MDEV-6321: close_temporary_tables() in format description event not serialised correctly

Follow-up patch, fixing a possible deadlock issue.

If the master crashes in the middle of an event group, there can be an active
transaction in a worker thread when we encounter the following master restart
format description event. In this case, we need to notify that worker thread
to abort and roll back the partial event group. Otherwise a deadlock occurs:
the worker thread waits for the commit that never arrives, and the SQL driver
thread waits for the worker thread to complete its event group, which it never
does.

2 files changed, 115 insertions, 0 deletions

diff --git a/mysql-test/suite/rpl/r/rpl_parallel_temptable.result b/mysql-test/suite/rpl/r/rpl_parallel_temptable.result
index 8aea6ccb1e8..61eba2cab2f 100644
--- a/mysql-test/suite/rpl/r/rpl_parallel_temptable.result
+++ b/mysql-test/suite/rpl/r/rpl_parallel_temptable.result
@@ -79,6 +79,43 @@ COUNT(*)
 SHOW STATUS LIKE 'Slave_open_temp_tables';
 Variable_name	Value
 Slave_open_temp_tables	0
+*** Test that if master logged partial event group before crash, we finish that group correctly before executing format description event ***
+include/stop_slave.inc
+CALL mtr.add_suppression("Statement accesses nontransactional table as well as transactional or temporary table, and writes to any of them");
+SET gtid_domain_id= 1;
+DELETE FROM t1;
+ALTER TABLE t1 ENGINE=InnoDB;
+CREATE TEMPORARY TABLE t2 (a INT PRIMARY KEY);
+INSERT INTO t2 VALUES (1);
+INSERT INTO t2 VALUES (2);
+SET gtid_domain_id= 2;
+CREATE TEMPORARY TABLE t3 (a INT PRIMARY KEY);
+INSERT INTO t3 VALUES (10);
+INSERT INTO t3 VALUES (20);
+INSERT INTO t1 SELECT a, 'server_1' FROM t2;
+INSERT INTO t1 SELECT a, 'default' FROM t3;
+INSERT INTO t1 SELECT a+2, '+server_1' FROM t2;
+FLUSH TABLES;
+SET SESSION debug_dbug="+d,crash_before_writing_xid";
+INSERT INTO t1 SELECT a+4, '++server_1' FROM t2;
+Got one of the listed errors
+INSERT INTO t1 VALUES (0, 1);
+include/save_master_gtid.inc
+include/start_slave.inc
+include/sync_with_master_gtid.inc
+SELECT * FROM t1 ORDER BY a;
+a	b
+0	1
+1	server_1
+2	server_1
+3	+server_1
+4	+server_1
+10	default
+20	default
+SHOW STATUS LIKE 'Slave_open_temp_tables';
+Variable_name	Value
+Slave_open_temp_tables	0
+FLUSH LOGS;
 include/stop_slave.inc
 SET GLOBAL slave_parallel_threads=@old_parallel_threads;
 include/start_slave.inc
diff --git a/mysql-test/suite/rpl/t/rpl_parallel_temptable.test b/mysql-test/suite/rpl/t/rpl_parallel_temptable.test
index c50bbd3f215..1504dad224e 100644
--- a/mysql-test/suite/rpl/t/rpl_parallel_temptable.test
+++ b/mysql-test/suite/rpl/t/rpl_parallel_temptable.test
@@ -1,3 +1,4 @@
+--source include/have_innodb.inc
 --source include/have_binlog_format_statement.inc
 --let $rpl_topology=1->2
 --source include/rpl_init.inc
@@ -132,6 +133,83 @@ SELECT COUNT(*) FROM t1 WHERE a BETWEEN 100+0 AND 100+256;
 SHOW STATUS LIKE 'Slave_open_temp_tables';
 
 
+--echo *** Test that if master logged partial event group before crash, we finish that group correctly before executing format description event ***
+
+--source include/stop_slave.inc
+
+--connection server_1
+CALL mtr.add_suppression("Statement accesses nontransactional table as well as transactional or temporary table, and writes to any of them");
+SET gtid_domain_id= 1;
+DELETE FROM t1;
+ALTER TABLE t1 ENGINE=InnoDB;
+CREATE TEMPORARY TABLE t2 (a INT PRIMARY KEY);
+INSERT INTO t2 VALUES (1);
+INSERT INTO t2 VALUES (2);
+
+--connection default
+SET gtid_domain_id= 2;
+CREATE TEMPORARY TABLE t3 (a INT PRIMARY KEY);
+INSERT INTO t3 VALUES (10);
+INSERT INTO t3 VALUES (20);
+
+--connection server_1
+INSERT INTO t1 SELECT a, 'server_1' FROM t2;
+
+--connection default
+INSERT INTO t1 SELECT a, 'default' FROM t3;
+
+--connection server_1
+INSERT INTO t1 SELECT a+2, '+server_1' FROM t2;
+
+# Crash the master server in the middle of writing an event group.
+--write_file $MYSQLTEST_VARDIR/tmp/mysqld.1.expect
+wait
+EOF
+
+FLUSH TABLES;
+SET SESSION debug_dbug="+d,crash_before_writing_xid";
+--error 2006,2013
+INSERT INTO t1 SELECT a+4, '++server_1' FROM t2;
+
+--source include/wait_until_disconnected.inc
+--connection default
+--source include/wait_until_disconnected.inc
+
+--append_file $MYSQLTEST_VARDIR/tmp/mysqld.1.expect
+restart
+EOF
+
+--connection default
+--enable_reconnect
+--source include/wait_until_connected_again.inc
+
+--connection server_1
+--enable_reconnect
+--source include/wait_until_connected_again.inc
+
+INSERT INTO t1 VALUES (0, 1);
+#--save_master_pos
+--source include/save_master_gtid.inc
+
+--connection server_2
+# Start the slave replicating the events.
+# The main thing to test here is that the slave will know that it
+# needs to abort the partially received event group, so that the
+# execution of format_description event will not wait infinitely
+# for a commit of the incomplete group that never happens.
+
+--source include/start_slave.inc
+#--sync_with_master
+--source include/sync_with_master_gtid.inc
+
+SELECT * FROM t1 ORDER BY a;
+SHOW STATUS LIKE 'Slave_open_temp_tables';
+
+--connection server_1
+# This FLUSH can be removed once MDEV-6608 is fixed.
+FLUSH LOGS;
+
+
 --connection server_2
 --source include/stop_slave.inc
 SET GLOBAL slave_parallel_threads=@old_parallel_threads;