Description
Percona Server provides ability to get replication coordinates with Binlog_snapshot_file and Binlog_snapshot_position.
In order to setup a gtid replication, it's not comfortable to convert (file,pos) to GTID form.
At the same time @@GLOBAL.GTID_EXECUTED is changing during --single-transaction mysqldump backup:
mysql> LOCK TABLES FOR BACKUP;SET SESSION TRANSACTION ISOLATION LEVEL REPEATABLE READ; START TRANSACTION /*!40100 WITH CONSISTENT SNAPSHOT */; SELECT @@GLOBAL.GTID_EXECUTED;select sleep(10);SELECT @@GLOBAL.GTID_EXECUTED;SELECT @@GLOBAL.GTID_EXECUTED; Query OK, 0 rows affected (0.04 sec) Query OK, 0 rows affected (0.00 sec) Query OK, 0 rows affected (0.00 sec) +------------------------------------------------+ | @@GLOBAL.GTID_EXECUTED | +------------------------------------------------+ | 1e7becbc-0dd6-11ea-96b1-0242ac110002:1-4090778 | +------------------------------------------------+ 1 row in set (0.00 sec) +-----------+ | sleep(10) | +-----------+ | 0 | +-----------+ 1 row in set (10.03 sec) +------------------------------------------------+ | @@GLOBAL.GTID_EXECUTED | +------------------------------------------------+ | 1e7becbc-0dd6-11ea-96b1-0242ac110002:1-4128520 | +------------------------------------------------+ 1 row in set (0.00 sec)
The feature breaks SELECT @@GLOBAL.GTID_EXECUTED compare to vanilla mysqldump implementation:
https://github.com/mysql/mysql-server/blob/8.0/client/mysqldump.cc#L5651-L5688
https://github.com/percona/percona-server/blob/8.0/client/mysqldump.cc#L6477-L6522
In percona server's mysqldump:
if (opt_single_transaction && opt_master_data) { /* See if we can avoid FLUSH TABLES WITH READ LOCK with Binlog_snapshot_* variables. */ consistent_binlog_pos = check_consistent_binlog_pos(nullptr, nullptr); } if ((opt_lock_all_tables || (opt_master_data && !consistent_binlog_pos) || (opt_single_transaction && flush_logs))) { if (do_flush_tables_read_lock(mysql)) goto err; } else if (opt_lock_for_backup && do_lock_tables_for_backup(mysql)) goto err; if (opt_master_data && do_show_master_status(mysql, consistent_binlog_pos)) goto err; if (opt_slave_data && do_show_slave_status(mysql)) goto err; if (opt_single_transaction && (!opt_lock_for_backup || opt_master_data) && do_unlock_tables(mysql)) /* unlock but no commit! */ goto err;
As you can see "FLUSH TABLES WITH READ LOCK" was disabled for (opt_single_transaction | consistent_binlog_pos), but unlock tables called only for (!opt_lock_for_backup || opt_master_data).
Until GTID_EXECUTED will be covered by Binlog_snapshot_GTID, is possible to change
https://github.com/percona/percona-server/blob/8.0/client/mysqldump.cc#L6477
to
if ((opt_lock_all_tables || (opt_master_data && !(consistent_binlog_pos && opt_lock_for_backup)) || (opt_single_transaction && flush_logs))) { if (do_flush_tables_read_lock(mysql)) goto err; } else if (opt_lock_for_backup && do_lock_tables_for_backup(mysql)) goto err;
This allows to have upstream mysqldump behavior for setting GTID_EXECUTED with single-transaction, master-data and --skip-lock-for-backup
Attachments
Issue Links
- relates to
-
PS-4464 Expose the last global transaction identifier (GTID) executed for a CONSISTENT SNAPSHOT.
-
- Done
-
