The cluster restoration program is implemented as a separate command-line utility ndb_restore, which can normally be found in the MySQL bin directory. This program reads the files created as a result of the backup and inserts the stored information into the database.

ndb_restore must be executed once for each of the backup files that were created by the START BACKUP command used to create the backup (see Section 18.5.3.2, “Using The NDB Cluster Management Client to Create a Backup”). This is equal to the number of data nodes in the cluster at the time that the backup was created.

The following table includes options that are specific to the NDB Cluster native backup restoration program ndb_restore. Additional descriptions follow the table. For options common to most NDB Cluster programs (including ndb_restore), see Section 18.4.26, “Options Common to NDB Cluster Programs — Options Common to NDB Cluster Programs”.

Typical options for this utility are shown here:

ndb_restore [-c connection_string] -n node_id -b backup_id \
      [-m] -r --backup_path=/path/to/backup/files

Normally, when restoring from an NDB Cluster backup, ndb_restore requires at a minimum the --nodeid (short form: -n), --backupid (short form: -b), and --backup_path options.

The -c option is used to specify a connection string which tells ndb_restore where to locate the cluster management server. (See Section 18.3.3.3, “NDB Cluster Connection Strings”, for information on connection strings.) If this option is not used, then ndb_restore attempts to connect to a management server on localhost:1186. This utility acts as a cluster API node, and so requires a free connection “slot” to connect to the cluster management server. This means that there must be at least one [api] or [mysqld] section that can be used by it in the cluster config.ini file. It is a good idea to keep at least one empty [api] or [mysqld] section in config.ini that is not being used for a MySQL server or other application for this reason (see Section 18.3.3.7, “Defining SQL and Other API Nodes in an NDB Cluster”).

You can verify that ndb_restore is connected to the cluster by using the SHOW command in the ndb_mgm management client. You can also accomplish this from a system shell, as shown here:

shell> ndb_mgm -e "SHOW"

The --nodeid or -n is used to specify the node ID of the data node on which the backup was taken.

When restoring to a cluster with different number of data nodes from that where the backup was taken, this information helps identify the correct set or sets of files to be restored to a given node. (In such cases, multiple files usually need to be restored to a single data node.) The next few paragraphs provide an example.

Restore to a different number of data nodes.  You can restore to a cluster having fewer data nodes than the original provided that the larger number of nodes is an even multiple of the smaller number. In the following example, we use a backup taken on a cluster having four data nodes to a cluster having two data nodes.

It is possible to restore data without restoring table metadata. The default behavior when doing this is for ndb_restore to fail with an error if table data do not match the table schema; this can be overridden using the --skip-table-check or -s option.

Some of the restrictions on mismatches in column definitions when restoring data using ndb_restore are relaxed; when one of these types of mismatches is encountered, ndb_restore does not stop with an error as it did previously, but rather accepts the data and inserts it into the target table while issuing a warning to the user that this is being done. This behavior occurs whether or not either of the options --skip-table-check or --promote-attributes is in use. These differences in column definitions are of the following types:

ndb_restore supports limited attribute promotion in much the same way that it is supported by MySQL replication; that is, data backed up from a column of a given type can generally be restored to a column using a “larger, similar” type. For example, data from a CHAR(20) column can be restored to a column declared as VARCHAR(20), VARCHAR(30), or CHAR(30); data from a MEDIUMINT column can be restored to a column of type INT or BIGINT. See Section 17.4.1.10.2, “Replication of Columns Having Different Data Types”, for a table of type conversions currently supported by attribute promotion.

Attribute promotion by ndb_restore must be enabled explicitly, as follows:

Prior to NDB 7.2.14, conversions between character data types and TEXT or BLOB were not handled correctly (Bug #17325051).

Prior to NDB 7.2.18, demotion of TEXT to TINYTEXT was not handled correctly (Bug #18875137).

When converting between character data types and TEXT or BLOB, only conversions between character types (CHAR and VARCHAR) and binary types (BINARY and VARBINARY) can be performed at the same time. For example, you cannot promote an INT column to BIGINT while promoting a VARCHAR column to TEXT in the same invocation of ndb_restore.

Converting between TEXT columns using different character sets is not supported. Beginning with NDB 7.2.18, it is expressly disallowed (Bug #18875137).

When performing conversions of character or binary types to TEXT or BLOB with ndb_restore, you may notice that it creates and uses one or more staging tables named table_name$STnode_id. These tables are not needed afterwards, and are normally deleted by ndb_restore following a successful restoration.

--lossy-conversions, -L

This option is intended to complement the --promote-attributes option. Using --lossy-conversions allows lossy conversions of column values (type demotions or changes in sign) when restoring data from backup. With some exceptions, the rules governing demotion are the same as for MySQL replication; see Section 17.4.1.10.2, “Replication of Columns Having Different Data Types”, for information about specific type conversions currently supported by attribute demotion.

ndb_restore reports any truncation of data that it performs during lossy conversions once per attribute and column.

The --preserve-trailing-spaces option (short form -R) causes trailing spaces to be preserved when promoting a fixed-width character data type to its variable-width equivalent—that is, when promoting a CHAR column value to VARCHAR or a BINARY column value to VARBINARY. Otherwise, any trailing spaces are dropped from such column values when they are inserted into the new columns.

The -b option is used to specify the ID or sequence number of the backup, and is the same number shown by the management client in the Backup backup_id completed message displayed upon completion of a backup. (See Section 18.5.3.2, “Using The NDB Cluster Management Client to Create a Backup”.)

--restore_epoch (short form: -e) adds (or restores) epoch information to the cluster replication status table. This is useful for starting replication on an NDB Cluster replication slave. When this option is used, the row in the mysql.ndb_apply_status having 0 in the id column is updated if it already exists; such a row is inserted if it does not already exist. (See Section 18.6.9, “NDB Cluster Backups With NDB Cluster Replication”.)

--restore_data

This option causes ndb_restore to output NDB table data and logs.

--restore_meta

This option causes ndb_restore to print NDB table metadata.

The first time you run the ndb_restore restoration program, you also need to restore the metadata. In other words, you must re-create the database tables—this can be done by running it with the --restore_meta (-m) option. Restoring the metadata need be done only on a single data node; this is sufficient to restore it to the entire cluster.

--restore-privilege-tables

ndb_restore does not by default restore distributed MySQL privilege tables (NDB 7.2.0 and later). This option causes ndb_restore to restore the privilege tables.

This works only if the privilege tables were converted to NDB before the backup was taken. For more information, see Section 18.5.14, “Distributed MySQL Privileges for NDB Cluster”.

--backup_path

The path to the backup directory is required; this is supplied to ndb_restore using the --backup_path option, and must include the subdirectory corresponding to the ID backup of the backup to be restored. For example, if the data node's DataDir is /var/lib/mysql-cluster, then the backup directory is /var/lib/mysql-cluster/BACKUP, and the backup files for the backup with the ID 3 can be found in /var/lib/mysql-cluster/BACKUP/BACKUP-3. The path may be absolute or relative to the directory in which the ndb_restore executable is located, and may be optionally prefixed with backup_path=.

It is possible to restore a backup to a database with a different configuration than it was created from. For example, suppose that a backup with backup ID 12, created in a cluster with two database nodes having the node IDs 2 and 3, is to be restored to a cluster with four nodes. Then ndb_restore must be run twice—once for each database node in the cluster where the backup was taken. However, ndb_restore cannot always restore backups made from a cluster running one version of MySQL to a cluster running a different MySQL version. See Section 18.2.7, “Upgrading and Downgrading NDB Cluster 7.2”, for more information.

For more rapid restoration, the data may be restored in parallel, provided that there is a sufficient number of cluster connections available. That is, when restoring to multiple nodes in parallel, you must have an [api] or [mysqld] section in the cluster config.ini file available for each concurrent ndb_restore process. However, the data files must always be applied before the logs.

--no-upgrade

When using ndb_restore to restore a backup, VARCHAR columns created using the old fixed format are resized and recreated using the variable-width format now employed. This behavior can be overridden using the --no-upgrade option (short form: -u) when running ndb_restore.

--print_data

The --print_data option causes ndb_restore to direct its output to stdout.

TEXT and BLOB column values are always truncated. In NDB 7.2.18 and earlier, such values are truncated to the first 240 bytes in the output; in NDB 7.2.19 and later, they are truncated to 256 bytes. (Bug #14571512, Bug #65467) This cannot currently be overridden when using --print_data.

Several additional options are available for use with the --print_data option in generating data dumps, either to stdout, or to a file. These are similar to some of the options used with mysqldump, and are shown in the following list:

--print_meta

This option causes ndb_restore to print all metadata to stdout.

--print_log

The --print_log option causes ndb_restore to output its log to stdout.

--print

Causes ndb_restore to print all data, metadata, and logs to stdout. Equivalent to using the --print_data, --print_meta, and --print_log options together.

--dont_ignore_systab_0

Normally, when restoring table data and metadata, ndb_restore ignores the copy of the NDB system table that is present in the backup. --dont_ignore_systab_0 causes the system table to be restored. This option is intended for experimental and development use only, and is not recommended in a production environment.

--ndb-nodegroup-map, -z

This option can be used to restore a backup taken from one node group to a different node group. Its argument is a list of the form source_node_group, target_node_group.

--no-binlog

This option prevents any connected SQL nodes from writing data restored by ndb_restore to their binary logs.

--no-restore-disk-objects, -d

This option stops ndb_restore from restoring any NDB Cluster Disk Data objects, such as tablespaces and log file groups; see Section 18.5.12, “NDB Cluster Disk Data Tables”, for more information about these.

--parallelism=#, -p

ndb_restore uses single-row transactions to apply many rows concurrently. This parameter determines the number of parallel transactions (concurrent rows) that an instance of ndb_restore tries to use. By default, this is 128; the minimum is 1, and the maximum is 1024.

The work of performing the inserts is parallelized across the threads in the data nodes involved. This mechanism is employed for restoring bulk data from the .Data file—that is, the fuzzy snapshot of the data; it is not used for building or rebuilding indexes. The change log is applied serially; index drops and builds are DDL operations and handled separately. There is no thread-level parallelism on the client side of the restore.

--progress-frequency=N

Print a status report each N seconds while the backup is in progress. 0 (the default) causes no status reports to be printed. The maximum is 65535.

--verbose=#

Sets the level for the verbosity of the output. The minimum is 0; the maximum is 255. The default value is 1.

It is possible to restore only selected databases, or selected tables from a single database, using the syntax shown here:

ndb_restore other_options db_name,[db_name[,...] | tbl_name[,tbl_name][,...]]

In other words, you can specify either of the following to be restored:

--include-databases=db_name[,db_name][,...]

--include-tables=db_name.tbl_name[,db_name.tbl_name][,...]

Use the --include-databases option or the --include-tables option for restoring only specific databases or tables, respectively. --include-databases takes a comma-delimited list of databases to be restored. --include-tables takes a comma-delimited list of tables (in database.table format) to be restored.

When --include-databases or --include-tables is used, only those databases or tables named by the option are restored; all other databases and tables are excluded by ndb_restore, and are not restored.

The following table shows several invocations of ndb_restore using --include-* options (other options possibly required have been omitted for clarity), and the effects these have on restoring from an NDB Cluster backup:

You can also use these two options together. For example, the following causes all tables in databases db1 and db2, together with the tables t1 and t2 in database db3, to be restored (and no other databases or tables):

shell> ndb_restore [...] --include-databases=db1,db2 --include-tables=db3.t1,db3.t2

(Again we have omitted other, possibly required, options in the example just shown.)

--exclude-databases=db_name[,db_name][,...]

--exclude-tables=db_name.tbl_name[,db_name.tbl_name][,...]

It is possible to prevent one or more databases or tables from being restored using the ndb_restore options --exclude-databases and --exclude-tables. --exclude-databases takes a comma-delimited list of one or more databases which should not be restored. --exclude-tables takes a comma-delimited list of one or more tables (using database.table format) which should not be restored.

When --exclude-databases or --exclude-tables is used, only those databases or tables named by the option are excluded; all other databases and tables are restored by ndb_restore.

This table shows several invocations of ndb_restore usng --exclude-* options (other options possibly required have been omitted for clarity), and the effects these options have on restoring from an NDB Cluster backup:

You can use these two options together. For example, the following causes all tables in all databases except for databases db1 and db2, and tables t1 and t2 in database db3, to be restored:

shell> ndb_restore [...] --exclude-databases=db1,db2 --exclude-tables=db3.t1,db3.t2

(Again, we have omitted other possibly necessary options in the interest of clarity and brevity from the example just shown.)

You can use --include-* and --exclude-* options together, subject to the following rules:

For example, the following set of options causes ndb_restore to restore all tables from database db1 except db1.t1, while restoring no other tables from any other databases:

        
--include-databases=db1 --exclude-tables=db1.t1

However, reversing the order of the options just given simply causes all tables from database db1 to be restored (including db1.t1, but no tables from any other database), because the --include-databases option, being farthest to the right, is the first match against database db1 and thus takes precedence over any other option that matches db1 or any tables in db1:

        
--exclude-tables=db1.t1 --include-databases=db1

--exclude-missing-columns

It is also possible to restore only selected table columns using the --exclude-missing-columns option. When this option is used, ndb_restore ignores any columns missing from tables being restored as compared to the versions of those tables found in the backup. This option applies to all tables being restored. If you wish to apply this option only to selected tables or databases, you can use it in combination with one or more of the options described in the previous paragraph to do so, then restore data to the remaining tables using a complementary set of these options.

--exclude-missing-tables

Beginning with NDB 7.2.17, it is also possible to restore only selected tables columns using this option, which causes ndb_restore to ignore any tables from the backup that are not found in the target database.

--disable-indexes

Disable restoration of indexes during restoration of the data from a native NDB backup. Afterwards, you can restore indexes for all tables at once with multi-threaded building of indexes using --rebuild-indexes, which should be faster than rebuilding indexes concurrently for very large tables.

--rebuild-indexes

You can use this option with ndb_restore to cause multi-threaded rebuilding of the ordered indexes while restoring a native NDB backup. The number of threads used for building ordered indexes by ndb_restore with this option is controlled by the BuildIndexThreads data node configuration parameter (NDB 6.3.30 and later; NDB 7.0.11 and later).

It is necessary to use this option only for the first run of ndb_restore; this causes all ordered indexes to be rebuilt without using --rebuild-indexes again when restoring subsequent nodes. You should use this option prior to inserting new rows into the database; otherwise, it is possible for a row to be inserted that later causes a unique constraint violation when trying to rebuild the indexes.

Building of ordered indices is parallelized with the number of LDMs by default. Offline index builds performed during node and system restarts can be made faster using the BuildIndexThreads data node configuration parameter; this parameter has no effect on dropping and rebuilding of indexes by ndb_restore, which is performed online.

Rebuilding of unique indexes uses disk write bandwidth for redo logging and local checkpointing. An insufficient amount of this bandwith can lead to redo buffer overload or log overload errors. In such cases you can run ndb_restore --rebuild-indexes again; the process resumes at the point where the error occurred. You can also do this when you have encountered temporary errors. You can repeat execution of ndb_restore --rebuild-indexes indefinitely; you may be able to stop such errors by reducing the value of DiskCheckpointSpeed to provide additional disk bandwidth to redo logging, or reducing the --parallelism. If the problem is insufficient space, you can increase the size of the redo log (FragmentLogFileSize node configuration parameter), or you can increase the speed at which LCPs are performed (DiskCheckpointSpeed), in order to free space more quickly.

--skip-broken-objects

This option causes ndb_restore to ignore corrupt tables while reading a native NDB backup, and to continue restoring any remaining tables (that are not also corrupted). Currently, the --skip-broken-objects option works only in the case of missing blob parts tables.

--skip-unknown-objects

This option causes ndb_restore to ignore any schema objects it does not recognize while reading a native NDB backup. This can be used for restoring a backup made from a cluster running NDB Cluster 7.2 to a cluster running NDB Cluster 7.1.

--rewrite-database=old_dbname,new_dbname

This option makes it possible to restore to a database having a different name from that used in the backup. For example, if a backup is made of a database named products, you can restore the data it contains to a database named inventory, use this option as shown here (omitting any other options that might be required):

shell> ndb_restore --rewrite-database=product,inventory

The option can be employed multiple times in a single invocation of ndb_restore. Thus it is possible to restore simultaneously from a database named db1 to a database named db2 and from a database named db3 to one named db4 using --rewrite-database=db1,db2 --rewrite-database=db3,db4. Other ndb_restore options may be used between multiple occurrences of --rewrite-database.

In the event of conflicts between multiple --rewrite-database options, the last --rewrite-database option used, reading from left to right, is the one that takes effect. For example, if --rewrite-database=db1,db2 --rewrite-database=db1,db3 is used, only --rewrite-database=db1,db3 is honored, and --rewrite-database=db1,db2 is ignored. It is also possible to restore from multiple databases to a single database, so that --rewrite-database=db1,db3 --rewrite-database=db2,db3 restores all tables and data from databases db1 and db2 into database db3.

--exclude-intermediate-sql-tables[=TRUE|FALSE]

When performing copying ALTER TABLE operations, mysqld creates intermediate tables (whose names are prefixed with #sql-). When TRUE, the --exclude-intermediate-sql-tables option keeps ndb_restore from restoring such tables that may have been left over from such operations. This option is TRUE by default.

The --exclude-intermediate-sql-tables option was introduced in NDB 7.2.17. (Bug #17882305)

Error reporting.  ndb_restore reports both temporary and permanent errors. In the case of temporary errors, it may able to recover from them, and reports Restore successful, but encountered temporary error, please look at configuration in such cases.