DRAFT: MDEV-36025: Server start fails on mariabackup with prepared tr…

mysql-test/suite/mariabackup/slave_provision_prepared_trx.cnf

@@ -0,0 +1,14 @@
+[mysqld.1]
+log-slave-updates
+loose-innodb
+
+[mysqld.2]
+log-slave-updates
+loose-innodb
+skip-slave-start
+
+[ENV]
+SERVER_MYPORT_1=		@mysqld.1.port
+SERVER_MYSOCK_1=		@mysqld.1.socket
+SERVER_MYPORT_2=		@mysqld.2.port
+SERVER_MYSOCK_2=		@mysqld.2.socket

mysql-test/suite/mariabackup/slave_provision_prepared_trx.result

@@ -0,0 +1,70 @@
+### MDEV-36025 backup taken from a slave with optimistic parallel replication fails to restore most of the time
+### Test provisioning a slave from an existing server running optimistic
+### parallel replication so that prepared transactions are left in the
+### backup when the restored server is started.
+RESET MASTER;
+CREATE TABLE t1(a INT PRIMARY KEY, b VARCHAR(60)) ENGINE INNODB;
+INSERT INTO t1 VALUES(1, NULL);
+INSERT INTO t1 VALUES(2, NULL);
+INSERT INTO t1 VALUES(3, NULL);
+connect  server2,127.0.0.1,root,,,$SERVER_MYPORT_2;
+SET STATEMENT sql_log_bin=0 FOR ALTER TABLE mysql.gtid_slave_pos ENGINE=InnoDB;
+RESET MASTER;
+SET GLOBAL gtid_slave_pos= '';
+CHANGE MASTER TO
+master_port=PORT, master_host='127.0.0.1', master_user='root',
+master_use_gtid= slave_pos;
+SET GLOBAL slave_parallel_threads=3;
+SET GLOBAL slave_parallel_mode=optimistic;
+START SLAVE;
+connection default;
+include/save_master_gtid.inc
+connection server2;
+include/sync_with_master_gtid.inc
+connect  server2a,127.0.0.1,root,,,$SERVER_MYPORT_2;
+BEGIN;
+SELECT * FROM t1 WHERE a=1 FOR UPDATE;
+a	b
+1	NULL
+connection default;
+UPDATE t1 SET b='xyzzy' WHERE a=1;
+INSERT INTO t1 VALUES (4, NULL);
+UPDATE t1 SET b='foobar' WHERE a=2;
+include/save_master_gtid.inc
+*** Doing backup...
+*** Doing prepare...
+connection server2a;
+ROLLBACK;
+include/sync_with_master_gtid.inc
+connection server2;
+*** Provision a new slave from the backup
+*** Stopping provisioned server
+*** Removing old datadir for provisioned server
+*** Provision new server from backup
+disconnect server2a;
+# restart
+CHANGE MASTER TO
+master_port=PORT, master_host='127.0.0.1', master_user='root',
+master_use_gtid= slave_pos;
+START SLAVE;
+connection default;
+include/save_master_gtid.inc
+SELECT * FROM t1 ORDER BY a;
+a	b
+1	xyzzy
+2	foobar
+3	NULL
+4	NULL
+connection server2;
+include/sync_with_master_gtid.inc
+SELECT * FROM t1 ORDER BY a;
+a	b
+1	xyzzy
+2	foobar
+3	NULL
+4	NULL
+STOP SLAVE;
+RESET SLAVE ALL;
+DROP TABLE t1;
+connection default;
+DROP TABLE t1;

mysql-test/suite/mariabackup/slave_provision_prepared_trx.test

@@ -0,0 +1,99 @@
+--source include/have_innodb.inc
+--source include/have_log_bin.inc
+
+--echo ### MDEV-36025 backup taken from a slave with optimistic parallel replication fails to restore most of the time
+--echo ### Test provisioning a slave from an existing server running optimistic
+--echo ### parallel replication so that prepared transactions are left in the
+--echo ### backup when the restored server is started.
+
+
+RESET MASTER;
+CREATE TABLE t1(a INT PRIMARY KEY, b VARCHAR(60)) ENGINE INNODB;
+INSERT INTO t1 VALUES(1, NULL);
+INSERT INTO t1 VALUES(2, NULL);
+INSERT INTO t1 VALUES(3, NULL);
+
+--connect (server2,127.0.0.1,root,,,$SERVER_MYPORT_2)
+let MYSQLD_DATADIR=`select @@datadir`;
+let $basedir=$MYSQLTEST_VARDIR/tmp/backup;
+SET STATEMENT sql_log_bin=0 FOR ALTER TABLE mysql.gtid_slave_pos ENGINE=InnoDB;
+RESET MASTER;
+SET GLOBAL gtid_slave_pos= '';
+
+--replace_result $SERVER_MYPORT_1 PORT
+eval CHANGE MASTER TO
+  master_port=$SERVER_MYPORT_1, master_host='127.0.0.1', master_user='root',
+  master_use_gtid= slave_pos;
+SET GLOBAL slave_parallel_threads=3;
+SET GLOBAL slave_parallel_mode=optimistic;
+START SLAVE;
+
+--connection default
+--source include/save_master_gtid.inc
+--connection server2
+--source include/sync_with_master_gtid.inc
+
+# Block transaction T1 from committing on the slave, forcing a following
+# T2 to remain in "prepared" state when the backup is taken.
+--connect (server2a,127.0.0.1,root,,,$SERVER_MYPORT_2)
+BEGIN;
+SELECT * FROM t1 WHERE a=1 FOR UPDATE;
+
+--connection default
+UPDATE t1 SET b='xyzzy' WHERE a=1;
+INSERT INTO t1 VALUES (4, NULL);
+UPDATE t1 SET b='foobar' WHERE a=2;
+--source include/save_master_gtid.inc
+
+# Sleep a small amount to give a higher chance for the transactions to reach
+# their prepared state when the backup runs.
+--sleep 0.3
+
+--echo *** Doing backup...
+--exec $XTRABACKUP --defaults-file=$MYSQLTEST_VARDIR/my.cnf --socket=$SERVER_MYSOCK_2 --backup --target-dir=$basedir
+--echo *** Doing prepare...
+--exec $XTRABACKUP --prepare --binlog-info=1 --target-dir=$basedir
+
+--connection server2a
+ROLLBACK;
+--source include/sync_with_master_gtid.inc
+
+--connection server2
+--echo *** Provision a new slave from the backup
+
+--echo *** Stopping provisioned server
+--source include/shutdown_mysqld.inc
+
+--echo *** Removing old datadir for provisioned server
+--rmdir $MYSQLD_DATADIR
+
+--echo *** Provision new server from backup
+--exec $XTRABACKUP --copy-back --datadir=$MYSQLD_DATADIR --target-dir=$basedir
+
+--disconnect server2a
+--source include/start_mysqld.inc
+
+--replace_result $SERVER_MYPORT_1 PORT
+eval CHANGE MASTER TO
+  master_port=$SERVER_MYPORT_1, master_host='127.0.0.1', master_user='root',
+  master_use_gtid= slave_pos;
+# The slave GTID position was saved transactionally to mysql.gtid_slave_pos
+# as part of the backup.
+START SLAVE;
+
+--connection default
+--source include/save_master_gtid.inc
+SELECT * FROM t1 ORDER BY a;
+
+--connection server2
+--source include/sync_with_master_gtid.inc
+SELECT * FROM t1 ORDER BY a;
+
+STOP SLAVE;
+RESET SLAVE ALL;
+DROP TABLE t1;
+
+--connection default
+DROP TABLE t1;
+
+rmdir $basedir;

sql/log.cc

@@ -10039,7 +10039,8 @@ int TC_LOG_MMAP::open(const char *opt_name)
     inited= 1;
     crashed= TRUE;
     sql_print_information("Recovering after a crash using %s", opt_name);
-    if (tc_heuristic_recover)
+    if (tc_heuristic_recover &&
+        tc_heuristic_recover != TC_HEURISTIC_RECOVER_AUTO)
     {
       sql_print_error("Cannot perform automatic crash recovery when "
                       "--tc-heuristic-recover is used");
@@ -10562,7 +10563,8 @@ TC_LOG_MMAP  tc_log_mmap;
 
 int TC_LOG::using_heuristic_recover()
 {
-  if (!tc_heuristic_recover)
+  if (!tc_heuristic_recover ||
+      tc_heuristic_recover == TC_HEURISTIC_RECOVER_AUTO)
     return 0;
 
   sql_print_information("Heuristic crash recovery mode");

sql/mysqld.cc

@@ -294,7 +294,7 @@ const char *show_comp_option_name[]= {"YES", "NO", "DISABLED"};
 
 static const char *tc_heuristic_recover_names[]=
 {
-  "OFF", "COMMIT", "ROLLBACK", NullS
+  "OFF", "COMMIT", "ROLLBACK", "AUTO", NullS
 };
 static TYPELIB tc_heuristic_recover_typelib=
 {
@@ -7027,7 +7027,7 @@ struct my_option my_long_options[]=
   {"tc-heuristic-recover", 0,
    "Decision to use in heuristic recover process",
    &tc_heuristic_recover, &tc_heuristic_recover,
-   &tc_heuristic_recover_typelib, GET_ENUM, REQUIRED_ARG, 0, 0, 0, 0, 0, 0},
+   &tc_heuristic_recover_typelib, GET_ENUM, REQUIRED_ARG, 3, 0, 0, 0, 0, 0},
   {"temp-pool", 0,
 #if (ENABLE_TEMP_POOL)
    "Using this option will cause most temporary files created to use a small "

sql/sql_class.h

@@ -280,6 +280,7 @@ class Recreate_info
 
 #define TC_HEURISTIC_RECOVER_COMMIT   1
 #define TC_HEURISTIC_RECOVER_ROLLBACK 2
+#define TC_HEURISTIC_RECOVER_AUTO     3
 extern ulong tc_heuristic_recover;
 
 typedef struct st_user_var_events