It is possible to set a number of options specific to NDB Cluster in the table comment or column comments of an NDB table. Table-level options for controlling read from any replica and partition balance can be embedded in a table comment using NDB_TABLE.

NDB_COLUMN can be used in a column comment to set the size of the blob parts table column used for storing parts of blob values by NDB to its maximum. This works for BLOB, MEDIUMBLOB, LONGBLOB, TEXT, MEDIUMTEXT, LONGTEXT, and JSON columns. Beginning with NDB 8.0.30, a column comment can also be used to control the inline size of a blob column. NDB_COLUMN comments do not support TINYBLOB or TINYTEXT columns, since these have an inline part (only) of fixed size, and no separate parts to store elsewhere.

NDB_TABLE can be used in a table comment to set options relating to partition balance and whether the table is fully replicated, among others.

The remainder of this section describes these options and their use.

NDB_COLUMN Options

In NDB Cluster, a column comment in a CREATE TABLE or ALTER TABLE statement can also be used to specify an NDB_COLUMN option. Beginning with version 8.0.30, NDB supports two column comment options BLOB_INLINE_SIZE and MAX_BLOB_PART_SIZE. (Prior to NDB 8.0.30, only MAX_BLOB_PART_SIZE is supported.) Syntax for this option is shown here:

COMMENT 'NDB_COLUMN=speclist'

speclist := spec[,spec]

spec := 
    BLOB_INLINE_SIZE=value
  | MAX_BLOB_PART_SIZE[={0|1}]

BLOB_INLINE_SIZE specifies the number of bytes to be stored inline by the column; its expected value is an integer in the range 1 - 29980, inclusive. Setting a value greater than 29980 raises an error; setting a value less than 1 is allowed, but causes the default inline size for the column type to be used.

You should be aware that the maximum value for this option is actually the maximum number of bytes that can be stored in one row of an NDB table; every column in the row contributes to this total.

You should also keep in mind, especially when working with TEXT columns, that the value set by MAX_BLOB_PART_SIZE or BLOB_INLINE_SIZE represents column size in bytes. It does not indicate the number of characters, which varies according to the character set and collation used by the column.

To see the effects of this option, first create a table with two BLOB columns, one (b1) with no extra options, and another (b2) with a setting for BLOB_INLINE_SIZE, as shown here:

mysql> CREATE TABLE t1 (
    ->    a INT NOT NULL PRIMARY KEY,
    ->    b1 BLOB,
    ->    b2 BLOB COMMENT 'NDB_COLUMN=BLOB_INLINE_SIZE=8000'
    ->  ) ENGINE NDB;
Query OK, 0 rows affected (0.32 sec)

You can see the BLOB_INLINE_SIZE settings for the BLOB columns by querying the ndbinfo.blobs table, like this:

mysql> SELECT 
    ->   column_name AS 'Column Name', 
    ->   inline_size AS 'Inline Size', 
    ->   part_size AS 'Blob Part Size' 
    -> FROM ndbinfo.blobs 
    -> WHERE table_name = 't1';
+-------------+-------------+----------------+
| Column Name | Inline Size | Blob Part Size |
+-------------+-------------+----------------+
| b1          |         256 |           2000 |
| b2          |        8000 |           2000 |
+-------------+-------------+----------------+
2 rows in set (0.01 sec)

You can also check the output from the ndb_desc utility, as shown here, with the relevant lines displayed using emphasized text:

$> ndb_desc -d test t1
-- t --
Version: 1
Fragment type: HashMapPartition
K Value: 6
Min load factor: 78
Max load factor: 80
Temporary table: no
Number of attributes: 3
Number of primary keys: 1
Length of frm data: 945
Max Rows: 0
Row Checksum: 1
Row GCI: 1
SingleUserMode: 0
ForceVarPart: 1
PartitionCount: 2
FragmentCount: 2
PartitionBalance: FOR_RP_BY_LDM
ExtraRowGciBits: 0
ExtraRowAuthorBits: 0
TableStatus: Retrieved
Table options: readbackup
HashMap: DEFAULT-HASHMAP-3840-2
-- Attributes --
a Int PRIMARY KEY DISTRIBUTION KEY AT=FIXED ST=MEMORY
b1 Blob(256,2000,0) NULL AT=MEDIUM_VAR ST=MEMORY BV=2 BT=NDB$BLOB_64_1
b2 Blob(8000,2000,0) NULL AT=MEDIUM_VAR ST=MEMORY BV=2 BT=NDB$BLOB_64_2
-- Indexes -- 
PRIMARY KEY(a) - UniqueHashIndex
PRIMARY(a) - OrderedIndex

BLOB_INLINE_SIZE has no effect on TINYBLOB columns. In NDB 8.0.41 and later, it is disallowed with TINYBLOB, and causes a warning if used.

For MAX_BLOB_PART_SIZE, the = sign and the value following it are optional. Using any value other than 0 or 1 results in a syntax error.

The effect of using MAX_BLOB_PART_SIZE in a column comment is to set the blob part size of a TEXT or BLOB column to the maximum number of bytes supported for this by NDB (13948). This option can be applied to any blob column type supported by MySQL except TINYBLOB or TINYTEXT (BLOB, MEDIUMBLOB, LONGBLOB, TEXT, MEDIUMTEXT, LONGTEXT). Unlike BLOB_INLINE_SIZE, MAX_BLOB_PART_SIZE has no effect on JSON columns.

To see the effects of this option, we first run the following SQL statement in the mysql client to create a table with two BLOB columns, one (c1) with no extra options, and another (c2) with MAX_BLOB_PART_SIZE:

mysql> CREATE TABLE test.t2 (
    ->   p INT PRIMARY KEY, 
    ->   c1 BLOB, 
    ->   c2 BLOB COMMENT 'NDB_COLUMN=MAX_BLOB_PART_SIZE'
    -> ) ENGINE NDB;
Query OK, 0 rows affected (0.32 sec)

From the system shell, run the ndb_desc utility to obtain information about the table just created, as shown in this example:

$> ndb_desc -d test t2
-- t --
Version: 1
Fragment type: HashMapPartition
K Value: 6
Min load factor: 78
Max load factor: 80
Temporary table: no
Number of attributes: 3
Number of primary keys: 1
Length of frm data: 324
Row Checksum: 1
Row GCI: 1
SingleUserMode: 0
ForceVarPart: 1
FragmentCount: 2
ExtraRowGciBits: 0
ExtraRowAuthorBits: 0
TableStatus: Retrieved
HashMap: DEFAULT-HASHMAP-3840-2
-- Attributes --
p Int PRIMARY KEY DISTRIBUTION KEY AT=FIXED ST=MEMORY
c1 Blob(256,2000,0) NULL AT=MEDIUM_VAR ST=MEMORY BV=2 BT=NDB$BLOB_22_1
c2 Blob(256,13948,0) NULL AT=MEDIUM_VAR ST=MEMORY BV=2 BT=NDB$BLOB_22_2
-- Indexes -- 
PRIMARY KEY(p) - UniqueHashIndex
PRIMARY(p) - OrderedIndex

Column information in the output is listed under Attributes; for columns c1 and c2 it is displayed here in emphasized text. For c1, the blob part size is 2000, the default value; for c2, it is 13948, as set by MAX_BLOB_PART_SIZE.

You can also query the ndbinfo.blobs table to see this, as shown here:

mysql> SELECT 
    ->   column_name AS 'Column Name', 
    ->   inline_size AS 'Inline Size', 
    ->   part_size AS 'Blob Part Size' 
    -> FROM ndbinfo.blobs 
    -> WHERE table_name = 't2';
+-------------+-------------+----------------+
| Column Name | Inline Size | Blob Part Size |
+-------------+-------------+----------------+
| c1          |         256 |           2000 |
| c2          |         256 |          13948 |
+-------------+-------------+----------------+
2 rows in set (0.00 sec)

You can change the blob part size for a given blob column of an NDB table using an ALTER TABLE statement such as this one, and verifying the changes afterwards using SHOW CREATE TABLE:

mysql> ALTER TABLE test.t2 
    ->    DROP COLUMN c1, 
    ->     ADD COLUMN c1 BLOB COMMENT 'NDB_COLUMN=MAX_BLOB_PART_SIZE',
    ->     CHANGE COLUMN c2 c2 BLOB AFTER c1;
Query OK, 0 rows affected (0.47 sec)
Records: 0  Duplicates: 0  Warnings: 0

mysql> SHOW CREATE TABLE test.t2\G
*************************** 1. row ***************************
       Table: t
Create Table: CREATE TABLE `t2` (
  `p` int(11) NOT NULL,
  `c1` blob COMMENT 'NDB_COLUMN=MAX_BLOB_PART_SIZE',
  `c2` blob,
  PRIMARY KEY (`p`)
) ENGINE=ndbcluster DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_0900_ai_ci
1 row in set (0.00 sec)

mysql> EXIT
Bye

The output of ndb_desc shows that the blob part sizes of the columns have been changed as expected:

$> ndb_desc -d test t2
-- t --
Version: 16777220
Fragment type: HashMapPartition
K Value: 6
Min load factor: 78
Max load factor: 80
Temporary table: no
Number of attributes: 3
Number of primary keys: 1
Length of frm data: 324
Row Checksum: 1
Row GCI: 1
SingleUserMode: 0
ForceVarPart: 1
FragmentCount: 2
ExtraRowGciBits: 0
ExtraRowAuthorBits: 0
TableStatus: Retrieved
HashMap: DEFAULT-HASHMAP-3840-2
-- Attributes --
p Int PRIMARY KEY DISTRIBUTION KEY AT=FIXED ST=MEMORY
c1 Blob(256,13948,0) NULL AT=MEDIUM_VAR ST=MEMORY BV=2 BT=NDB$BLOB_26_1
c2 Blob(256,2000,0) NULL AT=MEDIUM_VAR ST=MEMORY BV=2 BT=NDB$BLOB_26_2
-- Indexes -- 
PRIMARY KEY(p) - UniqueHashIndex
PRIMARY(p) - OrderedIndex

You can also see the change by running the query against ndbinfo.blobs again:

mysql> SELECT 
    ->   column_name AS 'Column Name', 
    ->   inline_size AS 'Inline Size', 
    ->   part_size AS 'Blob Part Size' 
    -> FROM ndbinfo.blobs 
    -> WHERE table_name = 't2';
+-------------+-------------+----------------+
| Column Name | Inline Size | Blob Part Size |
+-------------+-------------+----------------+
| c1          |         256 |          13948 |
| c2          |         256 |           2000 |
+-------------+-------------+----------------+
2 rows in set (0.00 sec)

It is possible to set both BLOB_INLINE_SIZE and MAX_BLOB_PART_SIZE for a blob column, as shown in this CREATE TABLE statement:

mysql> CREATE TABLE test.t3 (
    ->   p INT NOT NULL PRIMARY KEY,
    ->   c1 JSON,
    ->   c2 JSON COMMENT 'NDB_COLUMN=BLOB_INLINE_SIZE=5000,MAX_BLOB_PART_SIZE'
    -> ) ENGINE NDB;
Query OK, 0 rows affected (0.28 sec)

Querying the blobs table shows us that the statement worked as expected:

mysql> SELECT 
    ->   column_name AS 'Column Name', 
    ->   inline_size AS 'Inline Size', 
    ->   part_size AS 'Blob Part Size' 
    -> FROM ndbinfo.blobs 
    -> WHERE table_name = 't3';
+-------------+-------------+----------------+
| Column Name | Inline Size | Blob Part Size |
+-------------+-------------+----------------+
| c1          |        4000 |           8100 |
| c2          |        5000 |           8100 |
+-------------+-------------+----------------+
2 rows in set (0.00 sec)

You can also verify that the statement worked by checking the output of ndb_desc.

Changing a column's blob part size must be done using a copying ALTER TABLE; this operation cannot be performed online (see Section 25.6.12, “Online Operations with ALTER TABLE in NDB Cluster”).

For more information about how NDB stores columns of blob types, see String Type Storage Requirements.

NDB_TABLE Options

For an NDB Cluster table, the table comment in a CREATE TABLE or ALTER TABLE statement can also be used to specify an NDB_TABLE option, which consists of one or more name-value pairs, separated by commas if need be, following the string NDB_TABLE=. Complete syntax for names and values syntax is shown here:

COMMENT="NDB_TABLE=ndb_table_option[,ndb_table_option[,...]]"

ndb_table_option: {
    NOLOGGING={1 | 0}
  | READ_BACKUP={1 | 0}
  | PARTITION_BALANCE={FOR_RP_BY_NODE | FOR_RA_BY_NODE | FOR_RP_BY_LDM
                      | FOR_RA_BY_LDM | FOR_RA_BY_LDM_X_2
                      | FOR_RA_BY_LDM_X_3 | FOR_RA_BY_LDM_X_4}
  | FULLY_REPLICATED={1 | 0}
}

Spaces are not permitted within the quoted string. The string is case-insensitive.

The four NDB table options that can be set as part of a comment in this way are described in more detail in the next few paragraphs.

NOLOGGING: By default, NDB tables are logged, and checkpointed. This makes them durable to whole cluster failures. Using NOLOGGING when creating or altering a table means that this table is not redo logged or included in local checkpoints. In this case, the table is still replicated across the data nodes for high availability, and updated using transactions, but changes made to it are not recorded in the data node's redo logs, and its content is not checkpointed to disk; when recovering from a cluster failure, the cluster retains the table definition, but none of its rows—that is, the table is empty.

Using such nonlogging tables reduces the data node's demands on disk I/O and storage, as well as CPU for checkpointing CPU. This may be suitable for short-lived data which is frequently updated, and where the loss of all data in the unlikely event of a total cluster failure is acceptable.

It is also possible to use the ndb_table_no_logging system variable to cause any NDB tables created or altered while this variable is in effect to behave as though it had been created with the NOLOGGING comment. Unlike when using the comment directly, there is nothing in this case in the output of SHOW CREATE TABLE to indicate that it is a nonlogging table. Using the table comment approach is recommended since it offers per-table control of the feature, and this aspect of the table schema is embedded in the table creation statement where it can be found easily by SQL-based tools.

READ_BACKUP: Setting this option to 1 has the same effect as though ndb_read_backup were enabled; enables reading from any replica. Doing so greatly improves the performance of reads from the table at a relatively small cost to write performance. Beginning with NDB 8.0.19, 1 is the default for READ_BACKUP, and the default for ndb_read_backup is ON (previously, read from any replica was disabled by default).

You can set READ_BACKUP for an existing table online, using an ALTER TABLE statement similar to one of those shown here:

ALTER TABLE ... ALGORITHM=INPLACE, COMMENT="NDB_TABLE=READ_BACKUP=1";

ALTER TABLE ... ALGORITHM=INPLACE, COMMENT="NDB_TABLE=READ_BACKUP=0";

For more information about the ALGORITHM option for ALTER TABLE, see Section 25.6.12, “Online Operations with ALTER TABLE in NDB Cluster”.

PARTITION_BALANCE: Provides additional control over assignment and placement of partitions. The following four schemes are supported:

PARTITION_BALANCE is the preferred interface for setting the number of partitions per table. Using MAX_ROWS to force the number of partitions is deprecated but continues to be supported for backward compatibility; it is subject to removal in a future release of MySQL NDB Cluster. (Bug #81759, Bug #23544301)

FULLY_REPLICATED controls whether the table is fully replicated, that is, whether each data node has a complete copy of the table. To enable full replication of the table, use FULLY_REPLICATED=1.

This setting can also be controlled using the ndb_fully_replicated system variable. Setting it to ON enables the option by default for all new NDB tables; the default is OFF. The ndb_data_node_neighbour system variable is also used for fully replicated tables, to ensure that when a fully replicated table is accessed, we access the data node which is local to this MySQL Server.

An example of a CREATE TABLE statement using such a comment when creating an NDB table is shown here:

mysql> CREATE TABLE t1 (
     >     c1 INT NOT NULL AUTO_INCREMENT PRIMARY KEY,
     >     c2 VARCHAR(100),
     >     c3 VARCHAR(100) )
     > ENGINE=NDB
     >
COMMENT="NDB_TABLE=READ_BACKUP=0,PARTITION_BALANCE=FOR_RP_BY_NODE";

The comment is displayed as part of the output of SHOW CREATE TABLE. The text of the comment is also available from querying the MySQL Information Schema TABLES table, as in this example:

mysql> SELECT TABLE_NAME, TABLE_SCHEMA, TABLE_COMMENT
     > FROM INFORMATION_SCHEMA.TABLES WHERE TABLE_NAME="t1"\G
*************************** 1. row ***************************
   TABLE_NAME: t1
 TABLE_SCHEMA: test
TABLE_COMMENT: NDB_TABLE=READ_BACKUP=0,PARTITION_BALANCE=FOR_RP_BY_NODE
1 row in set (0.01 sec)

This comment syntax is also supported with ALTER TABLE statements for NDB tables, as shown here:

mysql> ALTER TABLE t1 COMMENT="NDB_TABLE=PARTITION_BALANCE=FOR_RA_BY_NODE";
Query OK, 0 rows affected (0.40 sec)
Records: 0  Duplicates: 0  Warnings: 0

Beginning with NDB 8.0.21, the TABLE_COMMENT column displays the comment that is required to re-create the table as it is following the ALTER TABLE statement, like this:

mysql> SELECT TABLE_NAME, TABLE_SCHEMA, TABLE_COMMENT
    ->     FROM INFORMATION_SCHEMA.TABLES WHERE TABLE_NAME="t1"\G
*************************** 1. row ***************************
   TABLE_NAME: t1
 TABLE_SCHEMA: test
TABLE_COMMENT: NDB_TABLE=READ_BACKUP=0,PARTITION_BALANCE=FOR_RP_BY_NODE
1 row in set (0.01 sec)

mysql> SELECT TABLE_NAME, TABLE_SCHEMA, TABLE_COMMENT
     > FROM INFORMATION_SCHEMA.TABLES WHERE TABLE_NAME="t1";
+------------+--------------+--------------------------------------------------+
| TABLE_NAME | TABLE_SCHEMA | TABLE_COMMENT                                    |
+------------+--------------+--------------------------------------------------+
| t1         | c            | NDB_TABLE=PARTITION_BALANCE=FOR_RA_BY_NODE       |
| t1         | d            |                                                  |
+------------+--------------+--------------------------------------------------+
2 rows in set (0.01 sec)

Keep in mind that a table comment used with ALTER TABLE replaces any existing comment which the table might have.

mysql> ALTER TABLE t1 COMMENT="NDB_TABLE=PARTITION_BALANCE=FOR_RA_BY_NODE";
Query OK, 0 rows affected (0.40 sec)
Records: 0  Duplicates: 0  Warnings: 0

mysql> SELECT TABLE_NAME, TABLE_SCHEMA, TABLE_COMMENT
     > FROM INFORMATION_SCHEMA.TABLES WHERE TABLE_NAME="t1";
+------------+--------------+--------------------------------------------------+
| TABLE_NAME | TABLE_SCHEMA | TABLE_COMMENT                                    |
+------------+--------------+--------------------------------------------------+
| t1         | c            | NDB_TABLE=PARTITION_BALANCE=FOR_RA_BY_NODE       |
| t1         | d            |                                                  |
+------------+--------------+--------------------------------------------------+
2 rows in set (0.01 sec)

Prior to NDB 8.0.21, the table comment used with ALTER TABLE replaced any existing comment which the table might have had. This meant that (for example) the READ_BACKUP value was not carried over to the new comment set by the ALTER TABLE statement, and that any unspecified values reverted to their defaults. (BUG#30428829) There was thus no longer any way using SQL to retrieve the value previously set for the comment. To keep comment values from reverting to their defaults, it was necessary to preserve any such values from the existing comment string and include them in the comment passed to ALTER TABLE.

You can also see the value of the PARTITION_BALANCE option in the output of ndb_desc. ndb_desc also shows whether the READ_BACKUP and FULLY_REPLICATED options are set for the table. See the description of this program for more information.