It is common when setting up a new instance of NDB Cluster to need to import data from an existing NDB Cluster, instance of MySQL, or other source. This data is most often available in one or more of the following formats:
An SQL dump file such as produced by mysqldump. This can be imported using the mysql client, as shown later in this section.
A CSV file produced by mysqldump or other export program. Such files can be imported into
NDB
usingLOAD DATA INFILE
in the mysql client, or with the ndb_import utility provided with the NDB Cluster distribution. For more information about the latter, see Section 25.5.13, “ndb_import — Import CSV Data Into NDB”.A native
NDB
backup produced usingSTART BACKUP
in theNDB
management client. To import a native backup, you must use the ndb_restore program that comes as part of NDB Cluster. See Section 25.5.23, “ndb_restore — Restore an NDB Cluster Backup”, for more about using this program.
When importing data from an SQL file, it is often not necessary to enforce transactions or foreign keys, and temporarily disabling these features can speed up the import process greatly. This can be done using the mysql client, either from a client session, or by invoking it on the command line. Within a mysql client session, you can perform the import using the following SQL statements:
SET ndb_use_transactions=0;
SET foreign_key_checks=0;
source path/to/dumpfile;
SET ndb_use_transactions=1;
SET foreign_key_checks=1;
When performing the import in this fashion, you
must enable
ndb_use_transaction
and
foreign_key_checks
again following execution of
the mysql client's
source
command. Otherwise, it is possible for
later statements in same session may also be executed without
enforcing transactions or foreign key constraints, and which could
lead to data inconcsistency.
From the system shell, you can import the SQL file while disabling
enforcement of transaction and foreign keys by using the
mysql client with the
--init-command
option, like this:
$> mysql --init-command='SET ndb_use_transactions=0; SET foreign_key_checks=0' < path/to/dumpfile
It is also possible to load the data into an
InnoDB
table, and convert it to use
the NDB
storage engine afterwards using ALTER
TABLE ... ENGINE NDB). You should take into account, especially
for many tables, that this may require a number of such
operations; in addition, if foreign keys are used, you must mind
the order of the ALTER TABLE
statements
carefully, due to the fact that foreign keys do not work between
tables using different MySQL storage engines.
You should be aware that the methods described previously in this
section are not optimized for very large data sets or large
transactions. Should an application really need big transactions
or many concurrent transactions as part of normal operation, you
may wish to increase the value of the
MaxNoOfConcurrentOperations
data node configuration parameter, which reserves more memory to
allow a data node to take over a transaction if its transaction
coordinator stops unexpectedly.
You may also wish to do this when performing bulk
DELETE
or
UPDATE
operations on NDB Cluster
tables. If possible, try to have applications perform these
operations in chunks, for example, by adding
LIMIT
to such statements.
If a data import operation does not complete successfully, for
whatever reason, you should be prepared to perform any necessary
cleanup including possibly one or more DROP
TABLE
statements, DROP
DATABASE
statements, or both. Failing to do so may leave
the database in an inconsistent state.