MySQL Cluster Manager 1.4 Release Notes  /  Changes in MySQL Cluster Manager 1.4.1 (2016-11-24)

Changes in MySQL Cluster Manager 1.4.1 (2016-11-24)

This section documents all changes and bug fixes that have been applied in MySQL Cluster Manager 1.4.1 since the release of MySQL Cluster Manager version 1.4.0.

Functionality Added or Changed

  • Microsoft Windows: A 64-bit Windows installer package for MySQL Cluster Manager is now available.

  • Agent: Quotes used in specifying an mcmd option were interpreted literally as parts of the option value. With this fix, mcmd now handles the quotes properly for the following options by stripping the quotes before interpreting the option values:

    (Bug #24529142)

  • Agent: When an mcmd agent failed to shutdown a mysqld node, it kept on retrying until the process timed out, at which point the error code from mysqld was returned to the client. With this fix, the error is returned immediately after the first shutdown attempt has failed. (Bug #24418005)

  • Agent: After a start process --added command was executed, if a show status --operational command was then run, in the Command column of the output, the ID for only one of the added nodes was shown even if multiple nodes had been added and started. With this fix, nodeids are no longer shown in the command output in that case, to avoid any possible confusions. (Bug #24372861)

  • Agent: The restore cluster command now supports a new option, --promote-attributes, which allows attributes to be promoted when MySQL Cluster Manager restores data from a backup. (Bug #24295481)

  • Agent: The collect logs command now also collects trace files, to provide more information in the event of a cluster failure. (Bug #23746394)

  • Agent: Information that is unrelated to debugging is no longer included into the agent log when --log-level is set to debug, but is included when --log-level is set to info. (Bug #23716969)

  • Agent: The capability for MySQL Cluster Manager to set the cluster parameter ndbinfo_table_prefix has been removed, as setting it might cause a timeout when mcmd tries to start a mysqld node. (Bug #23632067)

  • Agent: The import cluster command now supports the import of cluster nodes that have been started with options specified on the command line using underscores (_) instead of dashes (-) in the option names (for example, --ndb-connectstring instead of --ndb-connectstring). (Bug #23535372)

  • Agent: The --initial option can now be used with the start cluster command for starting clusters in the created status. (Bug #23138442)

  • Agent: The timer in mcmd is now monotonic, so interval measures are no longer affected by clock changes on the host. (Bug #22699245)

  • Agent: The version number for the mcmd agent and the name of the host it is running on are now always recorded in the agent log (mcmd.log) at startups or log rotations, irrespective of the --log-level setting. (Bug #22616530)

  • Agent: The backup cluster command now creates a logical backup for the metadata of the NDB tables in a MySQL Cluster, allowing more flexibility for cluster reconfiguration during a restore. See the description for backup cluster for more details. (Bug #21200829)

  • Agent: The --backupid option can now be used with the list backups command to specify the ID of the backup to be listed.

  • Client: Execution of the .mcm file created by the import config --dryrun command failed when a relative file path was given for a file-path attribute to be imported for a mysqld node. The mcm client now accepts relative file paths for those attributes, except for the socket value, and for any paths for a directory value (for example, plugin_dir). (Bug #18650848)

  • Client: Creating a single-host site with localhost as the argument for --hosts now results in a site that cannot be scaled up by the add hosts command; a warning for that is given by the mcm client at the creation of the site. See the description for the create site command for details. (Bug #18389510)

  • MySQL Cluster Manager now supports MySQL Cluster 7.5. At the same time, support for MySQL Cluster 7.2 has been deprecated. (Bug #24940329)

Bugs Fixed

  • Agent: When the backup agents command failed to copy some files, it finished without erroring out, but caused the subsequent mcmd agent restart to fail. This fix adds proper error-handling function to the backup agents command. (Bug #25057056)

  • Agent: When a start cluster or stop cluster command was run, if it took more than two minutes for any data node to be started or stopped, the mcmd agent threw a timeout for the command, and sometimes did that more than once, even though the data node was still in a normal process of being started or stopped. It was because the timeout extension needed was never granted within two minutes. With this fix, a timeout extension is granted at a much earlier time, and the amount of the extension is calculated based on the cluster's DataMemory size. (Bug #24749459)

  • Agent: When mcmd failed to create the users it needs on a mysqld node, it just kept retrying until a timeout was reached, and without notifying the mcm client. With this fix, a failure is declared immediately to the client in the situation, and there are no more retries. (Bug #24511041)

  • Agent: After a failure by mcmd to configure a cluster, the progress message from the process caused the next attempt to configure the cluster to fail. With this fix, the progress message from the last attempt only causes a warning to be issued to the agent log, and is then ignored. (Bug #24499097)

  • Agent: Attempts to upgrade MySQL Cluster Manager to version 1.4.0 by starting the 1.4.0 executables on pre-1.4 data failed, as the automatic upgrades for backup configurations were not successful. (Bug #24433400)

  • Agent: mcmd exited unexpectedly after failing to parse correctly a message of an unexpected format from a mysqld node. With this fix, sanity check is performed on the received message first, in order to avoid parsing the wrong information out of it. (Bug #24430610)

  • Agent: When attempting to set a global system variable on a mysqld node, the mcmd agent always returned a success message, even if it failed to contact the mysqld node actually. (Bug #24417856)

  • Agent: When running the list backups command, if one of the mcmd agents ran into permission errors on some of the backups' file paths, the returned results for the command became incomplete. With this fix, complete results are returned in those situations. (Bug #24414682)

  • Agent: When the mcm client and the mcmd agent it connected to were on the same host whose loopback address was not 127.0.0.1, if the host's name or loopback address was not explicitly contained in the host list in the create site command, attempts to restart the cluster later would fail with the complaint that the host name for the agent could not be resolved in the host list. With this fix, a warning is given in the situation when the create site command is issued, asking the user to include the proper host name. (Bug #24411504)

  • Agent: mcmd sometimes failed to report errors for its executions of SQL statements to the mcmd log. This fix tries to ensures that those errors are captured. (Bug #24375344)

  • Agent: When an add package command failed due to an error in loading the cluster configuration, mcmd exited unexpectedly on Windows platforms, and falsely reported that the package was added successfully on Unix-like platforms. (Bug #24361901)

  • Agent: When mcmd failed to write to the configuration file config.ini or my.cnf, mcmd either quit unexpectedly (for most Unix-like platforms), or returned a success message without actually writing to the file (on Solaris). (Bug #23632067)

  • Agent: When a configuration parameter was imported from a mysqld node using the import config command, the value got truncated when it was a quoted value containing a hash sign and was followed by a comment when listed inside the configuration file. (Bug #23591849)

  • Agent: On Windows platforms, the value for the system variable wait_timeout for mysqld nodes cannot be made larger than 2147483 using the set command. (Bug #23503256)

  • Agent: On Windows platforms, the import cluster command failed when a 32-bit mcmd was working a 64-bit MySQL Cluster, or vice versa. (Bug #23503256)

  • Agent: The import config --dryrun command produced set commands of wrong syntax for instance-level TCP configuration settings specified inside the config.ini file. (Bug #23341146)

  • Agent: When NoOfReplicas was 4 for a cluster and only one forth of all the data nodes were available, mcmd reported that the cluster was non-operational when the show status --cluster command was run. (Bug #23330032)

  • Agent: mcmd ignored status update from an ndbapi (or unmanaged mysqld) node when it reported an LSN lower than one already reported by some other agent. This was due to an unnecessary check on the reported LSNs, which has been removed by this fix. (Bug #23320387)

  • Agent: When working with MySQL Cluster 7.5, resetting the Datadir parameter for a mysqld node using a reset command caused an error when the default data directory had contents inside. (Bug #23283577)

  • Agent: The command start process --added failed with the error Unable to create nodegroup ... when the parameter NoOfReplicas was set to 3 or 4. (Bug #23257723)

  • Agent: Some ndbmtd processes were being excluded from rolling restarts when the parameter NoOfReplicas was set to 3 or 4. (Bug #23251630)

  • Agent: With the parameter NoOfReplicas set to 3 or 4, even when there was still one mirror node left in a nodegroup for a ndbmtd process, the stop process command issued for the ndbmtd node was rejected by mcmd. (Bug #23250053)

  • Agent: When a quorum of majority of mcmd agents no longer existed (due to, for example, a network failure), an mcmd agent reported wrong statuses of failed processes even when those processes were local to its own host and were accessible by the agent. (Bug #23222658)

  • Agent: When network connection to a host with a data node running was lost and then reestablished, the show status command reported falsely that the data node was running again even though it had actually stopped. This fix makes sure the current status of the data node is properly reflected. (Bug #23220981)

  • Agent: When there was a failure in setting the value of the parameter LogDestination, mcmd, when returning an error, also issued sometimes a warning that an earlier runtime error had been overwritten. This is now prevented by putting in proper checking and handling for any existing error as a new error is being thrown. (Bug #23211849)

  • Agent: If a warning was returned by mcmd after it executed a SET GLOBAL statement to a mysqld node, the same warning was issued to the agent log (mcmd.log) again and again each time a new SET GLOBAL statement was executed. (Bug #23211783)

  • Agent: When the parameter ArbitrationRank was being set for a mysqld node, the data nodes and management nodes were not restarted, so they had no knowledge that the arbitration rank for the mysqld node has been changed. (Bug #23148368)

  • Agent: After a failed execution of the import config command using an older version of MySQL Cluster Manager, if MySQL Cluster Manager was then upgraded, restarting would fail afterward for some of the mcmd agents. (Bug #23148061)

  • Agent: In the case where there was only one ndb_mgmd node in a cluster, during a rolling restart, after the ndb_mgmd node was just restarted and a first data node was stopped, a second data node might fail before it got restarted, complaining that there was no arbitrator for the cluster. It was because the management node was still in the process of asserting itself as an arbitrator. With this fix, the restarting of data nodes only begin after they have all seen an arbitrator established. (Bug #23148061)

  • Agent: A new ndbd or ndbmtd node could not be added to a cluster if BackupDataDir specified a non-default location for cluster backups, and there were backups existing at that location. (Bug #23123364)

  • Agent: When already existing nodes were added again to a cluster without being started and the cluster was stopped, a subsequent start cluster --initial command failed. This was because mcmd attempted to recreate the already existing node groups, which this fix prevents. (Bug #23024367)

  • Agent: While the my.cnf configuration files for added mysqld processes generated by mcmd used group suffixes for group titles in the files (for example [mysqld.50]), the mysqld nodes were started without using the --defaults-group-suffix option, causing the generated configurations to be unread by the mysqld node. With this fix, the group suffixes are no longer used. (Bug #22931198)

  • Agent: On Windows platforms, an mcmd agent quit unexpectedly after it was unable to open a process handler to a PID. This fix makes mcmd able to handle the situation and not quit. (Bug #22886512)

  • Agent: During the time when an upgrade cluster command is being performed, if a mysqld node that had not yet been upgraded failed somehow, it was mistakenly restarted with the newer instead of the older binary package. With this fix, the node is restarted with the original binary package it was running on. (Bug #22880634)

  • Agent: Under some conditions, setting a configuration attribute for a cluster using the set command made mcmd quit unexpectedly, as more than one thread in MySQL Cluster Manager tried to change the cluster configurations together within a very short period of time, with one thread running into a checksum error for the repository directory and causing the agent to quit. This fix adds error checks, retries, and also exponential backoffs to handle the situation, in order to allow proper execution of the set command. (Bug #22865068)

  • Agent: A restore cluster command failed because one of the management nodes in the cluster was in the starting status, which should not prevent a cluster from being restored. (Bug #22755257)

  • Agent: When the start process --added command was used to start data nodes that were newly added, the command failed if there was an ndbapi node that was in the connected status, with the complaint that the ndbapi node was already running. (Bug #22726592)

  • Agent: If an earlier attempt to stop a mysqld node failed and left the node in a status of stopping, a subsequent stop cluster command failed with a timeout. With this fix, the stop cluster command actually reattempts twice the shutdown of the stopping process, and throws a proper error if the attempts fail. The timeout period for each attempt is also reduced to 5 seconds. (Bug #22682222, Bug #24735542)

    References: See also: Bug #19805950.

  • Agent: When performing a create cluster command for a large cluster with many processes, mcmd tried to resolve the host name and check the validity of the package name for every single process, which could take a long time and cause the create cluster command to fail with a timeout. mcmd now avoids repeated lookups for the same host name or package name, so the timeouts are prevented. (Bug #22671177)

  • Agent: When a backup was actually completed after an abort backup command had just been issued through the mcmd agent, the agent failed to respond to the backup completion, resulting in a timeout for the abort backup command. With this fix, the abort backup command will error out under the situation. (Bug #22655696)

  • Agent: When setting a value for the log-error attribute for a mysqld node using the set command, mcmd failed to, as expected, append the extension .err to the supplied file name when it did not have an extension. (Bug #22588267)

  • Agent: The mysql client hung as a status command was issued through it to a mysqld node, due to some communication packages remaining unhandled by mcmd. (Bug #22539167)

  • Agent: The autotune --dryrun command did not write to the .mcm script file the TCP connection attributes (for example, SendBufferMemory and ReceiveBufferMemory) it would set for the cluster when the --dryrun option was not used. (Bug #22517603)

  • Agent: With two clusters running separately on two different hosts in the same site, the autotune --dryrun command for a cluster failed with an internal error, complaining that a dump file [for the other cluster running on the other host] could not be opened because it did not exist on the host—which was to be expected. (Bug #22465053, Bug #79586)

  • Agent: If a cluster log rotation happened while MySQL Cluster Manager was starting the cluster, an mcmd agent failure would occur. This fix makes sure log rotation is properly detected and handled by mcmd. (Bug #22296243)

  • Agent: If distributed privileges for mysqld nodes were used and the root password has been set on the cluster's mysqld nodes, after a new mysqld process was added, MySQL Cluster Manager failed to connect to it, as mcmd tried in vain to log on to the new mysqld node as root with an empty password. With this fix, mcmd then attempts in that situation to log on to the node as the user mcmd. Also, even if the root account is not secured, when distributed privileges are used, mcmd now attempts first to connect to a new mysqld node as mcmcd before it tries to create the mcmd user on the new node. (Bug #22274785, Bug #81391)

  • Agent: For MySQL Cluster 7.4.8 and later, the import config command imported the deprecated PortNumber attribute under the [tcp] section of the cluster configuration file as 0.. The attribute is now skipped during a configuration import. (Bug #22274785)

  • Agent: A create cluster command failed with an error when one of the mcmd agents was down, even if that agent was not needed for the process. (Bug #22245706)

  • Agent: When the utility mysql_install_db was run by the mcmd agent at the creation of a mysqld node, it was not run asynchronously, resulting sometimes in unnecessary delay for other mcmd processes. With this fix, the utility is now run asynchronously. (Bug #22238508)

  • Agent: After a new host has been added to a cluster without a package being added for it, mcmd returned an error message to any get and set command, saying that the parameter to be get or set did not exist. With this fix, a proper error message is returned. (Bug #21894353)

  • Agent: On Windows platform, setting the datadir option for a mysqld node in the Windows file path format caused the mcmd agent to stop unexpectedly after it failed to restart the mysqld node. It was due to a mishandling of the Windows format file path, which has now been fixed. (Bug #19209870)

  • Agent: A rolling restart of the cluster performed by mcmd timed out while waiting for GCP and LCP takeover events to complete among the data nodes. With this fix, the timeout is avoided by having mcmd check for the status of the takeover events and wait until the involved data nodes are ready before trying to stop them. (Bug #14230789)

  • Client: Spaces in quoted option values for MySQL Cluster Manager were lost (for example, with --prompt='mcm1.4.1> ', the prompt for the mcm client became mcm1.4.1> [with no space at the end]). (Bug #24528495)

  • Client: When the mcm client was being started with the --debug option, if the mysql client could not be found at the expected location, the mcm client failed with a segmentation fault. (Bug #24522244)

  • Client: The success messages returned by the autotune --dryrun and import config --dryrun commands referred users to the agent log file for the proposed settings to be applied to the cluster, but the settings were not actually in the file. The success messages now give the path to the .mcm script file that contains the settings. (Bug #22280689)

  • Client: After the mcmd agent on a certain host failed, the mcm client continued to report the statuses of the processes on the host to running while the mcmd agent log and the ndb_mgmd queries already showed their statuses to be unknown.. (Bug #22174415)