MDEV-15242 Poor RBR update performance with partitioned tables

Observed and described
partitioned engine execution time difference
between master and slave was caused by excessive invocation
of base_engine::rnd_init which was done also for partitions
uninvolved into Rows-event operation.
The bug's slave slowdown therefore scales with the number of partitions.

Fixed with applying an upstream patch.

References:
----------
https://bugs.mysql.com/bug.php?id=73648
Bug#25687813 REPLICATION REGRESSION WITH RBR AND PARTITIONED TABLES

sql/ha_partition.cc

@@ -5090,7 +5090,8 @@ int ha_partition::rnd_pos_by_record(uchar *record)
   if (unlikely(get_part_for_delete(record, m_rec0, m_part_info, &m_last_part)))
     DBUG_RETURN(1);
 
-  DBUG_RETURN(handler::rnd_pos_by_record(record));
+  int err= m_file[m_last_part]->rnd_pos_by_record(record);
+  DBUG_RETURN(err);
 }
 
 

sql/handler.h

@@ -3042,9 +3042,17 @@ class handler :public Sql_alloc
   */
   virtual int rnd_pos_by_record(uchar *record)
   {
+    int error;
     DBUG_ASSERT(table_flags() & HA_PRIMARY_KEY_REQUIRED_FOR_POSITION);
+
+    error = ha_rnd_init(false);
+    if (error != 0)
+      return error;
+
     position(record);
-    return rnd_pos(record, ref);
+    error = ha_rnd_pos(record, ref);
+    ha_rnd_end();
+    return error;
   }
   virtual int read_first_row(uchar *buf, uint primary_key);
 public:

sql/log_event.cc

@@ -12135,10 +12135,6 @@ int Rows_log_event::find_row(rpl_group_info *rgi)
     int error;
     DBUG_PRINT("info",("locating record using primary key (position)"));
 
-    if (!table->file->inited &&
-        (error= table->file->ha_rnd_init_with_error(0)))
-      DBUG_RETURN(error);
-
     error= table->file->ha_rnd_pos_by_record(table->record[0]);
     if (error)
     {