Changes in MySQL Cluster Manager 1.4.8 (2020-01-24)

• Experimental support for MySQL NDB Cluster 8.0

• Functionality Added or Changed

• Bugs Fixed

Experimental support for MySQL NDB Cluster 8.0

• MySQL Cluster Manager now supports NDB Cluster 8.0 on an experimental basis; see Supported Platforms and MySQL NDB Cluster Versions for limitations.

Functionality Added or Changed

• Incompatible Change; Agent: When managing NDB Cluster 8.0.19 and later, the restore cluster command no longer restores shared users and grants to the mysql.ndb_sql_metadata table by default. A new command option, --include-stored-grants, has been added for overriding this behavior and enabling the restore of shared user and grant data and metadata. (Bug #30375672)

• Packaging: MySQL Cluster Manager is now built and shipped with OpenSSL 1.1.1d. (Bug #30612140)

• Agent: By default, a rolling restart for a cluster is performed on the nodes in a parallel manner (that is, half of the nodes are stopped and restarted together, followed by the second half of the nodes). Now, rolling restarts can be performed in a sequential manner by using the --sequential-restart option with commands that perform rolling restarts, in which case the nodes are stopped and restarted one after another. The option can be used with the restart cluster, change process, add process, set, reset, and autotune commands. (Bug #30062392)

• Agent: Some parameters could only be set before the initial start of a node (for example, NoOfFragmentLogparts); it then became impossible to set them after an initial start of a cluster had failed. It is now possible to set such parameters again, as long as the node involved had not yet been assigned to any nodegroup during the failed initial start. (Bug #29718127)

• Agent: Following changes in NDB Cluster 8.0.18, the maximum number of data nodes supported by a NDB 8.0 cluster managed by MySQL Cluster Manager is now 145, and the maximum number of node groups supported is now 72. With that change, when a NDB 8.0 Cluster is bootstrapped by mcmd with the --bootstrap option, ndb_mgmd, mysqld, and ndbapi nodes are assigned node IDs of 145 and above. (Bug #29459428)

• Client: The --disable-indexes option for the restore cluster command has been removed, to avoid having the cluster restored into a bad state. (Bug #30728256)

• Client: The message for ERROR 6020 has been expanded to include the reason why a configuration parameter cannot be set given the current process states. (Bug #30526906)

• Client: The log message for a failed connection to mcmd by an mcm client has been expanded, warning users to make sure the correct hostname or IP address is specified, and that no firewall is blocking connection attempts. (Bug #30288338)

Bugs Fixed

• Agent: The update process command failed when no ndb_mgmd node was running. It was due to an excessive restriction asserted by mcmd that this patch has relieved, so that a running ndb_mgmd node is required now only when update process is applied to a data node or mysqld node. (Bug #30723002)

• Agent: During an MySQL Cluster Manager upgrade, when mcmd was upgrading its repository. it was over aggressive in removing old configuration objects that did not need to be upgraded, resulting sometimes in inconsistency of the repository and mcmd quitting unexpectedly. With this fix, mcmd only deletes configuration objects that it is supposed to. (Bug #30666263)

• Agent: When running the import cluster command, there were a lot of unnecessary critical messages issued to the agent log. (Bug #30572723)

• Agent: A cluster restart failed after the cluster has been imported by mcmd if the Arbitration parameter was specified in the config.ini file of the imported cluster. It was because the import config command, which was always run before the import cluster command, imported the parameter's value wrongly, and that has been corrected by this fix. (Bug #30505003)

• Agent: After an add process command failed and the remove process command was used to clean up the processes that were not successfully added, a subsequent restart cluster command caused mcmd to quite unexpectedly. With this fix, mcmd issues a warning in the situation and then finishes the restart. (Bug #30498301)

• Agent: When the stop cluster command was run and mcmd had to force a mysqld node to stop, the node got stuck in the stopping state, while mcm reported that the cluster had been stopped successfully. (Bug #30375807)

• Agent: The restore cluster command timed out when the cluster to be backed up contained more than 2000 schemas. With this fix, the mcmd agents are informed of the progress of the restore on a more frequent basis, so that even if the process takes a longer time to finish, the agents would wait instead of timing out. (Bug #30360453)

• Agent: When running the restore cluster command, a schema or table name containing a comma was taken by mcmd as a comma-separated list of multiple schemas or tables, resulting in the schema or table being restored wrongly. With this fix, mcmd honors the comma as part of the schema or table name. (Bug #30360401)

• Agent: When managing an NDB 8.0 cluster, during the initial startup of the cluster, the start cluster command sometimes failed due to a failure to create the mcmd user on the cluster. That was because while the grants for the mcmd user are automatically synchronized to all mysqld nodes in NDB 8.0, mcmd still tried to create the user on individual nodes, thus creating conflicts. With this patch, the mcmd user is now only created on one mysqld node, and the grants are synchronized to other nodes by the cluster. (Bug #30015430)

• Agent: When running an add process command, if multiple new nodegroups are to be created, mcmd misassigned nodes already in a nodegroup to a new group, causing an error to be thrown. It was due to the way mcmd reported nodegroup membership information internally, which this patch corrects. (Bug #29723329)

• Agent: On Windows platforms, the rotate log command gave a wrong name to the new log file. (Bug #29478833)

• Agent: For MySQL NDB Cluster 8.0, after a set skip_grant_tables:mysqld=true command has been issued from the mcm client, mcmd lost connection to the mysqld node. It was because on a mysqld node, remote connections were disabled when skip-grant-tables was true. With this fix, for MySQL servers started by mcmd, setting or resetting skip-grant-tables no longer disables remote connections, as long as the setting or resetting is performed when the mysqld node has been stopped. (Bug #29329690)

• Agent: For a NDB Cluster without any mysqld nodes, a rolling restart might sometime cause some of the mcmd agents to quit unexpectedly. (Bug #29289945)

• Agent: A slow increase in memory use was observed over time for mcmd, which might potentially add up to a significant amount if mcmd was not restarted for a long time. This fix eliminates the problem by removing a task that was executed repeatedly and unnecessarily by mcmd. (Bug #29166492)

• Agent: The import config command took a long time to run on SPARC systems. With this fix the process for retrieving the configuration from the MySQL Server has been optimized for SPARC systems, so that the command does not run longer than on other platforms. (Bug #22673589)

• Client: When the delete package command specified with the --hosts option a host name that mcmd did not recognize as a site member, mcmd quit unexpectedly. With this fix, mcmd throws a proper error in the situation. (Bug #29584728)