MariaDB Enterprise Backup

Overview

Regular and reliable backups are essential to successful recovery of mission critical applications. MariaDB Enterprise Server backup and restore operations are performed using MariaDB Enterprise Backup, an enterprise-build of MariaDB Backup.

MariaDB Enterprise Backup is compatible with MariaDB Enterprise Server 10.2, 10.3, 10.4, 10.5, and 10.6.

Storage Engines and Backup Types

MariaDB Enterprise Backup creates a file-level backup of data from the MariaDB Enterprise Server data directory. This backup includes temporal data, and the encrypted and unencrypted tablespaces of supported storage engines (e.g., InnoDB, MyRocks, Aria).

MariaDB Enterprise Server implements:

  • Full backups, which contain all data in the database.

  • Incremental backups, which contain modifications since the last backup.

  • Partial backups, which contain a subset of the tables in the database.

Backup support is specific to storage engines. All supported storage engines enable full backup. The InnoDB storage engine additionally supports incremental backup.

Note

MariaDB Enterprise Backup does not support backups of MariaDB ColumnStore. Backup of MariaDB ColumnStore can be performed using MariaDB ColumnStore Tools. Backup of data ingested to MariaDB ColumnStore can also occur pre-ingestion, such as in the case of HTAP where backup could occur of transactional data in MariaDB Enterprise Server, and restore of data to MariaDB ColumnStore would then occur through reprocessing.

Non-blocking Backups

A feature of MariaDB Enterprise Backup and MariaDB Enterprise Server, non-blocking backups minimize workload impact during backups. When MariaDB Enterprise Backup connects to MariaDB Enterprise Server, staging operations are initiated to protect data during read.

Non-blocking backup functionality differs from historical backup functionality in the following ways:

  • MariaDB Enterprise Backup in MariaDB Enterprise Server includes enterprise-only optimizations to backup staging, including DDL statement tracking, which reduces lock-time during backups.

  • MariaDB Backup in MariaDB Community Server 10.4 and later will block writes, log tables, and statistics.

  • Older MariaDB Community Server releases used FLUSH TABLES WITH READ LOCK, which closed open tables and only allowed tables to be reopened with a read lock during the duration of backups.

Understanding Recovery

MariaDB Enterprise Backup creates complete or incremental backups of MariaDB Enterprise Server data, and is also used to restore data from backups produced using MariaDB Enterprise Backup.

Preparing Backups for Recovery

Full backups produced using MariaDB Enterprise Server are not initially point-in-time consistent, and an attempt to restore from a raw full backup will cause InnoDB to crash to protect the data.

Incremental backups produced using MariaDB Enterprise Backup contain only the changes since the last backup and cannot be used standalone to perform a restore.

To restore from a backup, you first need to prepare the backup for point-in-time consistency using the --prepare command:

  • Running the --prepare command on a full backup synchronizes the tablespaces, ensuring that they are point-in-time consistent and ready for use in recovery.

  • Running the --prepare command on an incremental backup synchronizes the tablespaces and also applies the updated data into the previous full backup, making it a complete backup ready for use in recovery.

  • Running the --prepare command on data that is to be used for a partial restore (when restoring only one or more selected tables) requires that you also use the --export option to create the necessary .cfg files to use in recovery.

Restore Requires Empty Data Directory

When MariaDB Enterprise Backup restores from a backup, it copies or moves the backup files into the MariaDB Enterprise Server data directory, as defined by the datadir system variable.

For MariaDB Enterprise Backup to safely restore data from full and incremental backups, the data directory must be empty. One way to achieve this is to move the data directory aside to a unique directory name:

  1. Make sure that the Server is stopped.

  2. Move the data directory to a unique name (for example, /var/lib/mysql-2020-01-01) OR remove the old data directory (depending on how much space you have available).

  3. Create a new (empty) data directory (for example, mkdir /var/lib/mysql).

  4. Run MariaDB Enterprise Backup to restore the databases into that directory.

  5. Change the ownership of all the restored files to the correct system user (for example, chown -R mysql:mysql /var/lib/mysql).

  6. Start MariaDB Enterprise Server, which now uses the restored data directory.

  7. When ready, and if you have not already done so, delete the old data directory to free disk space.

Creating the Backup User

When MariaDB Enterprise Backup performs a backup operation, it not only copies files from the data directory but also connects to the running MariaDB Enterprise Server.

This connection to MariaDB Enterprise Server is used to manage locks and backup staging that prevent the Server from writing to a file while being read for a backup.

MariaDB Enterprise Backup establishes this connection based on the user credentials specified with the --user and --password options when performing a backup.

It is recommended that a dedicated user be created and authorized to perform backups.

10.5 and Later

MariaDB Enterprise Backup 10.5 and later requires this user to have the RELOAD, PROCESS, LOCK TABLES, and BINLOG MONITOR privileges. (The BINLOG MONITOR privilege replaced the REPLICATION CLIENT privilege in MariaDB Enterprise Server 10.5.):

CREATE USER 'mariabackup'@'localhost'
IDENTIFIED BY 'mbu_passwd';

GRANT RELOAD, PROCESS, LOCK TABLES, BINLOG MONITOR
ON *.*
TO 'mariabackup'@'localhost';

In the above example, MariaDB Enterprise Backup would run on the local system that runs MariaDB Enterprise Server. Where backups may be run against a remote server, the user authentication and authorization should be adjusted.

While MariaDB Enterprise Backup requires a user for backup operations, no user is required for restore operations since restores occur while MariaDB Enterprise Server is not running.

10.4 and Earlier

MariaDB Enterprise Backup 10.4 and earlier requires this user to have the RELOAD, PROCESS, LOCK TABLES, and REPLICATION CLIENT privileges. (The BINLOG MONITOR privilege replaced the REPLICATION CLIENT privilege in MariaDB Enterprise Server 10.5.):

CREATE USER 'mariabackup'@'localhost'
IDENTIFIED BY 'mbu_passwd';

GRANT RELOAD, PROCESS, LOCK TABLES, REPLICATION CLIENT
ON *.*
TO 'mariabackup'@'localhost';

In the above example, MariaDB Enterprise Backup would run on the local system that runs MariaDB Enterprise Server. Where backups may be run against a remote server, the user authentication and authorization should be adjusted.

While MariaDB Enterprise Backup requires a user for backup operations, no user is required for restore operations since restores occur while MariaDB Enterprise Server is not running.

Full Backup and Restore

Full backups performed with MariaDB Enterprise Backup contain all table data present in the database.

When performing a full backup, MariaDB Enterprise Backup makes a file-level copy of the MariaDB Enterprise Server data directory. This backup omits log data such as the binary logs (binlog), error logs, general query logs, and slow query logs.

Performing Full Backups

When you perform a full backup, MariaDB Enterprise Backup writes the backup to the --target-dir path. The directory must be empty or non-existent and the operating system user account must have permission to write to that directory. A database user account is required to perform the backup.

The version of mariabackup or mariadb-backup should be the same version as the MariaDB Enterprise Server version. When the version does not match the server version, errors can sometimes occur, or the backup can sometimes be unusable.

To create a backup, execute mariabackup or mariadb-backup with the --backup option, and provide the database user account credentials using the --user and --password options:

$ sudo mariabackup --backup \
      --target-dir=/data/backups/full \
      --user=mariabackup \
      --password=mbu_passwd

Subsequent to the above example, the backup is now available in the designated --target-dir path.

Preparing a Full Backup for Recovery

A raw full backup is not point-in-time consistent and must be prepared before it can be used for a restore. The backup can be prepared any time after the backup is created and before the backup is restored. However, MariaDB recommends preparing a backup immediately after taking the backup to ensure that the backup is consistent.

The backup should be prepared with the same version of MariaDB Enterprise Backup that was used to create the backup.

To prepare the backup, execute mariabackup or mariadb-backup with the --prepare option:

$ sudo mariabackup --prepare \
   --use-memory=34359738368 \
   --target-dir=/data/backups/full

For best performance, the --use-memory option should be set to the server's innodb_buffer_pool_size value.

Restoring from Full Backups

Once a full backup has been prepared to be point-in-time consistent, MariaDB Enterprise Backup is used to copy backup data to the MariaDB Enterprise Server data directory.

To restore from a full backup:

  1. Stop the MariaDB Enterprise Server

  2. Empty the data directory

  3. Restore from the "full" directory using the --copy-back option:

    $ sudo mariabackup --copy-back --target-dir=/data/backups/full
    

MariaDB Enterprise Backup writes to the data directory as the current user, which can be changed using sudo. To confirm that restored files are properly owned by the user that runs MariaDB Enterprise Server, run a command like this (adapted for the correct user/group):

$ sudo chown -R mysql:mysql /var/lib/mysql

Once this is done, start MariaDB Enterprise Server:

$ sudo systemctl start mariadb

When the Server starts, it works from the restored data directory.

Incremental Backup and Restore

Full backups of large data-sets can be time-consuming and resource-intensive. MariaDB Enterprise Backup supports the use of incremental backups to minimize this impact.

While full backups are resource-intensive at time of backup, the resource burden around incremental backups occurs when preparing for restore. First, the full backup is prepared for restore, then each incremental backup is applied.

Performing Incremental Backups

When you perform an incremental backup, MariaDB Enterprise Backup compares a previous full or incremental backup to what it finds on MariaDB Enterprise Server. It then creates a new backup containing the incremental changes.

Incremental backup is supported for InnoDB tables. Tables using other storage engines receive full backups even during incremental backup operations.

To increment a full backup, use the --incremental-basedir option to indicate the path to the full backup and the --target-dir option to indicate where you want to write the incremental backup:

$ sudo mariabackup --backup \
      --incremental-basedir=/data/backups/full \
      --target-dir=/data/backups/inc1 \
      --user=mariabackup \
      --password=mbu_passwd

In this example, MariaDB Enterprise Backup reads the /data/backups/full directory, and MariaDB Enterprise Server then creates an incremental backup in the /data/backups/inc1 directory.

Preparing an Incremental Backup

An incremental backup must be applied to a prepared full backup before it can be used in a restore operation. If you have multiple full backups to choose from, pick the nearest full backup prior to the incremental backup that you want to restore. You may also want to back up your full-backup directory, as it will be modified by the updates in the incremental data.

If your full backup directory is not yet prepared, run this to make it consistent:

$ sudo mariabackup --prepare --target-dir=/data/backups/full

Then, using the prepared full backup, apply the first incremental backup's data to the full backup in an incremental preparation step:

$ sudo mariabackup --prepare \
      --target-dir=/data/backups/full \
      --incremental-dir=/data/backups/inc1

Once the incremental backup has been applied to the full backup, the full backup directory contains the changes from the incremental backup (that is, the inc1/ directory). Feel free to remove inc1/ to save disk space.

Restoring from Incremental Backups

Once you have prepared the full backup directory with all the incremental changes you need (as described above), stop the MariaDB Enterprise Server, empty its data directory, and restore from the original full backup directory using the --copy-back option:

$ sudo mariabackup --copy-back --target-dir=/data/backups/full

MariaDB Enterprise Backup writes files into the data directory using either the current user or root (in the case of a sudo operation), which may be different from the system user that runs the database. Run the following to recursively update the ownership of the restored files and directories:

$ sudo chown -R mysql:mysql /var/lib/mysql

Then, start MariaDB Enterprise Server. When the Server starts, it works from the restored data directory.

Partial Backup and Restore

In a partial backup, MariaDB Enterprise Backup copies a specified subset of tablespaces from the MariaDB Enterprise Server data directory. Partial backups are useful in establishing a higher frequency of backups on specific data, at the expense of increased recovery complexity. In selecting tablespaces for a partial backup, please consider referential integrity.

Performing a Partial Backup

Command-line options can be used to narrow the set of databases or tables to be included within a backup:

Option

Description

--databases

List of databases to include

--databases-exclude

List of databases to omit from the backup

--databases-file

Path to file listing the databases to include

--tables

List of tables to include

--tables-exclude

List of tables to exclude

--tables-file

Path to file listing the tables to include

For example, you may wish to produce a partial backup, which excludes a specific database:

$ sudo mariabackup --backup \
      --target-dir=/data/backups/part \
      --user=mariabackup \
      --password=mbu_passwd \
      --database-exclude=test

Partial backups can also be incremental:

$ sudo mariabackup --backup \
      --incremental-basedir=/data/backups/part \
      --target-dir=/data/backups/part_inc1 \
      --user=mariabackup \
      --password=mbu_passwd  \
      --database-exclude=test

Preparing a Backup Before a Partial Restore

As with full and incremental backups, partial backups are not point-in-time consistent. A partial backup must be prepared before it can be used for recovery.

A partial restore can be performed from a full backup or partial backup.

The preparation step for either partial or full backup restoration requires the use of transportable tablespaces for InnoDB. As such, each prepare operation requires the --export option:

$ sudo mariabackup --prepare --export --target-dir=/data/backups/part

When using a partial incremental backup for restore, the incremental data must be applied to its prior partial backup data before its data is complete. If performing partial incremental backups, run the prepare statement again to apply the incremental changes onto the partial backup that served as the base.

$ sudo mariabackup --prepare --export \
      --target-dir=/data/backups/part \
      --incremental-dir=/data/backups/part_inc1

Performing a Partial Restore

Unlike full and incremental backups, you cannot restore partial backups directly using MariaDB Enterprise Backup. Further, as a partial backup does not contain a complete data directory, you cannot restore MariaDB Enterprise Server to a startable state solely with a partial backup.

To restore from a partial backup, you need to prepare a table on the MariaDB Enterprise Server, then manually copy the files into the data directory.

The details of the restore procedure depend on the characteristics of the table:

As partial restores are performed while the server is running, not stopped, care should be taken to prevent production workloads during restore activity.

Note

You can also use data from a full backup in a partial restore operation if you have prepared the data using the --export option as described above.

Partial Restore Non-partitioned Tables

To restore a non-partitioned table from a backup, first create a new table on MariaDB Enterprise Server to receive the restored data. It should match the specifications of the table you're restoring.

Be extra careful if the backup data is from a server with a different version than the restore server, as some differences (such as a differing ROW_FORMAT) can cause an unexpected result.

  1. Create an empty table for the data being restored:

    CREATE TABLE test.address_book (
       id INT PRIMARY KEY AUTO_INCREMENT,
       name VARCHAR(255),
       email VARCHAR(255));
    
  2. Modify the table to discard the tablespace:

    ALTER TABLE test.address_book DISCARD TABLESPACE;
    
  3. You can copy (or move) the files for the table from the backup to the data directory:

    $ sudo cp /data/backups/part_inc1/test/address_book.* /var/lib/mysql/test
    
  4. Use a wildcard to include both the .ibd and .cfg files. Then, change the owner to the system user running MariaDB Enterprise Server:

    $ sudo chown mysql:mysql /var/lib/mysql/test/address_book.*
    
  5. Lastly, import the new tablespace:

    ALTER TABLE test.address_book IMPORT TABLESPACE;
    

    MariaDB Enterprise Server looks in the data directory for the tablespace you copied in, then imports it for use. If the table is encrypted, it also looks for the encryption key with the relevant key ID that the table data specifies.

  6. Repeat this step for every table you wish to restore.

Partial Restore Partitioned Tables

Restoring a partitioned table from a backup requires a few extra steps compared to restoring a non-partitioned table.

To restore a partitioned table from a backup, first create a new table on MariaDB Enterprise Server to receive the restored data. It should match the specifications of the table you're restoring, including the partition specification.

Be extra careful if the backup data is from a server with a different version than the restore server, as some differences (such as a differing ROW_FORMAT) can cause an unexpected result.

  1. Create an empty table for the data being restored:

    CREATE TABLE test.students (
       id INT NOT NULL AUTO_INCREMENT,
       name VARCHAR(255),
       email VARCHAR(255),
       graduating_year YEAR,
       PRIMARY KEY (id, graduating_year)
    ) ENGINE = InnoDB
    PARTITION BY RANGE (graduating_year) (
       PARTITION p0 VALUES LESS THAN (2019),
       PARTITION p1 VALUES LESS THAN MAXVALUE
    );
    
  2. Then create a second empty table matching the column specification, but without partitions. This will be your working table:

    CREATE TABLE test.students_work AS
    SELECT * FROM test.students WHERE NULL;
    
  3. For each partition you want to restore, discard the working table's tablespace:

    ALTER TABLE test.students_work DISCARD TABLESPACE;
    
  4. Then, copy the table files from the backup, using the new name:

    $ sudo cp /data/backups/part_inc1/test/students.ibd /var/lib/mysql/test/students_work.ibd
    $ sudo cp /data/backups/part_inc1/test/students.cfg /var/lib/mysql/test/students_work.cfg
    
  5. Change the owner to that of the user running MariaDB Enterprise Server:

    $ sudo chown mysql:mysql /var/lib/mysql/test/students_work.*
    
  6. Import the copied tablespace:

    ALTER TABLE test.students_work IMPORT TABLESPACE;
    
  7. Lastly, exchange the partition, copying the tablespace from the working table into the partition file for the target table:

    ALTER TABLE test.students EXCHANGE PARTITION p0 WITH TABLE test.students_work;
    
  8. Repeat the above process for each partition until you have them all exchanged into the target table. Then delete the working table, as it's no longer necessary:

    DROP TABLE test.students_work;
    

    This restores a partitioned table.

Partial Restore of Tables with Full-Text Indexes

When restoring a table with a full-text search (FTS) index, InnoDB may throw a schema mismatch error.

In this case, to restore the table, it is recommended to:

  • Remove the corresponding .cfg file.

  • Restore data to a table without any secondary indexes including FTS.

  • Add the necessary secondary indexes to the restored table.

For example, to restore table t1 with FTS index from database db1:

  1. In the MariaDB shell, drop the table you are going to restore:

    DROP TABLE IF EXISTS db1.t1;
    
  2. Create an empty table for the data being restored:

    CREATE TABLE db1.t1(f1 CHAR(10)) ENGINE=INNODB;
    
  3. Modify the table to discard the tablespace:

    ALTER TABLE db1.t1 DISCARD TABLESPACE;
    
  4. In the operating system shell, copy the table files from the backup to the data directory of the corresponding database:

    $ sudo cp /data/backups/part/db1/t1.* /var/lib/mysql/db1
    
  5. Remove the .cfg file from the data directory:

    $ sudo rm /var/lib/mysql/db1/t1.cfg
    
  6. Change the owner of the newly copied files to the system user running MariaDB Enterprise Server:

    $ sudo chown mysql:mysql /var/lib/mysql/db1/t1.*
    
  7. In the MariaDB shell, import the copied tablespace:

    ALTER TABLE db1.t1 IMPORT TABLESPACE;
    
  8. Verify that the data has been successfully restored:

    SELECT * FROM db1.t1;
    
    +--------+
    | f1     |
    +--------+
    | ABC123 |
    +--------+
    
  9. Add the necessary secondary indexes:

    ALTER TABLE db1.t1 FORCE, ADD FULLTEXT INDEX f_idx(f1);
    
  10. The table is now fully restored:

    SHOW CREATE TABLE db1.t1\G
    
    *************************** 1. row ***************************
           Table: t1
    Create Table: CREATE TABLE `t1` (
      `f1` char(10) DEFAULT NULL,
      FULLTEXT KEY `f_idx` (`f1`)
    ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_general_ci
    

Point-in-Time Recoveries

Recovering from a backup restores the data directory at a specific point-in-time, but it does not restore the binary log. In a point-in-time recovery, you begin by restoring the data directory from a full or incremental backup, then use the mysqlbinlog utility to recover the binary log data to a specific point in time.

  1. First, prepare the backup as you normally would for a full or incremental backup:

    $ sudo mariabackup --prepare --target-dir=/data/backups/full
    
  2. When MariaDB Enterprise Backup runs on a MariaDB Enterprise Server where binary logs is enabled, it stores binary log information in the xtrabackup_binlog_info file. Consult this file to find the name of the binary log position to use. In the following example, the log position is 321.

    $ sudo cat /data/backups/full/xtraback_binlog_info
    
    mariadb-node4.00001     321
    
  3. Update the configuration file to use a new data directory.

    [mysqld]
    datadir=/var/lib/mysql_new
    
  4. Using MariaDB Enterprise Backup, restore from the backup to the new data directory:

    $ sudo mariabackup --copy-back --target-dir=/data/backups/full
    
  5. Then change the owner to the MariaDB Enterprise Server system user:

    $ sudo chown -R mysql:mysql /var/lib/mysql_new
    
  6. Start MariaDB Enterprise Server:

    $ sudo systemctl start mariadb
    
  7. Using the binary log file in the old data directory, the start position in the xtrabackup_binlog_info file, the date and time you want to restore to, and the mysqlbinlog utility to create an SQL file with the binary log changes:

    $ mysqlbinlog --start-position=321 \
          --stop-datetime="2019-06-28 12:00:00" \
          /var/lib/mysql/mariadb-node4.00001 \
          > mariadb-binlog.sql
    
  8. Lastly, run the binary log SQL to restore the databases:

    $ mysql -u root -p < mariadb-binlog.sql