What You Should Know About MySQL Replication Types

For companies that live and die by their databases, a one-shot backup isn’t really the perfect solution. Typically for such companies (think Yahoo! or Google), database access is a near-constant process, and database content changes continually, often on a second-by-second basis. Data replication, which involves continual data transfer between two (or more) servers to maintain a replica of the original database, is a better backup solution for these situations.

MySQL supports two (or three, depending on how you look at it) different methods of replicating databases from master to slave. All of these methods use the binary log; however, they differ in the type of data that is written to the master’s binary log.

• Statement-based replication Under this method, the binary log stores the SQL statements used to change databases on the master server. The slave reads this data and reexecutes these SQL statements to produce a copy of the master database. This is the default replication method in MySQL 5.1.11 and earlier and MySQL 5.1.29 onwards.
• Row-based replication Under this method, the binary log stores the record-level changes that occur to database tables on the master server. The slave reads this data and manipulates its records accordingly to produce a copy of the master database.
• Mixed-format replication Under this method, the server can dynamically choose between statement-based replication and row-based replication, depending on certain conditions. Some of these conditions include using a user-defined function (UDF), using an INSERT command with the DELAYED clause, using temporary tables, or using a statement that uses system variables. This is the default replication method in MySQL 5.1.12 to MySQL 5.1.28.

If you’re unsure which replication method to use and your replication needs aren’t complex, it’s best to stick to statement-based replication, as it’s been around longest and therefore has had the most time to have its kinks worked out. That said, certain types of statements cannot be replicated using this method, and it also tends to require a higher number of table locks. Row-based replication is useful for these situations. Because it replicates changes to rows, any change can be replicated, and it also requires fewer table locks.

The replication method currently in use on the server is listed in the binlog_format server variable.

mysql> SHOW VARIABLES LIKE 'binlog_format';

1 row in set (0.08 sec)

To alter the replication method, set a new value for this variable, as shown, using the SET command with either GLOBAL or SESSION scope. Note that using GLOBAL scope requires a server restart for the change in method to take effect.

mysql> SET binlog_format = 'MIXED';
Query OK, 0 rows affected (0.02 sec)
mysql> SELECT @@SESSION.binlog_format;
+-------------------------+
| @@SESSION.binlog_format |
+-------------------------+
| MIXED                   |
+-------------------------+
1 row in set (0.00 sec)

mysql> SET GLOBAL binlog_format = 'ROW';
Query OK, 0 rows affected (0.00 sec)
mysql> SELECT @@GLOBAL.binlog_format;;
+------------------------+
| @@GLOBAL.binlog_format |
+------------------------+
| ROW                    |
+------------------------+
1 row in set (0.00 sec)