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. 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. NDB
supports two column comment options
BLOB_INLINE_SIZE
and
MAX_BLOB_PART_SIZE
. Syntax for these options
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
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. 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:
FOR_RP_BY_NODE
: One partition per node.Only one LDM on each node stores a primary partition. Each partition is stored in the same LDM (same ID) on all nodes.
FOR_RA_BY_NODE
: One partition per node group.Each node stores a single partition, which can be either a primary replica or a backup replica. Each partition is stored in the same LDM on all nodes.
FOR_RP_BY_LDM
: One partition for each LDM on each node; the default.This is the setting used if
READ_BACKUP
is set to 1.FOR_RA_BY_LDM
: One partition per LDM in each node group.These partitions can be primary or backup partitions.
FOR_RA_BY_LDM_X_2
: Two partitions per LDM in each node group.These partitions can be primary or backup partitions.
FOR_RA_BY_LDM_X_3
: Three partitions per LDM in each node group.These partitions can be primary or backup partitions.
FOR_RA_BY_LDM_X_4
: Four partitions per LDM in each node group.These partitions can be primary or backup partitions.
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
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)
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.