[Optimize](Variant) optimize schema update performance

When update schema with high concurrency, updaing schemas cost is expensive. 1. update schema only when rows is not 0 2. copy_from is expensive, use copy constructor
eldenmoon · Dec 16, 2024 · 8930824 · 8930824
1 parent cf6dba7
commit 8930824
Show file tree

Hide file tree

Showing 2 changed files with 11 additions and 11 deletions.
diff --git a/be/src/olap/rowset_builder.cpp b/be/src/olap/rowset_builder.cpp
@@ -346,21 +346,22 @@ Status RowsetBuilder::commit_txn() {
     SCOPED_TIMER(_commit_txn_timer);
 
     const RowsetWriterContext& rw_ctx = _rowset_writer->context();
-    if (rw_ctx.tablet_schema->num_variant_columns() > 0) {
+    if (rw_ctx.tablet_schema->num_variant_columns() > 0 && _rowset->num_rows() > 0) {
         // Need to merge schema with `rw_ctx.merged_tablet_schema` in prior,
         // merged schema keeps the newest merged schema for the rowset, which is updated and merged
         // during flushing segments.
         if (rw_ctx.merged_tablet_schema != nullptr) {
             RETURN_IF_ERROR(tablet()->update_by_least_common_schema(rw_ctx.merged_tablet_schema));
+        } else {
+            // We should merge rowset schema further, in case that the merged_tablet_schema maybe null
+            // when enable_memtable_on_sink_node is true, the merged_tablet_schema will not be passed to
+            // the destination backend.
+            // update tablet schema when meet variant columns, before commit_txn
+            // Eg. rowset schema:       A(int),    B(float),  C(int), D(int)
+            // _tabelt->tablet_schema:  A(bigint), B(double)
+            //  => update_schema:       A(bigint), B(double), C(int), D(int)
+            RETURN_IF_ERROR(tablet()->update_by_least_common_schema(rw_ctx.tablet_schema));
         }
-        // We should merge rowset schema further, in case that the merged_tablet_schema maybe null
-        // when enable_memtable_on_sink_node is true, the merged_tablet_schema will not be passed to
-        // the destination backend.
-        // update tablet schema when meet variant columns, before commit_txn
-        // Eg. rowset schema:       A(int),    B(float),  C(int), D(int)
-        // _tabelt->tablet_schema:  A(bigint), B(double)
-        //  => update_schema:       A(bigint), B(double), C(int), D(int)
-        RETURN_IF_ERROR(tablet()->update_by_least_common_schema(rw_ctx.tablet_schema));
     }
 
     // Transfer ownership of `PendingRowsetGuard` to `TxnManager`

diff --git a/be/src/vec/common/schema_util.cpp b/be/src/vec/common/schema_util.cpp
@@ -415,8 +415,7 @@ Status get_least_common_schema(const std::vector<TabletSchemaSPtr>& schemas,
     // Ensure that the output schema also excludes these extracted columns. This approach prevents
     // duplicated paths following the update_least_common_schema process.
     auto build_schema_without_extracted_columns = [&](const TabletSchemaSPtr& base_schema) {
-        output_schema = std::make_shared<TabletSchema>();
-        output_schema->copy_from(*base_schema);
+        output_schema = std::make_shared<TabletSchema>(*base_schema);
         // Merge columns from other schemas
         output_schema->clear_columns();
         // Get all columns without extracted columns and collect variant col unique id