This section discusses the manner in which NDB Cluster divides and duplicates data for storage.
A number of concepts central to an understanding of this topic are discussed in the next few paragraphs.
Data node. An ndbd or ndbmtd process, which stores one or more fragment replicas—that is, copies of the partitions (discussed later in this section) assigned to the node group of which the node is a member.
Each data node should be located on a separate computer. While it is also possible to host multiple data node processes on a single computer, such a configuration is not usually recommended.
It is common for the terms “node” and “data node” to be used interchangeably when referring to an ndbd or ndbmtd process; where mentioned, management nodes (ndb_mgmd processes) and SQL nodes (mysqld processes) are specified as such in this discussion.
The number of node groups in an NDB Cluster is not directly
configurable; it is a function of the number of data nodes and of
the number of fragment replicas
configuration parameter), as shown here:
[# of node groups] = [# of data nodes] / NoOfReplicas
Thus, an NDB Cluster with 4 data nodes has 4 node groups if
NoOfReplicas is set to 1
config.ini file, 2 node groups if
NoOfReplicas is set to 2,
and 1 node group if
NoOfReplicas is set to 4.
Fragment replicas are discussed later in this section; for more
Section 126.96.36.199, “Defining NDB Cluster Data Nodes”.
All node groups in an NDB Cluster must have the same number of data nodes.
You can add new node groups (and thus new data nodes) online, to a running NDB Cluster; see Section 18.6.7, “Adding NDB Cluster Data Nodes Online”, for more information.
Partition. This is a portion of the data stored by the cluster. Each node is responsible for keeping at least one copy of any partitions assigned to it (that is, at least one fragment replica) available to the cluster.
The number of partitions used by default by NDB Cluster depends on the number of data nodes and the number of LDM threads in use by the data nodes, as shown here:
[# of partitions] = [# of data nodes] * [# of LDM threads]
When using data nodes running ndbmtd, the
number of LDM threads is controlled by the setting for
When using ndbd there is a single LDM thread,
which means that there are as many cluster partitions as nodes
participating in the cluster. This is also the case when using
MaxNoOfExecutionThreads set to 3 or less. (You
should be aware that the number of LDM threads increases with the
value of this parameter, but not in a strictly linear fashion, and
that there are additional constraints on setting it; see the
for more information.)
NDB and user-defined partitioning.
NDB Cluster normally partitions
NDBCLUSTER tables automatically.
However, it is also possible to employ user-defined partitioning
NDBCLUSTER tables. This is
subject to the following limitations:
LINEAR KEYpartitioning schemes are supported in production with
The maximum number of partitions that may be defined explicitly for any
8 * [, the number of node groups in an NDB Cluster being determined as discussed previously in this section. When running ndbd for data node processes, setting the number of LDM threads has no effect (since
number of LDM threads] * [
number of node groups]
ThreadConfigapplies only to ndbmtd); in such cases, this value can be treated as though it were equal to 1 for purposes of performing this calculation.
See Section 18.5.3, “ndbmtd — The NDB Cluster Data Node Daemon (Multi-Threaded)”, for more information.
For more information relating to NDB Cluster and user-defined partitioning, see Section 18.2.7, “Known Limitations of NDB Cluster”, and Section 19.6.2, “Partitioning Limitations Relating to Storage Engines”.
Fragment Replica. This is a copy of a cluster partition. Each node in a node group stores a replica. Also sometimes known as a partition replica. The number of fragment replicas is equal to the number of nodes per node group.
A fragment replica belongs entirely to a single node; a node can (and usually does) store several fragment replicas.
The following diagram illustrates an NDB Cluster with four data nodes running ndbd, arranged in two node groups of two nodes each; nodes 1 and 2 belong to node group 0, and nodes 3 and 4 belong to node group 1.
Only data nodes are shown here; although a working NDB Cluster requires an ndb_mgmd process for cluster management and at least one SQL node to access the data stored by the cluster, these have been omitted from the figure for clarity.
The data stored by the cluster is divided into four partitions, numbered 0, 1, 2, and 3. Each partition is stored—in multiple copies—on the same node group. Partitions are stored on alternate node groups as follows:
Partition 0 is stored on node group 0; a primary fragment replica (primary copy) is stored on node 1, and a backup fragment replica (backup copy of the partition) is stored on node 2.
Partition 1 is stored on the other node group (node group 1); this partition's primary fragment replica is on node 3, and its backup fragment replica is on node 4.
Partition 2 is stored on node group 0. However, the placing of its two fragment replicas is reversed from that of Partition 0; for Partition 2, the primary fragment replica is stored on node 2, and the backup on node 1.
Partition 3 is stored on node group 1, and the placement of its two fragment replicas are reversed from those of partition 1. That is, its primary fragment replica is located on node 4, with the backup on node 3.
What this means regarding the continued operation of an NDB Cluster is this: so long as each node group participating in the cluster has at least one node operating, the cluster has a complete copy of all data and remains viable. This is illustrated in the next diagram.
In this example, the cluster consists of two node groups each consisting of two data nodes. Each data node is running an instance of ndbd. Any combination of at least one node from node group 0 and at least one node from node group 1 is sufficient to keep the cluster “alive”. However, if both nodes from a single node group fail, the combination consisting of the remaining two nodes in the other node group is not sufficient. In this situation, the cluster has lost an entire partition and so can no longer provide access to a complete set of all NDB Cluster data.