MDEV-10630 rpl.rpl_mdev6020 fails in buildbot with timeout

The issue was that when running with valgrind the wait for master_pos_Wait()
was not long enough.

This patch also fixes two other failures that could affect rpl_mdev6020:
- check_if_conflicting_replication_locks() didn't properly check domains
- 'did_mark_start_commit' was after signals to other threads was sent which could
  get the variable read too early.

client/mysqltest.cc

@@ -181,6 +181,7 @@ static uint my_end_arg= 0;
 static uint opt_tail_lines= 0;
 
 static uint opt_connect_timeout= 0;
+static uint opt_wait_for_pos_timeout= 0;
 
 static char delimiter[MAX_DELIMITER_LENGTH]= ";";
 static uint delimiter_length= 1;
@@ -4659,7 +4660,7 @@ void do_sync_with_master2(struct st_command *command, long offset,
   MYSQL_ROW row;
   MYSQL *mysql= cur_con->mysql;
   char query_buf[FN_REFLEN+128];
-  int timeout= 300; /* seconds */
+  int timeout= opt_wait_for_pos_timeout;
 
   if (!master_pos.file[0])
     die("Calling 'sync_with_master' without calling 'save_master_pos'");
@@ -7098,6 +7099,10 @@ static struct my_option my_long_options[] =
    "Number of seconds before connection timeout.",
    &opt_connect_timeout, &opt_connect_timeout, 0, GET_UINT, REQUIRED_ARG,
    120, 0, 3600 * 12, 0, 0, 0},
+  {"wait_for_pos_timeout", 0,
+   "Number of seconds to wait for master_pos_wait",
+   &opt_wait_for_pos_timeout, &opt_wait_for_pos_timeout, 0, GET_UINT,
+   REQUIRED_ARG, 300, 0, 3600 * 12, 0, 0, 0},
   {"plugin_dir", 0, "Directory for client-side plugins.",
     &opt_plugin_dir, &opt_plugin_dir, 0,
    GET_STR, REQUIRED_ARG, 0, 0, 0, 0, 0, 0},

mysql-test/mysql-test-run.pl

@@ -5899,7 +5899,7 @@ sub start_mysqltest ($) {
   {
     # We are running server under valgrind, which causes some replication
     # test to be much slower, notable rpl_mdev6020.  Increase timeout.
-    mtr_add_arg($args, "--wait-for-pos-timeout=1500");
+    mtr_add_arg($args, "--wait-for-pos-timeout=0");
   }
 
   if ( $opt_ssl )

mysql-test/suite/rpl/t/rpl_mdev6020.test

@@ -1,8 +1,10 @@
+# Running this with valgrind can take > 5000 seconds with xtradb
+--source include/not_valgrind.inc
+
 --source include/have_innodb.inc
 --source include/have_partition.inc
 --source include/have_binlog_format_mixed_or_row.inc
 --source include/master-slave.inc
-
 --connection slave
 --source include/stop_slave.inc
 

sql/mdl.cc

@@ -443,7 +443,9 @@ public:
   virtual void notify_conflicting_locks(MDL_context *ctx) = 0;
 
   virtual bitmap_t hog_lock_types_bitmap() const = 0;
+#ifndef DBUG_OFF
   bool check_if_conflicting_replication_locks(MDL_context *ctx);
+#endif
 
   /** List of granted tickets for this lock. */
   Ticket_list m_granted;
@@ -2303,16 +2305,23 @@ void MDL_scoped_lock::notify_conflicting_locks(MDL_context *ctx)
   and trying to get an exclusive lock for the table.
 */
 
+#ifndef DBUG_OFF
 bool MDL_lock::check_if_conflicting_replication_locks(MDL_context *ctx)
 {
   Ticket_iterator it(m_granted);
   MDL_ticket *conflicting_ticket;
+  rpl_group_info *rgi_slave= ctx->get_thd()->rgi_slave;
+
+  if (!rgi_slave->gtid_sub_id)
+    return 0;
 
   while ((conflicting_ticket= it++))
   {
     if (conflicting_ticket->get_ctx() != ctx)
     {
       MDL_context *conflicting_ctx= conflicting_ticket->get_ctx();
+      rpl_group_info *conflicting_rgi_slave;
+      conflicting_rgi_slave= conflicting_ctx->get_thd()->rgi_slave;
 
       /*
         If the conflicting thread is another parallel replication
@@ -2320,15 +2329,18 @@ bool MDL_lock::check_if_conflicting_replication_locks(MDL_context *ctx)
         the current transaction has started too early and something is
         seriously wrong.
       */
-      if (conflicting_ctx->get_thd()->rgi_slave &&
-          conflicting_ctx->get_thd()->rgi_slave->rli ==
-          ctx->get_thd()->rgi_slave->rli &&
-          !conflicting_ctx->get_thd()->rgi_slave->did_mark_start_commit)
+      if (conflicting_rgi_slave &&
+          conflicting_rgi_slave->gtid_sub_id &&
+          conflicting_rgi_slave->rli == rgi_slave->rli &&
+          conflicting_rgi_slave->current_gtid.domain_id ==
+          rgi_slave->current_gtid.domain_id &&
+          !conflicting_rgi_slave->did_mark_start_commit)
         return 1;                               // Fatal error
     }
   }
   return 0;
 }
+#endif
 
 
 /**

sql/rpl_rli.cc

@@ -1921,8 +1921,8 @@ rpl_group_info::mark_start_commit_no_lock()
 {
   if (did_mark_start_commit)
     return;
-  mark_start_commit_inner(parallel_entry, gco, this);
   did_mark_start_commit= true;
+  mark_start_commit_inner(parallel_entry, gco, this);
 }
 
 
@@ -1933,12 +1933,12 @@ rpl_group_info::mark_start_commit()
 
   if (did_mark_start_commit)
     return;
+  did_mark_start_commit= true;
 
   e= this->parallel_entry;
   mysql_mutex_lock(&e->LOCK_parallel_entry);
   mark_start_commit_inner(e, gco, this);
   mysql_mutex_unlock(&e->LOCK_parallel_entry);
-  did_mark_start_commit= true;
 }
 
 
@@ -1981,12 +1981,12 @@ rpl_group_info::unmark_start_commit()
 
   if (!did_mark_start_commit)
     return;
+  did_mark_start_commit= false;
 
   e= this->parallel_entry;
   mysql_mutex_lock(&e->LOCK_parallel_entry);
   --e->count_committing_event_groups;
   mysql_mutex_unlock(&e->LOCK_parallel_entry);
-  did_mark_start_commit= false;
 }