3.9.1. Migrating from MySQL Native Replication 'In-Place'

If you are migrating an existing MySQL native replication deployment to use Tungsten Cluster the configuration of the Tungsten Cluster replication must be updated to match the status of the slave.

  1. Deploy Tungsten Cluster using the model or system appropriate according to Chapter 2, Deployment. Ensure that the Tungsten Cluster is not started automatically by excluding the --start or --start-and-report options from the tpm commands.

  2. On each slave

    Confirm that native replication is working on all slave nodes :

    shell> echo 'SHOW SLAVE STATUS\G' | tpm mysql | \
    egrep ' Master_Host| Last_Error| Slave_SQL_Running' 
                      Master_Host: tr-ssl1
                Slave_SQL_Running: Yes

  3. On the master and each slave

    Reset the Tungsten Replicator position on all servers :

    shell> replicator start offline
    shell> trepctl -service alpha reset -all -y

  4. On the master

    Login and start Tungsten Cluster services and put the Tungsten Replicator online:

    shell> startall
    shell> trepctl online
  5. On the master

    Put the cluster into maintenance mode using cctrl to prevent Tungsten Cluster automatically reconfiguring services:

    cctrl > set policy maintenance
  6. On each slave

    Record the current slave log position (as reported by the Master_Log_File and Exec_Master_Log_Pos output from SHOW SLAVE STATUS. Ideally, each slave should be stopped at the same position:

    shell> echo 'SHOW SLAVE STATUS\G' | tpm mysql | \
    egrep ' Master_Host| Last_Error| Master_Log_File| Exec_Master_Log_Pos' 
                      Master_Host: tr-ssl1
                  Master_Log_File: mysql-bin.000025
                       Last_Error: Error executing row event: 'Table 'tungsten_alpha.heartbeat' doesn't exist'
              Exec_Master_Log_Pos: 181268

    If you have multiple slaves configured to read from this master, record the slave position individually for each host. Once you have the information for all the hosts, determine the earliest log file and log position across all the slaves, as this information will be needed when starting Tungsten Cluster replication. If one of the servers does not show an error, it may be replicating from an intermediate server. If so, you can proceed normally and assume this server stopped at the same position as the host is replicating from.

  7. On the master

    Take the replicator offline and clear the THL:

    shell> trepctl offline
    shell> trepctl -service alpha reset -all -y
  8. On the master

    Start replication, using the lowest binary log file and log position from the slave information determined in step 6.

    shell> trepctl online -from-event 000025:181268

    Tungsten Replicator will start reading the MySQL binary log from this position, creating the corresponding THL event data.

  9. On each slave

    1. Disable native replication to prevent native replication being accidentally started on the slave.

      On MySQL 5.0 or MySQL 5.1:

      shell> echo "STOP SLAVE; CHANGE MASTER TO MASTER_HOST='';" | tpm mysql

      On MySQL 5.5 or later:

      shell> echo "STOP SLAVE; RESET SLAVE ALL;" | tpm mysql
    2. If the final position of MySQL replication matches the lowest across all slaves, start Tungsten Cluster services :

      shell> trepctl online
      shell> startall

      The slave will start reading from the binary log position configured on the master.

      If the position on this slave is different, use trepctl online -from-event to set the online position according to the recorded position when native MySQL was disabled. Then start all remaining services with startall.

      shell> trepctl online -from-event 000025:188249
      shell> startall

  10. Use cctrl to confirm that replication is operating correctly across the dataservice on all hosts.

  11. Put the cluster back into automatic mode:

    cctrl> set policy automatic
  12. Update your applications to use the installed connector services rather than a direct connection.

  13. Remove the master.info file on each slave to ensure that when a slave restarts, it does not connect up to the master MySQL server again.

Once these steps have been completed, Tungsten Cluster should be operating as the replication service for your MySQL servers. Use the information in Chapter 5, Operations Guide to monitor and administer the service.