[Fix](Variant) variant fallthrough with inverted index (apache#40069)

When reading from segment, the schema type is variant, if we check type valid in `get_inverted_index`, the result should always return nullptr(since variant type it self does not support inverted index), but the actual storage could be `string` or etc.So we should ignore the type check and return the correct inverted index iterators introduced by apache#36163
eldenmoon · Sep 3, 2024 · 3314c8b · 3314c8b
1 parent e3b1d0c
commit 3314c8b
Show file tree

Hide file tree

Showing 6 changed files with 30 additions and 5 deletions.
diff --git a/be/src/olap/rowset/segment_v2/segment.cpp b/be/src/olap/rowset/segment_v2/segment.cpp
@@ -203,7 +203,8 @@ Status Segment::new_iterator(SchemaSPtr schema, const StorageReadOptions& read_o
         ColumnReader* reader = nullptr;
         if (col.is_extracted_column()) {
             auto relative_path = col.path_info_ptr()->copy_pop_front();
-            const auto* node = _sub_column_tree[col.unique_id()].find_exact(relative_path);
+            int32_t unique_id = col.unique_id() > 0 ? col.unique_id() : col.parent_unique_id();
+            const auto* node = _sub_column_tree[unique_id].find_exact(relative_path);
             reader = node != nullptr ? node->data.reader.get() : nullptr;
         } else {
             reader = _column_readers.contains(col.unique_id())
@@ -775,8 +776,9 @@ ColumnReader* Segment::_get_column_reader(const TabletColumn& col) {
     // init column iterator by path info
     if (col.has_path_info() || col.is_variant_type()) {
         auto relative_path = col.path_info_ptr()->copy_pop_front();
+        int32_t unique_id = col.unique_id() > 0 ? col.unique_id() : col.parent_unique_id();
         const auto* node = col.has_path_info()
-                                   ? _sub_column_tree[col.unique_id()].find_exact(relative_path)
+                                   ? _sub_column_tree[unique_id].find_exact(relative_path)
                                    : nullptr;
         if (node != nullptr) {
             return node->data.reader.get();

diff --git a/be/src/olap/rowset/segment_v2/segment_iterator.cpp b/be/src/olap/rowset/segment_v2/segment_iterator.cpp
@@ -1488,10 +1488,15 @@ Status SegmentIterator::_init_inverted_index_iterators() {
     }
     for (auto cid : _schema->column_ids()) {
         if (_inverted_index_iterators[cid] == nullptr) {
+            // Not check type valid, since we need to get inverted index for related variant type when reading the segment.
+            // If check type valid, we can not get inverted index for variant type, and result nullptr.The result for calling
+            // get_inverted_index with variant suffix should return corresponding inverted index meta.
+            bool check_inverted_index_by_type = false;
             // Use segment’s own index_meta, for compatibility with future indexing needs to default to lowercase.
             RETURN_IF_ERROR(_segment->new_inverted_index_iterator(
                     _opts.tablet_schema->column(cid),
-                    _segment->_tablet_schema->get_inverted_index(_opts.tablet_schema->column(cid)),
+                    _segment->_tablet_schema->get_inverted_index(_opts.tablet_schema->column(cid),
+                                                                 check_inverted_index_by_type),
                     _opts, &_inverted_index_iterators[cid]));
         }
     }