Fix: dictionary changed size during iteration in GCSObjectMetadataClient

gokart/gcs_obj_metadata_client.py

@@ -159,10 +159,14 @@ def _get_label_size(label_name: str, label_value: str) -> int:
         )
         if current_total_metadata_size <= max_gcs_metadata_size:
             return labels
-        for label_name, label_value in reversed(labels.items()):
+        to_remove = []
+        for label_name, label_value in reversed(list(labels.items())):
             size = _get_label_size(label_name, label_value)
-            del labels[label_name]
+            to_remove.append(label_name)
             current_total_metadata_size -= size
             if current_total_metadata_size <= max_gcs_metadata_size:
                 break
+
+        for key in to_remove:
+            del labels[key]
         return labels

test/test_gcs_obj_metadata_client.py

@@ -140,6 +140,16 @@ def test_get_patched_obj_metadata_with_nested_required_task_outputs(self):
             got['__required_task_outputs'], '{"nested_task": {"nest": {"__gokart_task_name": "task1", "__gokart_output_path": "path/to/output1"}}}'
         )
 
+    def test_adjust_gcs_metadata_limit_size_runtime_error(self):
+        large_labels = {}
+        for i in range(100):
+            large_labels[f'key_{i}'] = 'x' * 1000
+
+        result = GCSObjectMetadataClient._adjust_gcs_metadata_limit_size(large_labels)
+
+        total_size = sum(len(k.encode('utf-8')) + len(v.encode('utf-8')) for k, v in result.items())
+        self.assertLessEqual(total_size, 8 * 1024)
+
 
 class TestGokartTask(unittest.TestCase):
     @patch.object(_DummyTaskOnKart, '_get_output_target')