Scalability bottleneck in ha_innodb::general_fetch

With "WL#6047 - Do not allocate trx id for read-only transactions"
m_prebuilt->trx->id is always 0 for read-only transactions. This makes
it useless as an index for fuzzy counters.

Use server thread id instead similarly to MySQL.

Author: svoj

storage/innobase/handler/ha_innodb.cc

@@ -10453,10 +10453,10 @@ ha_innobase::index_read(
 		table->status = 0;
 		if (m_prebuilt->table->is_system_db) {
 			srv_stats.n_system_rows_read.add(
-				(size_t) m_prebuilt->trx->id, 1);
+				thd_get_thread_id(m_prebuilt->trx->mysql_thd), 1);
 		} else {
 			srv_stats.n_rows_read.add(
-				(size_t) m_prebuilt->trx->id, 1);
+				thd_get_thread_id(m_prebuilt->trx->mysql_thd), 1);
 		}
 		break;
 
@@ -10778,10 +10778,10 @@ ha_innobase::general_fetch(
 		table->status = 0;
 		if (m_prebuilt->table->is_system_db) {
 			srv_stats.n_system_rows_read.add(
-				(size_t) m_prebuilt->trx->id, 1);
+				thd_get_thread_id(trx->mysql_thd), 1);
 		} else {
 			srv_stats.n_rows_read.add(
-				(size_t) m_prebuilt->trx->id, 1);
+				thd_get_thread_id(trx->mysql_thd), 1);
 		}
 		break;
 	case DB_RECORD_NOT_FOUND: