diff --git a/be/src/olap/rowset/segment_v2/segment_iterator.cpp b/be/src/olap/rowset/segment_v2/segment_iterator.cpp
index 82a3c8ba509b2d..de12ebb96c1b5b 100644
--- a/be/src/olap/rowset/segment_v2/segment_iterator.cpp
+++ b/be/src/olap/rowset/segment_v2/segment_iterator.cpp
@@ -1942,7 +1942,8 @@ Status SegmentIterator::_read_columns(const std::vector<ColumnId>& column_ids,
 }
 
 Status SegmentIterator::_init_current_block(
-        vectorized::Block* block, std::vector<vectorized::MutableColumnPtr>& current_columns) {
+        vectorized::Block* block, std::vector<vectorized::MutableColumnPtr>& current_columns,
+        uint32_t nrows_read_limit) {
     block->clear_column_data(_schema->num_column_ids());
 
     for (size_t i = 0; i < _schema->num_column_ids(); i++) {
@@ -1962,7 +1963,7 @@ Status SegmentIterator::_init_current_block(
                     column_desc->path() == nullptr ? "" : column_desc->path()->get_path());
             // TODO reuse
             current_columns[cid] = file_column_type->create_column();
-            current_columns[cid]->reserve(_opts.block_row_max);
+            current_columns[cid]->reserve(nrows_read_limit);
         } else {
             // the column in block must clear() here to insert new data
             if (_is_pred_column[cid] ||
@@ -1981,7 +1982,7 @@ Status SegmentIterator::_init_current_block(
                 } else if (column_desc->type() == FieldType::OLAP_FIELD_TYPE_DATETIME) {
                     current_columns[cid]->set_datetime_type();
                 }
-                current_columns[cid]->reserve(_opts.block_row_max);
+                current_columns[cid]->reserve(nrows_read_limit);
             }
         }
     }
@@ -2387,14 +2388,16 @@ Status SegmentIterator::_next_batch_internal(vectorized::Block* block) {
             }
         }
     }
-    RETURN_IF_ERROR(_init_current_block(block, _current_return_columns));
-    _converted_column_ids.assign(_schema->columns().size(), 0);
 
-    _current_batch_rows_read = 0;
     uint32_t nrows_read_limit = _opts.block_row_max;
     if (_can_opt_topn_reads()) {
         nrows_read_limit = std::min(static_cast<uint32_t>(_opts.topn_limit), nrows_read_limit);
     }
+
+    RETURN_IF_ERROR(_init_current_block(block, _current_return_columns, nrows_read_limit));
+    _converted_column_ids.assign(_schema->columns().size(), 0);
+
+    _current_batch_rows_read = 0;
     RETURN_IF_ERROR(_read_columns_by_index(
             nrows_read_limit, _current_batch_rows_read,
             _lazy_materialization_read || _opts.record_rowids || _is_need_expr_eval));
diff --git a/be/src/olap/rowset/segment_v2/segment_iterator.h b/be/src/olap/rowset/segment_v2/segment_iterator.h
index 5b89d1932fa262..d93f341b678644 100644
--- a/be/src/olap/rowset/segment_v2/segment_iterator.h
+++ b/be/src/olap/rowset/segment_v2/segment_iterator.h
@@ -222,7 +222,8 @@ class SegmentIterator : public RowwiseIterator {
                                                 bool set_block_rowid);
     void _replace_version_col(size_t num_rows);
     Status _init_current_block(vectorized::Block* block,
-                               std::vector<vectorized::MutableColumnPtr>& non_pred_vector);
+                               std::vector<vectorized::MutableColumnPtr>& non_pred_vector,
+                               uint32_t nrows_read_limit);
     uint16_t _evaluate_vectorization_predicate(uint16_t* sel_rowid_idx, uint16_t selected_size);
     uint16_t _evaluate_short_circuit_predicate(uint16_t* sel_rowid_idx, uint16_t selected_size);
     void _output_non_pred_columns(vectorized::Block* block);