Changes in MySQL NDB Cluster 9.1.0 (2024-10-16, Innovation Release)

MySQL NDB Cluster 9.1.0 is a new Innovation release of NDB Cluster, based on MySQL Server 9.1 and including features in version 9.1 of the NDB storage engine, as well as fixing recently discovered bugs in previous NDB Cluster releases.

Obtaining MySQL NDB Cluster 9.1.  NDB Cluster 9.1 source code and binaries can be obtained from https://dev.mysql.com/downloads/cluster/.

For an overview of major changes made in NDB Cluster 9.1, see What is New in MySQL NDB Cluster 9.1.

This release also incorporates all bug fixes and changes made in previous NDB Cluster releases, as well as all bug fixes and feature changes which were added in mainline MySQL 9 through MySQL 9.1.0 (see Changes in MySQL 9.1.0 (2024-10-15, Innovation Release)).

• Deprecation and Removal Notes

• Functionality Added or Changed

• Bugs Fixed

Deprecation and Removal Notes

• Use of an Ndb.cfg file for setting the connection string for an NDB process was not well documented or supported. With this release, this file is now formally deprecated, and a warning is issued whenever it is used. You should expect support for this file to be removed in a future release of MySQL Cluster. (WL #15765)

Functionality Added or Changed

• The ndbcluster plugin subscribes to all changes that occur in NDB and writes them epoch by epoch to the binary log. Each epoch received from NDB consists of a large number of changes, all of which are written to the binary log transaction cache before flushing them to the binary log. Previously, it was possible to configure the cache size for all threads, which often led to improper resource allocation for a MySQL Server used for writing a binary log of changes for NDB.

  To enable dimensioning and configuring the system properly, we introduce a new system variable ndb_log_cache_size which makes it possible to set the size of the transaction cache used by the NDB binary log injector, so that this size can be set separately for writing the binary log for NDB transactions and (using binlog_cache_size) for writing other transactions whose sizes are likely to be smaller. (Bug #36694848)

Bugs Fixed

• NDB Cluster APIs: Using NdbRecord and OO_SETVALUE from the NDB API to write the value of a Varchar, Varbinary, Longvarchar, or Longvarbinary column failed with error 829. (Bug #36989337)

• MySQL NDB ClusterJ: References to ClusterJPA and OpenJPA have been removed from the comments in the packaging files, as JPA code was already removed from ClusterJ some time ago. (Bug #36725675)

• MySQL NDB ClusterJ: ReconnectTest in the ClusterJ test suite failed sometimes due to a race condition. The test has been rewritten with proper synchronization. (Bug #28550140)

• Removed node management code from TRIX that was not actually used. (Bug #37006547)

• Submitting concurrent shutdown commands for individual nodes using ndb_mgm SHUTDOWN node_id or the MGM API sometimes had one or both of the following adverse results:

  • Cluster failure when all nodes in the same node group were stopped

  • Inability to recover when all nodes in the same node group were stopped, and the cluster had more than one node group

  This was due to the fact that the (planned) shutdown of a single node assumed that only one such shutdown occurred at a time, but did not actually check this limitation.

  We fix this so that concurrent single-node shutdown requests are serialized across the cluster, and any which would cause a cluster outage are rejected. (Bug #36943756)

  References: See also: Bug #36839995.

• Shutdown of a data node late in a schema transaction updating index statistics caused the president node to shut down as well. (Bug #36886242)

  References: See also: Bug #36877952.

• It was possible for duplicate events to be sent to user applications when a data node was shut down. (Bug #36750146)

• BLOB_INLINE_SIZE=0 set within a column comment was not honored, and the default for the blob type was used instead (such as 256 bytes for BLOB).

  See NDB_COLUMN Options, for more information. (Bug #36724336)

• Issues arose when an attempt was made to use a SHM transporter's wakeup socket before it was ready, due in part to error-handling when setting up the SHM transporter, which did not close the socket correctly prior to making another attempt at setup. (Bug #36568752, Bug #36623058)

• An error in a my.cnf file could cause the management node to shut down unexpectedly. (Bug #36508565)

• A race condition sometimes occurred between the watchdog thread and the signal execution thread trying to start node failure handling in parallel. (Bug #35728261)