MDEV-34042: Deadlock kill of XA PREPARE can break replication /...

MDEV-34042: Deadlock kill of XA PREPARE can break replication / rpl.rpl_parallel_multi_domain_xa sporadic failure

Refinement of the original patch.

Move the code to reset the kill up into the parent class
Xid_apply_log_event, to also fix the similar issue for XA COMMIT.

Increase the number of slave retries in the test case
rpl.rpl_parallel_multi_domain_xa to fix some sporadic failures. The test
generates massive amounts of conflicting transactions in multiple
independent domains, which can cause multiple rollback+retry for a
transaction as it conflicts with transactions in other domains one-by-one.
Signed-off-by: Kristian Nielsen <knielsen@knielsen-hq.org>

mysql-test/suite/rpl/r/rpl_parallel_multi_domain_xa.result

@@ -6,6 +6,8 @@ connection master;
 ALTER TABLE mysql.gtid_slave_pos ENGINE=InnoDB;
 connection slave;
 include/stop_slave.inc
+SET @old_transaction_retries                 = @@GLOBAL.slave_transaction_retries;
+SET @@global.slave_transaction_retries       = 1000;
 SET @old_parallel_threads                    = @@GLOBAL.slave_parallel_threads;
 SET @old_slave_domain_parallel_threads       = @@GLOBAL.slave_domain_parallel_threads;
 SET @@global.slave_parallel_threads          = 5;
@@ -45,6 +47,7 @@ include/stop_slave.inc
 SET @@global.slave_parallel_mode           = @old_parallel_mode;
 SET @@global.slave_parallel_threads        = @old_parallel_threads;
 SET @@global.slave_domain_parallel_threads = @old_slave_domain_parallel_threads;
+SET @@global.slave_transaction_retries     = @old_transaction_retries;
 include/start_slave.inc
 connection master;
 DROP TABLE t1;

mysql-test/suite/rpl/t/rpl_parallel_multi_domain_xa.test

@@ -21,6 +21,12 @@ ALTER TABLE mysql.gtid_slave_pos ENGINE=InnoDB;
 --connection slave
 --sync_with_master
 --source include/stop_slave.inc
+# This test runs huge number of transactions independently in parallel that
+# all conflict on a single row. This requires a large number of retries, as a
+# transaction can repeatedly conflict/deadlock with a large number of other
+# transactions (in a different domain) one by one.
+SET @old_transaction_retries                 = @@GLOBAL.slave_transaction_retries;
+SET @@global.slave_transaction_retries       = 1000;
 SET @old_parallel_threads                    = @@GLOBAL.slave_parallel_threads;
 SET @old_slave_domain_parallel_threads       = @@GLOBAL.slave_domain_parallel_threads;
 SET @@global.slave_parallel_threads          = 5;
@@ -160,6 +166,7 @@ SET @@global.slave_parallel_mode     = 'optimistic';
 SET @@global.slave_parallel_mode           = @old_parallel_mode;
 SET @@global.slave_parallel_threads        = @old_parallel_threads;
 SET @@global.slave_domain_parallel_threads = @old_slave_domain_parallel_threads;
+SET @@global.slave_transaction_retries     = @old_transaction_retries;
 --source include/start_slave.inc
 
 --connection master

sql/log_event_server.cc

@@ -4066,6 +4066,9 @@ int Xid_apply_log_event::do_apply_event(rpl_group_info *rgi)
   thd->wsrep_affected_rows= 0;
 #endif
 
+#ifndef DBUG_OFF
+  bool record_gtid_delayed_for_xa= false;
+#endif
   if (rgi->gtid_pending)
   {
     sub_id= rgi->gtid_sub_id;
@@ -4084,6 +4087,10 @@ int Xid_apply_log_event::do_apply_event(rpl_group_info *rgi)
                         return 1;
                       });
     }
+#ifndef DBUG_OFF
+    else
+      record_gtid_delayed_for_xa= true;
+#endif
   }
 
   general_log_print(thd, COM_QUERY, get_query());
@@ -4093,6 +4100,22 @@ int Xid_apply_log_event::do_apply_event(rpl_group_info *rgi)
   {
     DBUG_ASSERT(!thd->transaction->xid_state.is_explicit_XA());
 
+    DBUG_ASSERT(record_gtid_delayed_for_xa);
+    if (thd->rgi_slave->is_parallel_exec)
+    {
+      /*
+        With XA, since the transaction is prepared/committed without updating
+        the GTID pos (MDEV-32020...), we need here to clear any pending
+        deadlock kill.
+
+        Otherwise if the kill happened after the prepare/commit completed, it
+        might end up killing the subsequent GTID position update, causing the
+        slave to fail with error.
+      */
+      wait_for_pending_deadlock_kill(thd, thd->rgi_slave);
+      thd->reset_killed();
+    }
+
     if ((err= do_record_gtid(thd, rgi, false, &hton, true)))
       return err;
   }
@@ -4209,19 +4232,6 @@ int XA_prepare_log_event::do_commit()
   else
     res= trans_xa_commit(thd);
 
-  if (thd->rgi_slave->is_parallel_exec)
-  {
-    /*
-      Since the transaction is prepared/committed without updating the GTID pos
-      (MDEV-32020...), we need here to clear any pending deadlock kill.
-      Otherwise if the kill happened after the prepare/commit completed, it
-      might end up killing the subsequent GTID position update, causing the
-      slave to fail with error.
-    */
-    wait_for_pending_deadlock_kill(thd, thd->rgi_slave);
-    thd->reset_killed();
-  }
-
   return res;
 }
 #endif // HAVE_REPLICATION