilter::new(