17.4.3 ndbmtd — The MySQL Cluster Data Node Daemon (Multi-Threaded)

ndbmtd is a multi-threaded version of ndbd, the process that is used to handle all the data in tables using the NDBCLUSTER storage engine. ndbmtd is intended for use on host computers having multiple CPU cores. Except where otherwise noted, ndbmtd functions in the same way as ndbd; therefore, in this section, we concentrate on the ways in which ndbmtd differs from ndbd, and you should consult Section 17.4.1, “ndbd — The MySQL Cluster Data Node Daemon”, for additional information about running MySQL Cluster data nodes that apply to both the single-threaded and multi-threaded versions of the data node process.

Command-line options and configuration parameters used with ndbd also apply to ndbmtd. For more information about these options and parameters, see Section 17.4.1, “ndbd — The MySQL Cluster Data Node Daemon”, and Section, “Defining MySQL Cluster Data Nodes”, respectively.

ndbmtd is also file system-compatible with ndbd. In other words, a data node running ndbd can be stopped, the binary replaced with ndbmtd, and then restarted without any loss of data. (However, when doing this, you must make sure that MaxNoOfExecutionThreads is set to an apppriate value before restarting the node if you wish for ndbmtd to run in multi-threaded fashion.) Similarly, an ndbmtd binary can be replaced with ndbd simply by stopping the node and then starting ndbd in place of the multi-threaded binary. It is not necessary when switching between the two to start the data node binary using --initial.

Prior to MySQL Cluster NDB 7.0.6, there were known issues when using ndbmtd with MySQL Cluster Disk Data tables. If you wish to use multi-threaded data nodes with disk-based NDB tables, you should ensure that you are running MySQL Cluster NDB 7.0.6 or later (see Bug #41915 and Bug #44915)

Using ndbmtd differs from using ndbd in two key respects:

  1. Because ndbmtd runs by default in single-threaded mode (that is, it behaves like ndbd), you must configure it to use multiple threads. This can be done by setting an appropriate value in the config.ini file for the MaxNoOfExecutionThreads configuration parameter. For more information about this parameter and its use, see Multi-Threading Configuration Parameters (ndbmtd).

  2. Trace files are generated by critical errors in ndbmtd processes in a somewhat different fashion from how these are generated by ndbd failures. These differences are discussed in more detail in the next few paragraphs.

Like ndbd, ndbmtd generates a set of log files which are placed in the directory specified by DataDir in the config.ini configuration file. Except for trace files, these are generated in the same way and have the same names as those generated by ndbd.

In the event of a critical error, ndbmtd generates trace files describing what happened just prior to the error' occurrence. These files, which can be found in the data node's DataDir, are useful for analysis of problems by the MySQL Cluster Development and Support teams. One trace file is generated for each ndbmtd thread. The names of these files have the following pattern:


In this pattern, node_id stands for the data node's unique node ID in the cluster, trace_id is a trace sequence number, and thread_id is the thread ID. For example, in the event of the failure of an ndbmtd process running as a MySQL Cluster data node having the node ID 3 and with MaxNoOfExecutionThreads equal to 4, four trace files are generated in the data node's data directory. If the is the first time this node has failed, then these files are named ndb_3_trace.log.1_t1, ndb_3_trace.log.1_t2, ndb_3_trace.log.1_t3, and ndb_3_trace.log.1_t4. Internally, these trace files follow the same format as ndbd trace files.

The ndbd exit codes and messages that are generated when a data node process shuts down prematurely are also used by ndbmtd. See ndbd Error Messages, for a listing of these.


It is possible to use ndbd and ndbmtd concurrently on different data nodes in the same MySQL Cluster. However, such configurations have not been tested extensively; thus, we cannot recommend doing so in a production setting at this time.