Advantages of statement-based replication

Disadvantages of statement-based replication

Advantages of row-based replication

Disadvantages of row-based replication

Proven technology.

Less data written to log files. When updates or deletes affect many rows, this results in much less storage space required for log files. This also means that taking and restoring from backups can be accomplished more quickly.

Log files contain all statements that made any changes, so they can be used to audit the database.

Statements that are unsafe for SBR.  Not all statements which modify data (such as INSERT DELETE, UPDATE, and REPLACE statements) can be replicated using statement-based replication. Any nondeterministic behavior is difficult to replicate when using statement-based replication. Examples of such Data Modification Language (DML) statements include the following:

Statements that cannot be replicated correctly using statement-based replication are logged with a warning like the one shown here:

[Warning] Statement is not safe to log in statement format.

A similar warning is also issued to the client in such cases. The client can display it using SHOW WARNINGS.

INSERT ... SELECT requires a greater number of row-level locks than with row-based replication.

UPDATE statements that require a table scan (because no index is used in the WHERE clause) must lock a greater number of rows than with row-based replication.

For InnoDB: An INSERT statement that uses AUTO_INCREMENT blocks other nonconflicting INSERT statements.

For complex statements, the statement must be evaluated and executed on the slave before the rows are updated or inserted. With row-based replication, the slave only has to modify the affected rows, not execute the full statement.

If there is an error in evaluation on the slave, particularly when executing complex statements, statement-based replication may slowly increase the margin of error across the affected rows over time. See Section 17.4.1.28, “Slave Errors During Replication”.

Stored functions execute with the same NOW() value as the calling statement. However, this is not true of stored procedures.

Deterministic UDFs must be applied on the slaves.

Table definitions must be (nearly) identical on master and slave. See Section 17.4.1.9, “Replication with Differing Table Definitions on Master and Slave”, for more information.

All changes can be replicated. This is the safest form of replication.

Fewer row locks are required on the master, which thus achieves higher concurrency, for the following types of statements:

Fewer row locks are required on the slave for any INSERT, UPDATE, or DELETE statement.

RBR can generate more data that must be logged. To replicate a DML statement (such as an UPDATE or DELETE statement), statement-based replication writes only the statement to the binary log. By contrast, row-based replication writes each changed row to the binary log. If the statement changes many rows, row-based replication may write significantly more data to the binary log; this is true even for statements that are rolled back. This also means that making and restoring a backup can require more time. In addition, the binary log is locked for a longer time to write the data, which may cause concurrency problems. Use binlog_row_image=minimal to reduce the disadvantage considerably.

Deterministic UDFs that generate large BLOB values take longer to replicate with row-based replication than with statement-based replication. This is because the BLOB column value is logged, rather than the statement generating the data.

You cannot see on the slave what statements were received from the master and executed. However, you can see what data was changed using mysqlbinlog with the options --base64-output=DECODE-ROWS and --verbose.

Alternatively, use the binlog_rows_query_log_events variable, which if enabled adds a Rows_query event with the statement to mysqlbinlog output when the -vv option is used.

For tables using the MyISAM storage engine, a stronger lock is required on the slave for INSERT statements when applying them as row-based events to the binary log than when applying them as statements. This means that concurrent inserts on MyISAM tables are not supported when using row-based replication.