Percona TokuBackup is an open-source hot backup utility for MySQL servers running the TokuDB storage engine (including Percona Server for MySQL and MariaDB). It does not lock your database during backup. The TokuBackup library intercepts system calls that write files and duplicates the writes to the backup directory.
This feature is currently considered Experimental
The installation of TokuBackup can be performed with the ps-admin script.
To install Percona TokuBackup complete the following steps. Run the following commands as root or by using the sudo command.
Run ps-admin --enable-tokubackup to add the
preload-hotbackupoption into [mysqld_safe] section of
Checking SELinux status... INFO: SELinux is disabled. Checking if preload-hotbackup option is already set in config file... INFO: Option preload-hotbackup is not set in the config file. Checking TokuBackup plugin status... INFO: TokuBackup plugin is not installed. Adding preload-hotbackup option into /etc/my.cnf INFO: Successfully added preload-hotbackup option into /etc/my.cnf PLEASE RESTART MYSQL SERVICE AND RUN THIS SCRIPT AGAIN TO FINISH INSTALLATION!
Restart mysql service:
service mysql restart
Run ps-admin --enable-tokubackup again to finish the installation of the TokuBackup plugin.
Checking SELinux status... INFO: SELinux is disabled. Checking if preload-hotbackup option is already set in config file... INFO: Option preload-hotbackup is set in the config file. Checking TokuBackup plugin status... INFO: TokuBackup plugin is not installed. Checking if Percona Server is running with libHotBackup.so preloaded... INFO: Percona Server is running with libHotBackup.so preloaded. Installing TokuBackup plugin... INFO: Successfully installed TokuBackup plugin.
To run Percona TokuBackup, the backup destination directory must
exist, be writable and owned by the same user under which MySQL
server is running (usually
mysql) and empty.
Once this directory is created, the backup can be run using the following command:
mysql> set tokudb_backup_dir='/path_to_empty_directory';
tokudb_backup_dir variable automatically
starts the backup process to the specified directory. Percona
TokuBackup will take full backup each time, currently there is no
incremental backup option
If you get any error on this step (e.g. caused by some misconfiguration), the Reporting Errors section explains how to find out the reason.
Percona TokuBackup does not have any functionality for restoring a backup. You can use rsync or cp to restore the files. You should check that the restored files have the correct ownership and permissions.
Make sure that the datadir is empty and that MySQL server is shut down before restoring from backup. You can’t restore to a datadir of a running mysqld instance (except when importing a partial backup).
The following example shows how you might use the rsync command to restore the backup:
$ rsync -avrP /data/backup/ /var/lib/mysql/
Since attributes of files are preserved, in most cases you will need to change their ownership to mysql before starting the database server. Otherwise, the files will be owned by the user who created the backup.
$ chown -R mysql:mysql /var/lib/mysql
If you have changed default TokuDB data directory (
tokudb_data_dir) or TokuDB log directory (
tokudb_log_dir) or both of them, you will see separate folders for each setting in backup directory after taking backup. You’ll need to restore each folder separately:
$ rsync -avrP /data/backup/mysql_data_dir/ /var/lib/mysql/ $ rsync -avrP /data/backup/tokudb_data_dir/ /path/to/original/tokudb_data_dir/ $ rsync -avrP /data/backup/tokudb_log_dir/ /path/to/original/tokudb_log_dir/ $ chown -R mysql:mysql /var/lib/mysql $ chown -R mysql:mysql /path/to/original/tokudb_data_dir $ chown -R mysql:mysql /path/to/original/tokudb_log_dir
TokuBackup updates the PROCESSLIST state while the backup is in progress. You can see the output by running
SHOW PROCESSLIST or
SHOW FULL PROCESSLIST.
You can exclude certain files and directories based on a regular expression set in the
tokudb_backup_exclude session variable. If the source file name matches the excluded regular expression, then the source file is excluded from backup.
For example, to exclude all
lost+found directories from backup, use the following command:
mysql> SET tokudb_backup_exclude='/lost\\+found($|/)';
pid file is excluded by default. If you’re providing your own
additions to the exclusions and have the
pid file in the default
location, you will need to add the mysqld_safe.pid entry.
You can throttle the backup rate using the
tokudb_backup_throttle session-level variable. This variable throttles the write rate in bytes per second of the backup to prevent TokuBackup from crowding out other jobs in the system. The default and max value is 18446744073709551615.
mysql> SET tokudb_backup_throttle=1000000;
You can restrict the location of the destination directory where the backups can be located using the
tokudb_backup_allowed_prefix system-level variable. Attempts to backup to a location outside of the specified directory or its children will result in an error.
The default is
null, backups have no restricted locations. This read-only variable can be set in the
my.cnf configuration file and displayed with the
SHOW VARIABLES command:
mysql> SHOW VARIABLES LIKE 'tokudb_backup_allowed_prefix'; +------------------------------+-----------+ | Variable_name | Value | +------------------------------+-----------+ | tokudb_backup_allowed_prefix | /dumpdir | +------------------------------+-----------+
Percona TokuBackup uses two variables to capture errors. They are
tokudb_backup_last_error_string. When TokuBackup encounters an error, these will report on the error number and the error string respectively. For example, the following output shows these parameters following an attempted backup to a directory that was not empty:
mysql> SET tokudb_backup_dir='/tmp/backupdir'; ERROR 1231 (42000): Variable 'tokudb_backup_dir' can't be set to the value of '/tmp/backupdir' mysql> SELECT @@tokudb_backup_last_error; +----------------------------+ | @@tokudb_backup_last_error | +----------------------------+ | 17 | +----------------------------+ mysql> SELECT @@tokudb_backup_last_error_string; +---------------------------------------------------+ | @@tokudb_backup_last_error_string | +---------------------------------------------------+ | tokudb backup couldn't create needed directories. | +---------------------------------------------------+
TokuDB Hot Backup makes a transactionally consistent copy of the TokuDB files while applications read and write to these files. The TokuDB hot backup library intercepts certain system calls that writes files and duplicates the writes on backup files while copying files to the backup directory. The copied files contain the same content as the original files.
TokuDB Hot Backup also has an API. This API includes the
start capturing and
stop capturing commands. The “capturing” command starts the process, when a
portion of a file is copied to the backup location, and this portion is changed,
these changes are also applied to the backup location.
Replication often uses backup replication to create slaves. You must know the last executed global transaction identifier (GTID) or binary log position both for the slave and master configuration.
To lock tables, use
FLUSH TABLE WITH READ LOCK or use the smart locks like
LOCK TABLES FOR BACKUP or
LOCK BINLOG FOR BACKUP.
During the copy process, the binlog is flushed, and the changes are copied to
backup by the “capturing” mechanism. After everything has been copied, and the
“capturing” mechanism is still running, use the
LOCK BINLOG FOR BACKUP.
After this statement is executed, the binlog is flushed, the changes are
captured, and any queries that could change the binlog position or executed GTID
After this command, we can stop capturing and retrieve the last executed GTID or binlog log position and unlock the binlog.
After a backup is taken, there are the following files in the backup directory:
These files contain information for slave and master. You can use this information to start a new slave from the master or slave.
SHOW MASTER STATUS and
SHOW SLAVE STATUS commands provide the
In specific binlog formats, a binary log event can contain statements that produce temporary tables on the slave side, and the result of further statements may depend on the temporary table content. Typically, temporary tables are not selected for backup because they are created in a separate directory. A backup created with temporary tables created by binlog events can cause issues when restored because the temporary tables are not restored. The data may be inconsistent.
The following system variables
enables or disables the safe-slave mode, and
--tokudb-backup-safe-slave-timeout, which defines the maximum amount
of time in seconds to wait until temporary tables disappear. The
safe-slave mode, when used with
LOCK BINLOG FOR BACKUP, the slave SQL
thread is stopped and checked to see if temporary tables produced by the slave
exist or do not exist. If temporary tables exist, the slave SQL thread is
restarted until there are no temporary tables or a defined timeout is reached.
You should not use this option for group-replication. Create a Backup with a Timestamp *****************************
If you plan to store more than one backup in a location, you should add a timestamp to the backup directory name.
A sample Bash script has this information:
#!/bin/bash tm=$(date "+%Y-%m-%d-%H-%M-%S"); backup_dir=$PWD/backup/$tm; mkdir -p $backup_dir; bin/mysql -uroot -e "set tokudb_backup_dir='$backup_dir'"
- You must disable InnoDB asynchronous IO if backing up InnoDB tables with TokuBackup. Otherwise you will have inconsistent, unrecoverable backups. The appropriate setting is
- To be able to run Point-In-Time-Recovery you’ll need to manually get the binary log position.
- Transactional storage engines (TokuDB and InnoDB) will perform recovery on the backup copy of the database when it is first started.
- Tables using non-transactional storage engines (MyISAM) are not locked during the copy and may report issues when starting up the backup. It is best to avoid operations that modify these tables at the end of a hot backup operation (adding/changing users, stored procedures, etc.).
- The database is copied locally to the path specified in
/path/to/backup. This folder must exist, be writable, be empty, and contain enough space for a full copy of the database.
- TokuBackup always makes a backup of the MySQL
datadirand optionally the
tokudb_log_dir, and the binary log folder. The latter three are only backed up separately if they are not the same as or contained in the MySQL
datadir. None of these three folders can be a parent of the MySQL
- No other directory structures are supported. All InnoDB, MyISAM, and other storage engine files must be within the MySQL
- TokuBackup does not follow symbolic links.
- TokuBackup does not backup MySQL configuration file(s).
- TokuBackup does not backup tablespaces if they are out of
- Due to upstream bug #80183, TokuBackup can’t recover backed-up table data if backup was taken while running
ALTER TABLE ... TABLESPACE.
- TokuBackup doesn’t support incremental backups.