Traditional MySQL replication architectures (active-passive or source-replica topologies) introduce replication lag and require failover orchestrators (e.g., Orchestrator or ProxySQL) to handle master crashes.

MySQL NDB Cluster is a shared-nothing, distributed database that provides real-time in-memory performance, 99.999% high availability, and active-active multi-master read/write capabilities across all SQL nodes.

This guide details configuring and scaling a multi-master NDB cluster using Oracle MySQL Cluster Manager (MCM).

Core Architecture & Node Roles

An NDB cluster consists of three distinct node types:

  1. Management Nodes (ndb_mgmd): Distribute cluster configuration, manage node handshakes, handle arbitrations, and collect cluster logs.
  2. Data Nodes (ndbd / ndbmtd): Store table data, indexes, and partitions in memory (or disk-data tablespaces). ndbmtd is the high-performance multi-threaded data engine.
  3. SQL / API Nodes (mysqld): Stateless MySQL server daemons that parse SQL queries from application load balancers and interact with data nodes via the NDB storage engine.
               [ Application Layer / Load Balancer ]
                                │
        ┌───────────────────────┴───────────────────────┐
        ▼                                               ▼
 [SQL Node 1: 172.31.16.34]                   [SQL Node 2: 172.31.16.43]
 (mysqld + ndb_mgmd)                         (mysqld + ndb_mgmd)
        │                                               │
        └───────────────────────┬───────────────────────┘
                                ▼
         NDB Storage Engine Network (High-Speed Backplane)
                                │
        ┌───────────────────────┴───────────────────────┐
        ▼                                               ▼
 [Data Node 1: 172.31.23.137]                [Data Node 2: 172.31.24.183]
 (ndbmtd - Node Group 0)                     (ndbmtd - Node Group 0)

Step 1: Initializing the MCM Agents

Install the MySQL Cluster Manager daemon (mcmd) across all participant nodes and start the agent:

# Executed on all cluster hosts
./bin/mcmd --bootstrap

Once running, launch the MCM interactive client (mcm):

./bin/mcm -a localhost:1862

Step 2: Creating the Site and Registering Packages

Group your server hosts into a managed site:

mcm> create site -h 172.31.23.137,172.31.24.183,172.31.16.34,172.31.16.43 mysite;

Register the NDB cluster software package binary path:

mcm> add package -b /home/ops/mcm/cluster 7_6_8;

Step 3: Defining Node Topology and Starting the Cluster

Define node assignments mapping physical IPs to processes:

mcm> create cluster -P 7_6_8 \
     -R ndbd@172.31.23.137,ndbd@172.31.24.183,\
        ndb_mgmd@172.31.16.43,ndb_mgmd@172.31.16.34,\
        mysqld@172.31.16.34,mysqld@172.31.16.43 \
     mycluster;

Launch the cluster across all nodes:

mcm> start cluster mycluster;
mcm> show status -r mycluster;

Status Verification:

+--------+----------+---------------+---------+-----------+
| NodeId | Process  | Host          | Status  | Nodegroup |
+--------+----------+---------------+---------+-----------+
| 49     | ndb_mgmd | 172.31.16.43  | running |           |
| 50     | ndb_mgmd | 172.31.16.34  | running |           |
| 1      | ndbd     | 172.31.23.137 | running | 0         |
| 2      | ndbd     | 172.31.24.183 | running | 0         |
| 51     | mysqld   | 172.31.16.34  | running |           |
| 52     | mysqld   | 172.31.16.43  | running |           |
+--------+----------+---------------+---------+-----------+

Step 4: Synchronizing User Privileges Across All SQL Nodes

Because the mysql.user authentication tables traditionally run under standard storage engines rather than NDB, user privileges created on SQL Node 1 do not automatically mirror to SQL Node 2 by default.

Enable distributed privileges across the cluster by loading the NDB privilege distribution script on one SQL node:

mysql -u root -p -h 127.0.0.1 < cluster/share/ndb_dist_priv.sql

User grants (CREATE USER, GRANT) are now automatically distributed and kept consistent across all SQL nodes in real-time.

Step 5: Online Upgrades to Multi-Threaded Engine (ndbmtd)

To maximize CPU core utilization on multi-core data servers, convert the single-threaded ndbd engine to multi-threaded ndbmtd on the fly:

mcm> change process ndbd=ndbmtd mycluster;

MCM performs a rolling rolling restart across the cluster without interrupting client read/write transactions.