Changes in MySQL NDB Cluster 8.4.0 (2024-04-30, LTS Release)

• Deprecation and Removal Notes

• Functionality Added or Changed

• Bugs Fixed

Deprecation and Removal Notes

• Packaging; Linux: Removed the deprecated tool /usr/bin/pathfix.py from packages for Fedora 39. (Bug #35997178)

Functionality Added or Changed

• ndbinfo Information Database: Added the transporter_details table to the ndbinfo information database. This table is similar to the transporters table, but provides information about individual transporters rather than in the aggregate.

  For more information, see The ndbinfo transporter_details Table. (Bug #113163, Bug #36031560)

• NDB Client Programs: Added the --verbose option to the ndb_waiter test program to control the verbosity level of the output. (Bug #34547034)

• Improved logging related to purging of the binary log, including start and completions times, and whether it is the injector which has initiated the purge. (Bug #36176983)

Bugs Fixed

• NDB Cluster APIs: TLS connection errors were printed even though TLS was not specified for connections.

  To fix this issue, following an ignored TLS error, we explicitly reset the error condition in the management handle to NO_ERROR. (Bug #36354973)

• NDB Cluster APIs: The NdbEventOperation methods hasError() and clearError(), long deprecated, are effectively disabled: hasError() now returns a constant 0, and clearError() does nothing. To determine an event type, use getEventType2() instead.

• NDB Client Programs: In some cases, it was not possible to load cerificates generated using ndb_sign_keys. (Bug #36430004)

• NDB Client Programs: The following command-line options did not function correctly for the ndb_redo_log_reader utility program:

  • --mbyte

  • --page

  • --pageindex

  (Bug #36313427)

• NDB Client Programs: A certificate lifetime generated by ndb_sign_keys should consist of a fixed number of days, plus a random amount of extra time provided by the OpenSSL function RAND_bytes(), casting the result to a signed integer value. Because this value could sometimes be negative, this led to extra time being subtracted rather than added.

  We eliminate this problem by using an unsigned integer type to hold the value obtained from RAND_bytes(). (Bug #36270629)

• NDB Client Programs: Invoking ndb_mgmd with the --bind-address option could in some cases cause the program to terminate unexpectedly. (Bug #36263410)

• NDB Client Programs: Some NDB utilities such ndb_show_tables leaked memory from API connections when TLS was required by the data nodes, and with valid certificates. (Bug #36170703)

• NDB Client Programs: Work begun in NDB 8.0.18 and 8.0.20 to remove the unnecessary text NDBT_ProgramExit ... from the output of NDB programs is completed in this release. This message should no longer appear in the release binaries of any such programs. (Bug #36169823)

  References: See also: Bug #27096741.

• NDB Client Programs: The output from ndb_waiter --ndb-tls-search-path was not correctly formatted. (Bug #36132430)

• NDB Client Programs: On Windows hosts, ndb_sign_keys could not locate the ssh program. (Bug #36053948)

• NDB Client Programs: ndb_sign_keys did not handle the --CA-tool option correctly on Windows. (Bug #36053908)

• NDB Client Programs: The use of a strict 80-character limit for clang-format on the file CommandInterpreter.cpp broke the formatting of the interactive help text in the NDB management client. (Bug #36034395)

• NDB Client Programs: Trying to start ndb_mgmd with --bind-address=localhost failed with the error Illegal bind address, which was returned from the MGM API when attempting to parse the bind adress to split it into host and port parts. localhost is now accepted as a valid address in such cases. (Bug #36005903)

• An implicit rollback generated when refusing to discover a table in an ongoing transaction caused the entire transaction to roll back. This could happen when a table definition changed while a transaction was active. We also checked at such times to see whether the table already existed in the data dictionary, which also meant that a subsequent read from same table within the same transaction would (wrongly) allow discovery.

  Now in such cases, we skip checking whether or not a given table already exists in the data dictionary; instead, we now always refuse discovery of a table that is altered while a transaction is ongoing and return an error to the user. (Bug #36191370)

• When a backup was restored using ndb_restore with --disable-indexes and --restore-privilege-tables, the ordered index of the primary key was lost on the mysql.ndb_sql_metadata table, and could not be rebuilt even with --rebuild-indexes. (Bug #36157626)

• NDB maintains both a local and a global pool of free send buffers. When send buffers cannot be allocated from the local pool NDB allocates one from the global pool; likewise, buffers are freed and returned to the global pool when the local pool has too many free buffers. Both of these allocations require a mutex to be locked.

  In order to reduce contention on this global mutex, we attempt to over-allocate buffers from the global pool when needed, keeping the excess buffers in the local pool, when releasing excess buffers to the global pool this was done only to the limit determined by max_free. After having released to the global pool, such that the max_free limit was met, it was likely that additional buffers would soon be released, once again exceeding max_free. This caused extra contention on the global pool mutex.

  To address this issue, we now reduce the free buffers to 2/3 of the max_free limit in such cases. (Bug #36108639)

• SSL_pending() data from an SSL-enabled NdbSocket was not adequately checked for. (Bug #36076879)

• In certain cases, ndb_mgmd hung when attempting to sending a stop signal to ndbmtd. (Bug #36066725)

• Starting a replica to apply changes when NDB was not yet ready or had no yet started led to an unhelpful error message (Fatal error: Failed to run 'applier_start' hook). This happened when the replica started and the applier start hook waited for the number of seconds specified by --ndb-wait-setup for NDB to become ready; if it was not ready by then, the start hook reported the failure. Now in such cases, we let processing continue, instead, and allow the error to be returned from NDB, which better indicates its true source. (Bug #36054134)

• A mysqld process took much longer than expected to shut down when all data nodes were unreachable. (Bug #36052113)

• Negated the need for handling in the NDB binary log injector thread for a failure to instantiate an injector transaction by removing a potential point of failure in that operation. (Bug #36048889)

• It was possible in certain cases for the TRPMAN block to operate on transporters outside its own receive thread. (Bug #36028782)

• Removed a possible race condition between start_clients_thread() and update_connections(), due to both of these seeing the same transporter in the DISCONNECTING state. Now we make sure that disconnection is in fact completed before we set indicating that that the transporter has disconnected, so that update_connections() cannot close the NdbSocket before it has been completely shut down. (Bug #36009860)

• When a transporter was overloaded, the send thread did not yield to the CPU as expected, instead retrying the transporter repeatedly until reaching the hard-coded 200 microsecond timeout. (Bug #36004838)

• A MySQL server disconnected from schema distribution was unable to set up event operations because the table columns could not be found in the event. This could be made to happen by using ndb_drop_table or another means to drop a table directly from NDB that had been created using the MySQL server.

  We fix this by making sure in such cases that we properly invalidate the NDB table definition from the dictionary cache. (Bug #35948153)

• The ndb_sign_keys utility's --remote-openssl option did not function as expected. (Bug #35853405)

• A replica could not apply a row change while handling a Table definition changed error. Now any such error is handled as a temporary error which can be retried multiple times. (Bug #35826145)

• Repeated incomplete incomplete attempts to perform a system restart in some cases left the cluster in a state from which it could not recover without restoring it from backup. (Bug #35801548)

• The event buffer used by the NDB API maintains an internal pool of free memory to reduce the interactions with the runtime and operating system, while allowing memory that is no longer needed to be returned for other uses. This free memory is subtracted from the total allocated memory to determine the memory is use which is reported and used for enforcing buffer limits and other purposes; this was represented using a 32-bit value, so that if it exceeded 4 GB, the value wrapped, and the amount of free memory appeared to be reduced. This had potentially adverse effects on event buffer memory release to the runtime and OS, free memory reporting, and memory limit handling.

  This is fixed by using a 64-bit value to represent the amount of pooled free memory. (Bug #35483764)

  References: See also: Bug #35655162, Bug #35663761.

• START REPLICA, STOP REPLICA, and RESET REPLICA statements are now written to mysqld.log. (Bug #35207235)

• NDB transporter handling in mt.cpp differentiated between neighbor transporters carrying signals between nodes in the same node group, and all other transporters. This sometimes led to issues with multiple transporters when a transporter connected nodes that were neighbors with nodes that were not. (Bug #33800633)

• Removed unnecessary warnings generated by transient disconnections of data nodes during restore operations. (Bug #33144487)

• During setup of utility tables, the schema event handler sometimes hung waiting for the global schema lock (GSL) to become available. This could happen when the physical tables had been dropped from the cluster, or when the connection was lost for some other reason. Now we use a try lock when attempting to acquire the GSL in such cases, thus causing another setup check attempt to be made at a later time if the global schema lock is not available. (Bug #32550019, Bug #35949017)

• API nodes did not record any information in the log relating to disconnects due to missed heartbeats from the data nodes. (Bug #29623286)