The terms master and slave have historically been used in replication, but the terms terms primary and replica are now preferred. The old terms are used throughout the documentation, and in MariaDB commands, although MariaDB 10.5 has begun the process of renaming. The documentation will follow over time. See MDEV-18777 to follow progress on this effort.
Multi-source replication means that one server has many primaries from which it replicates.
You specify which primary connection you want to work with by either specifying the connection name in the command or setting default_master_connection to the connection you want to work with.
The connection name may include any characters and should be less than 64 characters. Connection names are compared without regard to case (case insensitive). You should preferably keep the connection name short as it will be used as a suffix for relay logs and primary info index files.
The new syntax introduced to handle many connections:
CHANGE MASTER ['connection_name'] TO .... This creates or modifies a connection to a primary.
FLUSH RELAY LOGS ['connection_name']
RESET SLAVE ['connection_name'] [ALL]. This is used to reset a replica's replication position or to remove a replica permanently.
SHOW RELAYLOG ['connection_name'] EVENTS
SHOW SLAVE ['connection_name'] STATUS
SHOW ALL SLAVES STATUS
START SLAVE ['connection_name'...]]
START ALL SLAVES ...
STOP SLAVE ['connection_name'] ...
STOP ALL SLAVES ...
The original old-style connection is an empty string
You don't have to use this connection if you don't want to.
Replication Variables for Multi-Source
The new replication variable default_master_connection
specifies which connection will be used for commands and variables if you don't
specify a connection. By default this is
'' (the default
The following replication variables are local for the connection. (In other
words, they show the value for the
@@default_master_connection connection). We are working on
making all the important ones local for the connection.
|Variable||max_relay_log_size||Max size of relay log. Is set at startup to |
|Variable||replicate_do_db||Tell the replica to restrict replication to updates of tables whose names appear in the comma-separated list. For statement-based replication, only the default database (that is, the one selected by USE) is considered, not any explicitly mentioned tables in the query. For row-based replication, the actual names of table(s) being updated are checked.|
|Variable||replicate_do_table||Tells the replica to restrict replication to tables in the comma-separated list|
|Variable||replicate_ignore_db||Tell the replica to restrict replication to updates of tables whose names do not appear in the comma-separated list. For statement-based replication, only the default database (that is, the one selected by USE) is considered, not any explicitly mentioned tables in the query. For row-based replication, the actual names of table(s) being updated are checked.|
|Variable||replicate_ignore_table||Tells the replica thread to not replicate any statement that updates the specified table, even if any other tables might be updated by the same statement.|
|Variable||replicate_wild_do_table||Tells the replica thread to restrict replication to statements where any of the updated tables match the specified database and table name patterns.|
|Variable||replicate_wild_ignore_table||Tells the replica thread to not replicate to the tables that match the given wildcard pattern.|
|Status||Slave_heartbeat_period||How often to request a heartbeat packet from the primary (in seconds).|
|Status||Slave_received_heartbeats||How many heartbeats we have got from the primary.|
|Status||Slave_running||Shows if the replica is running. YES means that the sql thread and the IO thread are active. No means either one is not running. '' means that |
|Variable||Sql_slave_skip_counter||How many entries in the replication log that should be skipped (mainly used in case of errors in the log).|
You can access all of the above variables with either
Note that in contrast to MySQL, all variables always show the correct active value!
set @@default_master_connection=''; show status like 'Slave_running'; set @@default_master_connection='other_connection'; show status like 'Slave_running';
@@default_master_connection contains a non existing name,
you will get a warning.
All other primary-related variables are global and affect either only the '' connections or all connections. For example, Slave_retried_transactions now shows the total number of retried transactions over all replicas.
If you need to set gtid_slave_pos you need to set this for all primaries at the same time.
New status variables:
|Number of executed |
|Number of executed |
|Number of executed |
|Number of executed |
SHOW ALL SLAVES STATUS has the following new columns:
|Name of the primary connection. This is the first variable.|
|State of SQL thread.|
|Number of retried transactions for this connection.|
|Max relay log size for this connection.|
|How many log entries the replica has executed.|
|How many heartbeats we have got from the primary.|
|How often to request a heartbeat packet from the primary (in seconds).|
The basic principle of the new files used by multi source replication is that
they have the same name as the original relay log files suffixed with
connection_name before the extension. The main exception is
the file that holds all connection is named as the normal
master-info-file with a
When you are using multi source, the following new files are created:
|Contains the current primary position for what's applied to in the replica. Extension is normally |
|The relay-log name with a connection_name suffix. The xxxxx is the relay log number. This contains the replication data read from the primary.|
| Contains the name of the active |
|Contains the current primary position for the relay log. Extension is normally |
When creating the file, the connection name is converted to lower case and all special characters in the connection name are converted, the same way as MySQL table names are converted. This is done to make the file name portable across different systems.
Instead of specifying names for
mysqld with --relay-log, --relay-log-index, --general-log-file, --slow-query-log-file,
--log-bin and --log-bin-index, you can just
specify --log-basename and all the other variables are set
with this as a prefix.
- All error messages from a replica with a connection name, that are written to the error log, are prefixed with
Master 'connection_name':. This makes it easy to see from where an error originated.
WARN_NO_MASTER_INFOnow includes connection_name.
- There is no conflict resolution. The assumption is that there are no conflicts in data between the different primaries.
- All executed commands are stored in the normal binary log (nothing new here).
- If the server variable
log_warnings> 1 then you will get some information in the log about how the multi-master-info file is updated (mainly for debugging).
- SHOW [FULL] SLAVE STATUS has one line per connection and more columns than before. Note that the first column is the
RESET SLAVEnow deletes all relay-log files.
- One can set the values for the
replicate-...variables from the command line or in
my.cnffor a given connection by prefixing the variable with the connection name.
- If one doesn't use any connection name prefix for a
replicate..variable, then the value will be used as the default value for all connections that don't have a value set for this variable.
mysqld --main_connection.replicate_do_db=main_database --replicate_do_db=other_database
The have sets the
replicate_do_db variable to
main_database for the connection named
main_connection. All other connections will use the value
One can also use this syntax to set
replicate-rewrite-db for a given connection.
Typical Use Cases
- You are partitioning your data over many primaries and would like to get it all together on one machine to do analytical queries on all data.
- You have many databases spread over many MariaDB/MySQL servers and would like to have all of them on one machine as an extra backup.
- In a Galera cluster the default replication filter rules like
replicate-do-dbdo not apply to replication connections, but also to Galera write set applier threads. By using a named multi-primary replication connection instead, even when replicating from just one primary into the cluster, the primary-replica replication rules can be kept separate from the Galera intra-node replication traffic.
- Each active connection will create 2 threads (as is normal for MariaDB replication).
- You should ensure that all primaries have different
server-id's. If you don't do this, you will get into trouble if you try to replicate from the multi-source replica back to your primaries.
- One can change max_relay_log_size for any active connection, but new connections will always use the server startup value for
max_relay_log_size, which can't be changed at runtime.
- Option innodb-recovery-update-relay-log (xtradb feature to store and restore relay log position for replicas) only works for the default connection ''. As this option is not really safe and can easily cause loss of data if you use storage engines other than InnoDB, we don't recommend this option be used.
- slave_net_timeout affects all connections. We don't check anymore if it's less than Slave_heartbeat_period, as this doesn't make sense in a multi-source setup.
Incompatibilities with MariaDB/MySQL 5.5
- max_relay_log_size is now (almost) a normal variable and not automatically changed if max_binlog_size is changed. To keep things compatible with old config files, we set it to
max_binlog_sizeat startup if its value is 0.
- You can now access replication variables that depend on the active connection with either
- We only write information about relay log positions for recovery if innodb-recovery-update-relay-log is set.
- Slave_retried_transactions now shows the total count of retried transactions over all replicas.
- The status variable
Com_slave_startis replaced with Com_start_slave.
- The status variable
Com_slave_stopis replaced with Com_stop_slave.
FLUSH RELAY LOGSare not replicated anymore. This is not safe as connection names may be different on the replica.