[pick](Serde-2.1) fix variant serde may lost num_rows when subcolumns…

… empty (apache#38413)

serialization object with empty subcolumns may lost num_rows, so need
to record num_rows and set back num_rows in serdes

backport apache#38413

be/src/agent/be_exec_version_manager.h

@@ -71,12 +71,13 @@ class BeExecVersionManager {
  *    g. do local merge of remote runtime filter
  *    h. "now": ALWAYS_NOT_NULLABLE -> DEPEND_ON_ARGUMENTS
 */
-constexpr inline int BeExecVersionManager::max_be_exec_version = 5;
+constexpr inline int BeExecVersionManager::max_be_exec_version = 6;
 constexpr inline int BeExecVersionManager::min_be_exec_version = 0;
 
 /// functional
 constexpr inline int BITMAP_SERDE = 3;
 constexpr inline int USE_NEW_SERDE = 4; // release on DORIS version 2.1
 constexpr inline int OLD_WAL_SERDE = 3; // use to solve compatibility issues, see pr #32299
+constexpr inline int VARIANT_SERDE = 6; // change variant serde to fix PR #38413
 
 } // namespace doris

be/src/vec/data_types/data_type_object.cpp

@@ -29,6 +29,7 @@
 #include <utility>
 #include <vector>
 
+#include "agent/be_exec_version_manager.h"
 #include "vec/columns/column_object.h"
 #include "vec/common/assert_cast.h"
 #include "vec/common/typeid_cast.h"
@@ -66,7 +67,6 @@ int64_t DataTypeObject::get_uncompressed_serialized_bytes(const IColumn& column,
         if (is_nothing(type)) {
             continue;
         }
-
         PColumnMeta column_meta_pb;
         column_meta_pb.set_name(entry->path.get_path());
         type->to_pb_column_meta(&column_meta_pb);
@@ -78,6 +78,10 @@ int64_t DataTypeObject::get_uncompressed_serialized_bytes(const IColumn& column,
         size += type->get_uncompressed_serialized_bytes(entry->data.get_finalized_column(),
                                                         be_exec_version);
     }
+    // serialize num of rows, only take effect when subcolumns empty
+    if (be_exec_version >= VARIANT_SERDE) {
+        size += sizeof(uint32_t);
+    }
 
     return size;
 }
@@ -121,6 +125,11 @@ char* DataTypeObject::serialize(const IColumn& column, char* buf, int be_exec_ve
     }
     // serialize num of subcolumns
     *reinterpret_cast<uint32_t*>(size_pos) = num_of_columns;
+    // serialize num of rows, only take effect when subcolumns empty
+    if (be_exec_version >= VARIANT_SERDE) {
+        *reinterpret_cast<uint32_t*>(buf) = column_object.rows();
+        buf += sizeof(uint32_t);
+    }
 
     return buf;
 }
@@ -155,6 +164,13 @@ const char* DataTypeObject::deserialize(const char* buf, IColumn* column,
         }
         column_object->add_sub_column(key, std::move(sub_column), type);
     }
+    size_t num_rows = 0;
+    // serialize num of rows, only take effect when subcolumns empty
+    if (be_exec_version >= VARIANT_SERDE) {
+        num_rows = *reinterpret_cast<const uint32_t*>(buf);
+        column_object->set_num_rows(num_rows);
+        buf += sizeof(uint32_t);
+    }
 
     column_object->finalize();
 #ifndef NDEBUG

fe/fe-common/src/main/java/org/apache/doris/common/Config.java

@@ -1859,7 +1859,7 @@ public class Config extends ConfigBase {
      * Max data version of backends serialize block.
      */
     @ConfField(mutable = false)
-    public static int max_be_exec_version = 5;
+    public static int max_be_exec_version = 6;
 
     /**
      * Min data version of backends serialize block.

regression-test/data/variant_p0/rqg/rqg5.out

@@ -0,0 +1,67 @@
+-- This file is automatically generated. You should know what you did if you want to edit this
+-- !rqg5 --
+0
+
+-- !rqg5_2 --
+50
+
+-- !rqg5_3 --
+50
+
+-- !rqg5_4 --
+50
+
+-- !rqg5_5 --
+50
+
+-- !rqg5_6 --
+50
+
+-- !rqg5_7 --
+0
+
+-- !rqg5_8 --
+0
+
+-- !rqg5_9 --
+50
+
+-- !rqg5_10 --
+50
+
+-- !rqg5_11 --
+50
+
+-- !rqg5_12 --
+50
+
+-- !rqg5_13 --
+50
+
+-- !rqg5_14 --
+50
+
+-- !rqg5_15 --
+0
+
+-- !rqg5_16 --
+100
+
+-- !rqg5_17 --
+100
+
+-- !rqg5_18 --
+100
+
+-- !rqg5_19 --
+100
+
+-- !rqg5_20 --
+100
+
+-- !rqg5_21 --
+100
+
+-- !rqg5_22 --
+0
+