3.7 Backing Up and Restoring MySQL Cluster Manager Agents

This section explains how to back up configuration data for mcmd agents and how to restore the backed-up agent data. Used together with the backup cluster command, the backup agents command allows you to backup and restore a complete cluster-plus-manager setup.

If no host names are given with the backup agents command, backups are created for all agents of the site:

mcm> backup agents mysite;
+-----------------------------------+
| Command result                    |
+-----------------------------------+
| Agent backup created successfully |
+-----------------------------------+
1 row in set (0.07 sec)

To backup one or more specific agents, specify them with the --hosts option:

mcm> backup agents --hosts=tonfisk mysite;
+-----------------------------------+
| Command result                    |
+-----------------------------------+
| Agent backup created successfully |
+-----------------------------------+
1 row in set (0.07 sec)

If no site name is given, only the agent that the mcm client is connected to is backed up.

The backup for each agent includes the following contents from the agent repository (mcm_data folder):

• The rep subfolder

• The metadata files high_water_mark and repchksum

The repository is locked while the backup are in progress, to avoid creating an inconsistent backup. The backup for each agent is created in a subfolder named rep_backup/timestamp under the agent's mcm_data folder, with timestamp reflecting the time the backup began. If you want the backup to be at another place, create a soft link from mcm_data/rep_backup to your desired storage location.

For MySQL Cluster Manager 1.4.6 and later: You can list agent backups using the list backups command with the --agent option and the site name:

mcm> list backups --agent mysite;
+------------+-------+---------+----------------------+---------+
| BackupId   | Agent | Host    | Timestamp            | Comment |
+------------+-------+---------+----------------------+---------+
| 1522914101 | 0     | tonfisk | 2018-04-05 07:41:41Z |         |
| 1522914105 | 0     | tonfisk | 2018-04-05 07:41:45Z |         |
| 1522914121 | 0     | tonfisk | 2018-04-05 07:42:01Z |         |
+------------+-------+---------+----------------------+---------+
3 rows in set (0.00 sec)

To restore the backup for an agent:

• Wipe the contents of the agent's mcm_data/rep folder

• Delete the metadata files high_water_mark and repchksum from the mcm_data folder

• Copy the contents in the mcm_data/rep_backup/timestamp/rep folder back into the mcm_data/rep folder

• Copy the metadata files high_water_mark and repchksum from the mcm_data/rep_backup/timestamp folder back into the mcm_data folder

• Restart the agent

The steps are illustrated below:

mysql@tonfisk$ cd mcm_data

mysql@tonfisk$ cp mcm_data/rep_backup/timestamp/rep/* ./rep/

mysql@tonfisk$ cp mcm_data/rep_backup/timestamp/high_water_mark ./

mysql@tonfisk$ cp mcm_data/rep_backup/timestamp/repchksum ./

mysql@tonfisk$ mcm1.4.8/bin/mcmd

The backup may be manually restored on just one, or more than one agents. If backup is restored for only one agent on, say, host A, host A will contact the other agents of the site to make them recover their repositories from host A using the usual mechanism for agent recovery. If all agents on all hosts are restored and restarted manually, the situation will be similar to the normal restarting all agents after stopping them at slightly different points in time.

If configuration changes has been made to the cluster since the restored backup was created, the same changes must be made again after the agent restores have been completed, to ensure that the agents' configurations match those of the actual running cluster. For example: sometime after a backup was done, a set MaxNoOfTables:ndbmtd=500 mycluster command was issued and soon afterward, something happened and corrupted the agent repository; after the agent backup was restored, the same set command has to be rerun in order to update the mcmd agents' configurations. While the command does not effectively change anything on the cluster itself, after it has been run, a rolling restart of the cluster processes using the restart cluster command is still required.