Binary logging can be row-based, statement-based, or a mix of the two. See Binary Log Formats for more details on the formats. If logging is statement-based, it is possible that a statement will have different effects on the master and on the slave.
Stored routines are particularly prone to this, for two main reasons:
- stored routines can be non-deterministic, in other words non-repeatable, and therefore have different results each time they are run.
- the slave thread executing the stored routine on the slave holds full privileges, while this may not be the case when the routine was run on the master.
The problems with replication will only occur with statement-based logging. If row-based logging is used, since changes are made to rows based on the master's rows, there is no possibility of the slave and master getting out of sync.
By default, with row-based replication, triggers run on the master, and the effects of their executions are replicated to the slaves. However, starting from MariaDB 10.1.1, it is possible to run triggers on the slaves. See Running triggers on the slave for Row-based events.
How MariaDB handles statement-based binary logging of routines
The following only applies if statement-based binary logging is enabled. If binary logging is not enabled, or the logging format is either row-based or mixed, these limitations do not apply.
- If binary logging is enabled, when a stored function is created, it must be declared as either
READS SQL DATA, or else an error will occur.
- MariaDB cannot check whether a function is deterministic, and relies on the correct definition being used.
- To create or modify a stored function, a user requires the SUPER privilege as well as the regular privileges. See Stored Routine Privileges for these details.
- The above conditions are not required if the server variable log_bin_trust_function_creators is set to
1- by default it is set to
- Triggers work in the same way, except that they are always assumed to be deterministic for logging purposes, even if this is obviously not the case, such as when they use the UUID function.
- Triggers can also update data. The slave uses the DEFINER attribute to determine which user is taken to have created the trigger.
- Note that the above limitations do no apply to stored procedures or to events.
A deterministic function:
CREATE FUNCTION trust_me(x INT) RETURNS INT DETERMINISTIC READS SQL DATA BEGIN RETURN x; END;
A non-deterministic function, since it uses the UUID_SHORT function:
CREATE FUNCTION dont_trust_me() RETURNS INT BEGIN RETURN UUID_SHORT(); END;