17.4.3 Upgrading a Replication Setup

When you upgrade servers that participate in a replication setup, the procedure for upgrading depends on the current server versions and the version to which you are upgrading.

This section applies to upgrading replication from older versions of MySQL to MySQL 5.6. A 4.0 server should be 4.0.3 or newer.

When you upgrade a master to 5.6 from an earlier MySQL release series, you should first ensure that all the slaves of this master are using the same 5.6.x release. If this is not the case, you should first upgrade the slaves. To upgrade each slave, shut it down, upgrade it to the appropriate 5.6.x version, restart it, and restart replication. Relay logs created by the slave after the upgrade are in 5.6 format.

Changes affecting operations in strict SQL mode may result in replication failure on an updated slave. For example, as of MySQL 5.6.13, the server restricts insertion of a DEFAULT value of 0 for temporal data types in strict mode (STRICT_TRANS_TABLES or STRICT_ALL_TABLES). A resulting incompatibility for replication if you use statement-based logging (binlog_format=STATEMENT) is that if a slave is upgraded, a nonupgraded master will execute statements without error that may fail on the slave and replication will stop. To deal with this, stop all new statements on the master and wait until the slaves catch up. Then upgrade the slaves. Alternatively, if you cannot stop new statements, temporarily change to row-based logging on the master (binlog_format=ROW) and wait until all slaves have processed all binary logs produced up to the point of this change. Then upgrade the slaves.

After the slaves have been upgraded, shut down the master, upgrade it to the same 5.6.x release as the slaves, and restart it. If you had temporarily changed the master to row-based logging, change it back to statement-based logging. The 5.6 master is able to read the old binary logs written prior to the upgrade and to send them to the 5.6 slaves. The slaves recognize the old format and handle it properly. Binary logs created by the master subsequent to the upgrade are in 5.6 format. These too are recognized by the 5.6 slaves.

In other words, when upgrading to MySQL 5.6, the slaves must be MySQL 5.6 before you can upgrade the master to 5.6. Note that downgrading from 5.6 to older versions does not work so simply: You must ensure that any 5.6 binary log or relay log has been fully processed, so that you can remove it before proceeding with the downgrade.

Downgrading a replication setup to a previous version cannot be done once you have switched from statement-based to row-based replication, and after the first row-based statement has been written to the binlog. See Section 17.1.2, “Replication Formats”.

Some upgrades may require that you drop and re-create database objects when you move from one MySQL series to the next. For example, collation changes might require that table indexes be rebuilt. Such operations, if necessary, will be detailed at Section 2.11.1.3, “Upgrading from MySQL 5.5 to 5.6”. It is safest to perform these operations separately on the slaves and the master, and to disable replication of these operations from the master to the slave. To achieve this, use the following procedure:

  1. Stop all the slaves and upgrade them. Restart them with the --skip-slave-start option so that they do not connect to the master. Perform any table repair or rebuilding operations needed to re-create database objects, such as use of REPAIR TABLE or ALTER TABLE, or dumping and reloading tables or triggers.

  2. Disable the binary log on the master. To do this without restarting the master, execute a SET sql_log_bin = 0 statement. Alternatively, stop the master and restart it without the --log-bin option. If you restart the master, you might also want to disallow client connections. For example, if all clients connect using TCP/IP, use the --skip-networking option when you restart the master.

  3. With the binary log disabled, perform any table repair or rebuilding operations needed to re-create database objects. The binary log must be disabled during this step to prevent these operations from being logged and sent to the slaves later.

  4. Re-enable the binary log on the master. If you set sql_log_bin to 0 earlier, execute a SET sql_log_bin = 1 statement. If you restarted the master to disable the binary log, restart it with --log-bin, and without --skip-networking so that clients and slaves can connect.

  5. Restart the slaves, this time without the --skip-slave-start option.

Replication with global transaction identifiers was introduced in MySQL 5.6.7. If you are upgrading an existing replication setup from a version of MySQL that does not support GTIDs to a version that does, you should not enable GTIDs on either the master or the slave before making sure that the setup meets all the requirements for GTID-based replication. See Section 17.1.3.2, “Setting Up Replication Using GTIDs”, which contains information about converting existing replication setups to use GTID-based replication.