Changes in MySQL 8.0.12 (2018-07-27, General Availability)

For general information about upgrades, downgrades, platform support, etc., please visit https://dev.mysql.com/doc/relnotes/mysql/8.0/en/.

• Audit Log Notes

• Compilation Notes

• Configuration Notes

• Data Dictionary Notes

• Deprecation and Removal Notes

• Firewall Notes

• SQL Function and Operator Notes

• Logging Notes

• Optimizer Notes

• Packaging Notes

• Plugin Notes

• Security Notes

• Spatial Data Support

• SQL Syntax Notes

• X Plugin Notes

• Functionality Added or Changed

• Bugs Fixed

Audit Log Notes

• For new MySQL installations, the USER and HOST columns in the audit_log_user table used by MySQL Enterprise Audit now have definitions that better correspond to the definitions of the User and Host columns in the mysql.user system table.

  For upgrades to an installation for which MySQL Enterprise Audit is already installed, it is recommended that you alter the table definitions as follows:

  ALTER TABLE mysql.audit_log_user
    DROP FOREIGN KEY audit_log_user_ibfk_1;
  ALTER TABLE mysql.audit_log_filter
    CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_as_ci;
  ALTER TABLE mysql.audit_log_user
    CONVERT TO CHARACTER SET utf8mb4 COLLATE utf8mb4_0900_as_ci;
  ALTER TABLE mysql.audit_log_user
    MODIFY COLUMN USER VARCHAR(32);
  ALTER TABLE mysql.audit_log_user
    ADD FOREIGN KEY (FILTERNAME) REFERENCES mysql.audit_log_filter(NAME);

  (Bug #23706056)

Compilation Notes

• Microsoft Windows: For OpenSSL 1.1, the library names have changed for Windows. The CMake configuration code now accounts for this. Also, it is now possible on Windows to use -DWITHOUT_SERVER=1 to build 32-bit client binaries. (Bug #28170711, Bug #91223)

• The minimum version of the Boost library for server builds is now 1.67.0. (Bug #27866110)

Configuration Notes

• The maximum value was increased from 256 to 1024 for these Performance Schema system variables:

  performance_schema_max_cond_classes
  performance_schema_max_file_classes
  performance_schema_max_mutex_classes
  performance_schema_max_rwlock_classes
  performance_schema_max_socket_classes
  performance_schema_max_stage_classes
  performance_schema_max_thread_classes

  The default value was increased from 250 to 300 for this Performance Schema system variable:

  performance_schema_max_mutex_classes

  (Bug #27647918)

• For RelWithDebInfo builds on Linux, REPRODUCIBLE_BUILD now defaults to ON. (Bug #27483447)

• The default value for the slave_pending_jobs_size_max system variable has increased from 16M to 128M. On a multi-threaded slave ( with slave_parallel_workers > 0), this value sets the maximum amount of memory that is made available to slave worker queues holding events that have not yet been applied. The potentially larger memory usage should be taken into account when calculating the resources required to deploy a multi-threaded slave. Masters and single-threaded slaves are not affected by this setting.

  The new default is double the recently increased default value specified by the max_allowed_packet system variable (now 64M). The value of slave_pending_jobs_size_max is a soft limit, so events larger than its value (consisting of one or more packets) can be queued and processed. However, a large transaction is held until all the slave workers have empty queues, and then processed. All subsequent transactions are held until the large transaction has been completed. The delay to clear the queues of all the slave workers and the wait to queue subsequent transactions can cause lag on the replication slave and decreased concurrency of the slave workers. The value of slave_pending_jobs_size_max should therefore be set high enough to accommodate most expected event sizes. (WL #11348)

Data Dictionary Notes

• The following data dictionary enhancements were introduced:

  • An internal method was added to protect against global read locks. The method is called before storing, updating, or dropping data dictionary objects.

  • Assertion code was added to the data dictionary API to protect against global read locks when storing, updating, or dropping data dictionary objects.

  • Protection is now implicitly acquired against backup locks and global read locks when an exclusive metadata lock is acquired by a storage engine.

  • A cost model cache validity check is now performed before releasing a cost model entry.

  • A data dictionary API function was added for committing and rolling back transactions, and for releasing transactional meta data locks.

  (Bug #27937059)

Deprecation and Removal Notes

• Explicit ASC or DESC qualifiers for GROUP BY clauses are now deprecated and will be removed in a future MySQL version. (WL #11459)

Firewall Notes

• The MySQL Enterprise Firewall firewall_whitelist table now contains a primary key column named ID. (Bug #27164826)

SQL Function and Operator Notes

• The BIT_AND(), BIT_OR(), and BIT_XOR() aggregate functions now can be used as window functions, if an OVER clause is present. See Aggregate Function Descriptions. (WL #11573)

Logging Notes

• Previously, successful assignments to the dragnet.log_error_filter_rules system variable resulted in a Note and a nonzero warning count. To avoid generating a diagnostic for a successful operation, this Note is no longer generated. Instead, the dragnet.Status status variable can be consulted to determine the result of the most recent assignment to dragnet.log_error_filter_rules. (Bug #27910708, Bug #90571)

• Previously, log components listed in the log_error_services system variable had to be separated by semicolons. Components now can be separated by commas. A given setting cannot use both semicolon and comma separators. (Bug #27788925, Bug #90268)

Optimizer Notes

• The filesort algorithm (used by the optimizer to satisfy an ORDER BY clause when no index can be used) is now more memory efficient. The primary change is that the optimizer allocates memory buffers incrementally as needed, up to the size indicated by the sort_buffer_size system variable, rather than allocating a fixed amount of sort_buffer_size bytes up front. This enables users to set sort_buffer_size to larger values to speed up larger sorts, without concern for excessive memory use for small sorts. (This benefit may not occur for multiple concurrent sorts on Windows, which has a weak multithreaded malloc.)

  In addition, filesort is less pessimistic with respect to worst-case scenarios. Previously, filesort produced an error without starting if merge operations or sort-key generation might fail. Now, because the worst case is unlikely to occur in practice, filesort produces an error only for actual rather than prospective failures. (WL #11590)

• Previously, extended EXPLAIN output was produced only for SELECT statements. Extended output is now produced for INSERT, REPLACE, UPDATE, and DELETE statements as well. (To display extended output, use SHOW WARNINGS following EXPLAIN. See Extended EXPLAIN Output Format.) (WL #6033)

Packaging Notes

• For Linux, the generic binary distribution compression algorithm changed from Gzip to XZ; and the file extension changed from .tar.gz to .tar.xz. (Bug #28450941, Bug #91889)

• Docker disables NUMA-related operations by default, so the mysqld included in Docker images now has NUMA support disabled so that the error log does not fill with NUMA warnings. (Bug #28081363)

• RPM .spec files were updated to reflect the removal of yaSSL and that OpenSSL is the default SSL library for all builds. (Bug #28025427)

• For generic Linux .tar.gz distributions, the mysqlxtest client has been moved from the server package to the test package. (Bug #27744137)

• An RPM package for installing ARM 64-bit (aarch64) binaries of MySQL Server on Oracle Linux 7 is now available in the MySQL Yum Repository and for direct download.

  Known Limitation for this ARM release: You must enable the Oracle Linux 7 Software Collections Repository (ol7_software_collections) to install this package, and must also adjust the libstdc++7 path. See Yum's Platform Specific Notes for additional details.

Plugin Notes

• Previously, the Rewriter query rewrite plugin permitted rewriting only for SELECT statements. Rewritable statements now include INSERT, REPLACE, UPDATE, and DELETE as well. See The Rewriter Query Rewrite Plugin. (WL #8776)

Security Notes

• The linked OpenSSL library for MySQL Server has been updated to version 1.0.2o. Issues fixed in the new OpenSSL version are described at http://www.openssl.org/news/vulnerabilities.html. (Bug #28025379)

Spatial Data Support

• Incompatible Change: The new spatial functions ST_Longitude() and ST_Latitude() return the longitude or latitude, respectively, of their Point argument. If called with a second argument, these functions use the argument to set the longitude or latitude, respectively, of their Point argument.

  In addition, the ST_X() and ST_Y() functions now interpret their Point argument according to its spatial reference system definition (SRS):

  • ST_X() returns the coordinate value of the axis that appears first in the SRS definition, and ST_Y() returns the coordinate value of the axis that appears second in the SRS definition.

  • If ST_X() and ST_Y() are used to set the longitude or latitude of a Point argument that has a geographic SRS, the longitude or latitude value must be in the proper range for the SRS, or an error occurs.

  Note

  If spatial data contains geometry values that now are interpreted differently by ST_X() and ST_Y(), existing queries using these functions will return different results, compared to previous MySQL versions.

  (Bug #27125600, Bug #88503, WL #11453)

• The EPSG data set containing spatial reference system data for spatial calculations has been upgraded from version 9.2 to 9.3. (Bug #27977721)

• Spatial functions that accept geometry arguments and previously returned an ER_LONGITUDE_OUT_OF_RANGE or ER_LATITUDE_OUT_OF_RANGE error for longitude or latitude values out of range now return ER_GEOMETRY_PARAM_LONGITUDE_OUT_OF_RANGE or ER_GEOMETRY_PARAM_LATITUDE_OUT_OF_RANGE instead. This does not apply to spatial functions that parse or modify geometries: ST_xxxFromText(), ST_xxxFromWKT(), ST_xxxFromWKB(), ST_GeomFromGeoJSON(), ST_X(), ST_Y(), ST_Longitude(), and ST_Latitude(). (Bug #27892138)

• Previously, the parser recognized USING RTREE in index specifications, but this could not be specified for any storage engine. Now USING RTREE is permitted for spatial index specifications. (Bug #27836608)

• Indexes on geometry columns now must be spatial indexes. Consequently, the SPATIAL keyword is optional but implicit for creating an index on a spatial column. A spatial index on a geometry column is not permitted for a primary key or unique index. Also, ASC and DESC are not permitted for spatial indexes. (Bug #21087676, WL #11808)

SQL Syntax Notes

• Important Change: MySQL now allows a query having a WITH ROLLUP modifier to use DISTINCT. See SELECT Statement, for more information. (Bug #87450, Bug #26640100, Bug #20671578, Bug #76229)

• Important Change: MySQL now allows the use of ORDER BY with grouping functions in SELECT. statements. (Bug #86312, Bug #27063525)

• Important Change: MySQL now allows ORDER BY to be used in a query having a WITH ROLLUP modifier. For more information, see SELECT Statement. (Bug #86311, Bug #26073513)

X Plugin Notes

• Authentication errors returned from a MySQL 8.0.11 server to an X Plugin client were marked as fatal, which caused the client to stop attempting its automatic sequence of authentication attempts using different methods. X Plugin now ignores fatal authentication errors, and subsequent expected errors such as write timeout errors, while the sequence of authentication attempts is in progress. If none of the authentication methods tried are successful, X Plugin now selects and returns the most relevant error message received from the server, which is not necessarily the last error message received. (Bug #28135006)

• During the X Plugin automated authentication sequence, connection attempts are made using different authentication methods in a pre-set sequence. X Plugin checks before starting the sequence whether the connection uses SSL or not, and does not attempt connection with the PLAIN authentication method if the connection is not encrypted. For additional verification, a check and error were added so that the PLAIN authentication method is disallowed if attempted when the connection is not encrypted. (Bug #27691189)

• In the event of an authentication failure, X Plugin always returned the error code ER_ACCESS_DENIED, with a custom X Protocol error message for the error situation. The error codes that are received from the MySQL server for authentication failures are now passed on to the client instead, with the standard MySQL error messages that go with them. (Bug #27675699)

• During the X Plugin automated authentication sequence, where connection attempts are made using different authentication methods in a pre-set sequence, the error returned after every failed connection attempt was marked with the severity FATAL. Now, the errors are marked as ERROR while there are still other authentication methods to be attempted, and only the final error that results in disconnection is marked as FATAL. (Bug #27636947)

• The X Plugin code was being compiled twice with different preprocessor definitions. (Bug #27267054)

• A number of improvements were made to the handling of authentication issues by X Plugin:

  • The X Plugin automated authentication sequence, where connection attempts are made using different authentication methods in a pre-set sequence, is now halted if an error response indicates that the connection has been broken or reads or writes are timing out.

  • The automated authentication sequence is also now halted in the event of any fatal error.

  • The last significant error is now reported for an abandoned authentication sequence, not just the last error that occurred.

  • A more specific error code and message are now reported if authentication with the caching_sha2_password authentication plugin is attempted over an unsecure connection.

  (Bug #27257774)

• In some circumstances, a MySQL server with X Plugin loaded could experience an unexpected halt during a shutdown process initiated by a failed start or by the RESTART command. The issue has now been fixed. (Bug #25474793, Bug #27259783)

Functionality Added or Changed

• Important Change; Group Replication: group_replication_exit_state_action determines how Group Replication behaves when a server instance leaves the group involuntarily, such as when it is expelled from the group due to an unstable network connection. When group_replication_exit_state_action is ABORT_SERVER (the default), the instance shuts itself down; when its value is READ_ONLY, the instance switches itself to super read-only mode instead, and goes into the Group Replication ERROR state. (WL #11568)

• InnoDB: InnoDB now supports ALGORITHM=INSTANT for the following ALTER TABLE operations:

  • Adding a column. This feature is referred to as “Instant ADD COLUMN”.

  • Adding or dropping a virtual column.

  • Adding or dropping a column default value.

  • Modifying the definition of an ENUM or SET column.

  • Changing the index type.

  • Renaming a table.

  Operations that support ALGORITHM=INSTANT only modify metadata in the data dictionary. No exclusive metadata locks are taken on the table during preparation and execution phases of the operation, and table data is unaffected, making the operations instantaneous. If not specified explicitly, ALGORITHM=INSTANT is used by default by operations that support it. If ALGORITHM=INSTANT is specified but not supported, the operation fails immediately with an error.

  For more information about operations that support ALGORITHM=INSTANT, see Online DDL Operations.

  Thanks to the Tencent Games DBA team for the contribution. (Bug #28100103, Bug #91074, WL #11250)

• InnoDB: Undo logging is now supported for small updates to large object (LOB) data, which improves performance of LOB updates that are 100 bytes in size or less. Previously, LOB updates were a minimum of one LOB page in size, which is less than optimal for updates that might only modify a few bytes. This enhancement builds upon support added in MySQL 8.0.4 for partial update of LOB data. (WL #11328)

• Replication: You can now use a shell pipe to supply an archive of compressed binary log files to mysqlbinlog as streamed input. Previously, the files had to be extracted separately before passing them to mysqlbinlog. Thanks to the Facebook team for this contribution.

  In this example, binlog-files_1.gz contains multiple binary log files for processing. The pipeline extracts the contents of binlog-files_1.gz, pipes the binary log files to mysqlbinlog as standard input, and pipes the output of mysqlbinlog into the mysql client for execution:

  gzip -cd binlog-files_1.gz | ./mysqlbinlog - | ./mysql -uroot -p

  (Bug #27445278, Bug #89423)

  References: See also: Bug #27836543.

• Replication: The SYSTEM_VARIABLES_ADMIN and PERSIST_RO_VARIABLES_ADMIN privileges have been added to the mysql.session reserved account. (Bug #89873, Bug #27621869)

• Replication: Information about XCom's memory usage has been added to the Performance Schema tables. The setup_instruments table now has an entry corresponding to the XCom cache, and the memory_summary_global_by_event_name table contains the memory usage statistics of the XCom cache. (WL #9855)

• Microsoft Windows: On Windows, the forking used to implement RESTART was suppressible using the --gdb option. However, this was a side effect, performed in addition to the other option actions done to set up a debugging environment. In non-debug settings, the new --no-monitor option may be used for the sole purpose of suppressing forking the monitor process. For a server started with either --gdb or --no-monitor, executing RESTART causes the server to simply exit without restarting. (Bug #27801043)

• Checking for foreign key relationships by DROP TABLE and DROP DATABASE was improved. The check now takes place before any tables are deleted, so that in the event of a problem the statement fails before making any changes. Parent and child tables now can be dropped in arbitrary order, as long as they are dropped by the same DROP TABLE statement. In addition, error reporting was improved for attempts to drop a parent table without dropping a child table. (Bug #27821060, Bug #17564464, Bug #70531, Bug #22359539, Bug #79610)

• When support for roles was added, the ADMIN keyword became a reserved keyword. ADMIN is now once again a nonreserved keyword. (Bug #27814204)

• When a client shuts down the server, the server now writes a message to the error log indicating which user performed this action. (Bug #26246628, Bug #86635)

• Previously, for the --ssl-mode=VERIFY_IDENTITY option, the client checked whether the host name that it used for connecting matched the Common Name value in the certificate but not the Subject Alternative Name value. Now, if the client uses OpenSSL 1.0.2 or higher, the client checks whether the host name matches either the Subject Alternative Name value or the Common Name value in the server certificate. Thanks to Daniël van Eeden for a patch on which this change was based. (Bug #16211011, Bug #68052, Bug #27511233, Bug #89578)

Bugs Fixed

• Important Change; JSON: The JSON_TABLE() function now automatically decodes base-64 values and prints them using the character set given by the column specification. (Bug #90157, Bug #27729112)

  References: See also: Bug #89847, Bug #27613276.

• InnoDB: New columns added to INFORMATION_SCHEMA.INNODB_TABLES and INFORMATION_SCHEMA.INNODB_COLUMNS tables in 8.0.12 did not appear after upgrading to 8.0.12. (Bug #28065244)

• InnoDB: Large object (LOB) version numbers stored in LOB index entries were left uninitialized at 0 instead of initialized at 1. (Bug #28046298)

• InnoDB: Adding a column to a table with a generated column raised an invalid assertion. (Bug #28040201)

• InnoDB: An attempt to access freed memory during a checkpoint operation caused the server to exit. (Bug #28039477)

• InnoDB: An assertion failure occurred when starting the server with experimental system variables that were unintentionally exposed in normal MySQL builds. (Bug #27899424, Bug #90526, Bug #27898396, Bug #90523, Bug #27898344, Bug #90522)

• InnoDB: An invalid system variable setting raised an assertion. (Bug #27898284)

• InnoDB: Attempting to move a table from a discarded file-per-table tablespace to a shared tablespace raised an assertion. The operation now produces an error instead. (Bug #27861972)

• InnoDB: Potential failures related to page allocation for large objects (LOBs) were not sufficiently handled. (Bug #27852003)

• InnoDB: The ngram full-text search parser permitted comma and period characters to be tokenized as words, which caused an inconsistency between boolean and natural language mode search results. Comma and period characters are no longer tokenized. (Bug #27847697)

• InnoDB: Checksum mismatches and other issues were caused by concurrent modification of a tablespace header page, and attempting to place a lock on a partially initialized page. (Bug #27823064)

• InnoDB: An I/O error returned by an fsync() operation is now treated as a hard error. (Bug #27805553, Bug #90296)

• InnoDB: In a single thread, two minitransactions latched the first page of a large object value (LOB) while another thread attempted to allocate the page, causing a deadlock. (Bug #27777959)

• InnoDB: After upgrading to MySQL 8.0.12, table columns including the FTS_DOC_ID column were missing from the INFORMATION_SCHEMA.INNODB_COLUMNS table. (Bug #27774145)

• InnoDB: An invalid assertion was raised when the optimizer chose a prefix index defined on a virtual column. (Bug #27755892)

  References: This issue is a regression of: Bug #19806106.

• InnoDB: An in-memory table object for a table being truncated could be opened under improper metadata locking protection.

  A new method of preserving the AUTOINC counter value during a TRUNCATE PARTITION operation was implemented. (Bug #27754995)

• InnoDB: An assertion was raised during a mini-transaction commit that included dirty pages but no redo log records. The mini-transaction did not wait for space in the recently-closed buffer and subsequently marked the page with an invalid current LSN value. After the invalid LSN value was added to the flush list, a delayed mini-transaction added dirty pages with an earlier LSN value, breaking the flush list order.

  The innodb_log_checkpoint_fuzzy_now debug option was added. Enabling this option forces InnoDB to write a fuzzy checkpoint. Also, new internal redo log module counters were added for tracking LSN values. The log_lsn_buf_dirty_pages_added counter tracks the LSN value up to which dirty pages have been added, and the log_lsn_buf_pool_oldest_lwm counter tracks the low watermark LSN for the oldest modified block in the buffer pool. (Bug #27664539)

• InnoDB: The Contention-Aware Transaction Scheduling algorithm (CATS) tracks the number of transactions waiting for another transaction. In the case of a deadlock, the method by which the number of waiting transactions is approximated produced inflated values. (Bug #27646322)

• InnoDB: Attempting to acquire an InnoDB ticket while holding a victim transaction lock caused a lock order violation. (Bug #27626681)

  References: This issue is a regression of: Bug #23476050.

• InnoDB: An attempt to read a zero-length BLOB value raised an assertion failure during recovery. The BLOB value was not yet completely inserted. (Bug #27617389)

• InnoDB: A race condition occurred when acquiring a table lock for a table that was already in memory. (Bug #27586419, Bug #27577704)

  References: This issue is a regression of: Bug #26848711.

• InnoDB: After restarting the server, attempting to create an encrypted table returned a generic error from the storage engine. (Bug #27577339)

• InnoDB: The use of Contention-Aware Transaction Scheduling (CATS) with a spatial index caused a transaction to wait indefinitely. (Bug #27572937)

• InnoDB: Incorrect parsing of a tablespace name that included multibyte characters caused a DROP DATABASE operation to fail. (Bug #27566937)

• InnoDB: A check performed to identify temporary tables caused an error when the data dictionary attempted to open a partitioned table. (Bug #27565997)

• InnoDB: A schema mismatch error reported during an import tablespace operation failed to print mismatched table flags in a readable format. (Bug #27542720)

• InnoDB: A DDL operation failed to wait for a FULLTEXT index optimization operation to finish. (Bug #27326796)

  References: This issue is a regression of: Bug #24938374.

• InnoDB: An unnecessary check for read-only transactions was removed from the trx_set_rw_mode() function. Thanks to Sandeep Sethia for the patch. (Bug #27211287, Bug #88739)

• InnoDB: A DDL operation that added a foreign key constraint raised an assertion when it accessed a stale memory object that belonged to the parent table. (Bug #27208858)

• InnoDB: A failing assertion occurred after initiating a memcached get operation. (Bug #26876594)

• InnoDB: A corrupt index ID encountered during a foreign key check raised an assertion. (Bug #26654685)

• InnoDB: A DROP TABLE operation on a table residing in a general tablespace raised an assertion when checking index status in the post-DDL phase of the operation. (Bug #26523254)

• InnoDB: An internal deadlock during a DDL operation resulted in a long semaphore wait followed by a server exit. (Bug #26225783)

• InnoDB: A DDL operation encountered a serious error due to an invalid lock upgrade. (Bug #26225783)

• InnoDB: On a Windows 64-bit system, invalid buffer pool configuration values caused the server to exit on startup. (Bug #26100239, Bug #86370)

• InnoDB: A debug code block related to mutex type definitions was removed. It no longer differed from the non-debug version of the code block. (Bug #24952279, Bug #83529)

• InnoDB: A server failure during recovery occurred due to a log parsing buffer overflow. Space is now allocated to the log parsing buffer dynamically and incremented as necessary until it reaches the size specified by innodb_log_buffer_size. (Bug #24734190)

• InnoDB: Table locking was optimized for concurrent updates on the same table by avoiding costly iterations over the table locks queue.

  Thanks to Zhai Weixiang for the contribution. (Bug #18955152, Bug #72948)

• Packaging: On Windows, the mysqld.exe --help command displayed unnecessary and unexpected details in the system output, which this fix eliminates. (Bug #27894020)

• Partitioning: Rollback of ALTER TABLE ... TRUNCATE PARTITION statements was not handled correctly. (Bug #27603025)

  References: See also: Bug #87562, Bug #26710839.

• Partitioning: For a partitioned table, partition update time could be incorrect after rebuilding the table or restarting the server. (Bug #27073100)

• Replication: If the option --skip-log-bin was specified at server startup to disable binary logging, no warning message was issued if both expire_logs_days and binlog_expire_logs_seconds were set to a nonzero value. The --skip-log-bin option meant there were no binary log files to expire, so the outcome of these binary log expiration period settings (which is that the expire_logs_days value is ignored) had no practical effect. However, the warning message is now issued, so that the situation can be corrected before binary logging is enabled for the server. (Bug #27699608)

• Replication: The log messages generated when a member fails to join the group have been improved, for example when group_replication_group_name on the member joining the group does not match the seed's group_replication_group_name this is now described in the log message. (Bug #27628695)

• Replication: If a MySQL server is started in crash recovery mode with an innodb_force_recovery setting of 4 or greater, which places InnoDB into read-only mode, the semisynchronous replication master plugin should not be initialized. Due to a change in plugin initialization order in MySQL 8.0, the server assumed that if the plugin was registered, it had also been initialized, and incorrectly attempted to deinitialize it. The handling of this plugin in crash recovery mode has now been corrected so that the deinitialization function is not executed when the plugin has not been initialized. (Bug #27481872)

• Replication: The use of replication filters or binary log filters can cause issues when they are applied to tables that are updated with XA transactions. Filtering of tables could cause an XA transaction to be empty on a replication slave, and empty XA transactions are not supported. Also, with the settings master_info_repository=TABLE and relay_log_info_repository=TABLE on a replication slave, which became the defaults in MySQL 8.0, the internal state of the data engine transaction is changed following a filtered XA transaction, and can become inconsistent with the replication transaction context state.

  Due to these issues, the use of replication filters or binary log filters in combination with XA transactions is not supported. This fix adds the new error ER_XA_REPLICATION_FILTERS, which is logged whenever an XA transaction is impacted by a replication filter, whether or not the transaction was empty as a result. If the transaction is not empty, the replication slave is able to continue running, but you should take steps to discontinue the use of replication filters with XA transactions in order to avoid potential issues. If the transaction is empty, the replication slave stops. In that event, the replication slave might be in an undetermined state in which the consistency of the replication process might be compromised. In particular, the gtid_executed set on a slave of the slave might be inconsistent with that on the master. To resolve this situation, isolate the master and stop all replication, then check GTID consistency across the replication topology. Undo the XA transaction that generated the error message, then restart replication. (Bug #27442477)

• Replication: When mysqlbinlog was used to read a binary log event larger than 128KB, the read failed because 128KB was returned as the size, instead of the actual size. The issue has now been fixed. Thanks to the Facebook team for the patch. (Bug #27417084, Bug #89326)

• Replication: When a transaction larger than the binary log transaction cache size (binlog_cache_size) was flushed to a temporary file during processing, and the flush failed due to a lack of space in the temporary directory, the flush error was not handled correctly. No message was written to the error log, and the binary log cache was not cleared after the transaction was rolled back. Now, in this situation, the server takes an appropriate action based on the binlog_error_action setting (shut down the server or halt logging), and writes a message to the error log. When the transaction is rolled back, the server checks for flush errors and clears the binary log cache if any occurred. (Bug #27399620, Bug #89272)

• Replication: When GTIDs are in use for replication, replicated transactions that are filtered out on the slave are persisted. If binary logging is enabled on the slave, the filtered-out transaction is written to the binary log as a Gtid_log_event followed by an empty transaction containing only BEGIN and COMMIT statements. If binary logging is disabled, the GTID of the filtered-out transaction is written to the mysql.gtid_executed table. This process ensures that there are no gaps in the set of executed GTIDs, and that the filtered-out transactions are not retrieved again if the slave reconnects to the master. Previously, this process was not done for CREATE DATABASE, ALTER DATABASE, and DROP DATABASE statements, but it is now carried out for those statements as well as for others. (Bug #27308751, Bug #88891)

• Replication: On a multithreaded slave, when a STOP SLAVE statement is executed on the slave, followed by a START SLAVE statement, the error log can report a different position in the binary log for the slave SQL thread when exiting, compared to the position reported for the slave SQL thread at the subsequent initialization.

  For a multithreaded slave, the position reported for the SQL thread on exit is a low water mark, up to which the replication stream is consistent and has no gaps. Transactions appearing before the position are guaranteed to have committed, but transactions after the position may have committed or not. However, this low water mark was being reported before the process to stop the worker threads was actually carried out, and the low water mark was subsequently updated by a checkpoint routine during that process. The timing of the log message has now been changed so that the final low water mark is reported as the position for the SQL thread on exit. (Bug #27300658)

• Replication: An online upgrade of a replication group that was running a version of MySQL earlier than 5.7.23 was not possible. When a member running a version earlier than either MySQL 5.7.22 or MySQL 8.0.11 tried to join a group with members running MySQL 5.7.21 or earlier, it failed to join the group because MySQL 5.7.21 does not send the value of it lower_case_table_names system variable. This mean that the later version compared the lower_case_table_names variable with an invalid value and exited the group. The work around is to take the group offline, upgrade to version 5.7.23 or 8.0.12 and later. Subsequent upgrades of the group can then be carried out without taking the group offline. See Upgrading Group Replication. (Bug #90794, Bug #27991334)

• Replication: It was possible to change the value of gtid_purged while running the Group Replication plugin, even though this variable should never be modified in such cases. (Bug #90146, Bug #27724561)

• Replication: The PIT_TRANSACTIONS_NEGATIVE_CERTIFIED, the PIT_TRANSACTIONS_ROWS_VALIDATING and the PIT_TRANSACTIONS_LOCAL_ROLLBACK member messages were not being correctly decoded. (Bug #90077, Bug #27692831)

• Replication: In certain situations, such as during distributed recovery procedure, the certification info garbage collection was purging more data than it should, resulting in conflicts not being detected. The garbage collection procedure has been improved to take this case in consideration. (Bug #89938, Bug #27652526)

• Replication: The ER_GRP_RPL_SQL_SERVICE_FAILED_TO_RUN_SQL_QUERY error was being logged incorrectly. (Bug #89788, Bug #27590534)

• Replication: Using group_replication_force_members to unblock a group, for example after losing majority, sometimes failed with error 1231. (Bug #86957, Bug #26394418)

• Group Replication: When the group_replication_applier channel's applier thread encountered an error, the master_log_name and end_log_pos values shown in the resulting error message were incorrect, since the events in a transaction are replicated before they are written to the binary log of the member where the transaction originated, with the result that the final values are unknown at the time they are applied on the replica by the applier thread. To avoid confusion, any such error messages encountered by a group_replication_applier channel no longer contain the binary log name or position. (Bug #89146, Bug #27368735)

• macOS: On macOS, the server autostarted at system boot time even when set not to in the MySQL preference pane. (Bug #27969174, Bug #90672)

• Microsoft Windows: For debug builds on Windows, CREATE SPATIAL REFERENCE SYSTEM with an empty definition string could raise assertions. (Bug #27672683)

• In the client library, signals were incorrectly unblocked for ppoll(). Thanks to Facebook for the patch. (Bug #28075623, Bug #90999, Bug #28096808, Bug #91067)

• Certain ALTER TABLE statements could disable table encryption. (Bug #28045585)

  References: This issue is a regression of: Bug #27389878.

• Under some conditions, MyISAM code compilation failed due to a missing include file. (Bug #28039150, Bug #90898)

• The REGEXP_REPLACE() function truncated its result when used in an UPDATE statement. (Bug #28027093, Bug #90870)

• REGEXP_REPLACE() results from one result set row could carry forward to the next row, resulting in accumulation of previous results in the current row. (Bug #27992118, Bug #90803)

• A heap overflow vulnerability in the MySQL client library was fixed. (Bug #27980823)

• Users who had the SUPER privilege were not permitted to modify the keyring_operations system variable. (Bug #27976270)

• Full-text search auxiliary tables created in MySQL 5.7 and MySQL 8.0.11 with timestamps of 0 caused a data dictionary upgrade failure when upgrading from MySQL 8.0.11 to MySQL 8.0.12. (Bug #27960500)

• A “source and destination overlap in memcpy” Valgrind error was returned when upgrading from MySQL 5.7. (Bug #27945658, Bug #90691)

• Linux binaries were unable to dump core after executing a setuid() call. (Bug #27929894, Bug #90642, Bug #23337428, Bug #21723)

• RESET PERSIST did not flush variable values from memory. (Bug #27924206)

• For debug builds, using CREATE TABLE ... SELECT to create a table with a long numeric string and indexing the column could lead to an assertion being raised. (Bug #27909771)

• For debug builds, RESET PERSIST for an unknown variable could cause an assertion to be raised. (Bug #27903874, Bug #90546)

  References: This issue is a regression of: Bug #27374791.

• Updates to column metadata of a view referencing a partitioned table could cause a server exit. (Bug #27903842)

• For debug builds, an assertion could be raised for UNION queries when computing the combined data type of a GEOMETRY column and SELECT * FROM (SELECT NULL). (Bug #27903792, Bug #90550)

• For debug builds, starting the server with thread_stack=0 and executing ALTER TABLE could raise an assertion. (Bug #27899274, Bug #90525)

• SHOW TABLE STATUS could cause a server exit. (Bug #27897815)

• For generated columns that used the INTERVAL() function, incorrect behavior could occur. (Bug #27881102)

• For Fedora 28, MySQL Community RPM packages would not install if Connector ODBC native packages were installed. (Bug #27850721)

• It was possible to drop the Performance Schema. (Bug #27830283)

• A runtime error could occur for calls to mysql_sys_var_longlong() or String::copy(). (Bug #27822413, Bug #27808412)

• An assertion was raised by error logging code that used an inappropriate error number. (Bug #27792903)

  References: This issue is a regression of: Bug #27462408.

• The CONVERT TO CHARACTER SET charset_name clause for ALTER TABLE produced a syntax error if charset_name was DEFAULT. (Bug #27760787, Bug #90222)

  References: This issue is a regression of: Bug #27389878.

• An unencrypted connection could result from a client connection attempt specifying that an encrypted connection was required, if the server was not configured to support SSL. (Bug #27759871)

• INSERT INTO ... SELECT did not perform as many checks as INSERT INTO ... VALUES() with respect to requiring valid geometry values for geometry columns. (Bug #27756083)

• REGEXP_INSTR() did not return an error message when used with invalid match mode flags. (Bug #27751277)

• Some data dictionary table definitions were modified to align more closely with their initial design. Additionally, the data dictionary version number was incremented, and code that refers to data dictionary table indexes now uses symbolic index names instead of numeric values. (Bug #27745526, Bug #90196)

• Executing a prepared statement to insert rows with large number of parameters was very slow. (Bug #27699248)

• ibd2sdi could exit due to performing a divide-by-zero operation. (Bug #27692051, Bug #90071)

• The mysql-boost-8.0.4.tar.gz source distribution omitted NDB sources. (Bug #27690232)

• SHOW GRANTS could cause a server exit if executed for an unprivileged user who inherited privileges from an anonymous user. (Bug #27678129)

• The parser incorrectly permitted the index name for CREATE INDEX to be empty. (Bug #27676427)

• Failures of server restarts initiated by systemd or mysqld_safe were not written to the error log. (Bug #27675050, Bug #90012)

• For debug builds, if a transaction was in read-only mode, trying to create or drop a histogram could raise an assertion. (Bug #27672693)

• If the server was started with --skip-grant-tables, SET ROLE DEFAULT caused a server exit. (Bug #27672154)

• Casting of VARBINARY values to TIME values was not performed correctly. (Bug #27665997)

• Error checking was not performed after creation of a condition for a const reference. (Bug #27665085)

• The dictionary information object managed by the MeCab model was used after the MeCab model was released. (Bug #27660368)

• Attempting to abort events of the MYSQL_AUDIT_AUTHENTICATION_CLASS class caused a server exit. These events are now not abortable. (Bug #27645636)

• These resource-group statements are now disallowed in stored functions because they cause an implicit commit, which is not permitted in a stored function: CREATE RESOURCE GROUP, ALTER RESOURCE GROUP, DROP RESOURCE GROUP. (Bug #27638623, Bug #89914)

• Audit log filter rules did not permit class names to be specified as an array of strings. That is now possible. Example:

  {
    "filter": {
      "class": [
        { "name": [ "connection", "general", "table_access" ] }
      ]
    }
  }

  (Bug #27628325)

• On Windows, if the Visual C++ Redistributable for Visual Studio had been removed, MySQL uninstallation using the MSI installer failed. (Bug #27621546)

• mysql_secure_installation now loads the validate_password component rather than the deprecated validate_password plugin. (Bug #27619667)

• REGEXP_LIKE() using an expression containing a question mark (?) was not handled correctly when included in a prepared statement. (Bug #27595368)

• REGEXP_SUBSTR() did not always handle NULL patterns correctly. (Bug #27572258)

• Some SHOW CREATE TABLE statements could raise an assertion. (Bug #27569314)

• Spurious stages could be seen in the Performance Schema events_stages_current table. (Bug #27566220)

• Triggers for a table could execute in an order different from that specified by their PRECEDES/FOLLOWS clauses. (Bug #27544152)

• Attempts to set some persisted variables at startup (such as keyring_operations and mandatory_roles) produced a privilege violation error and the server did not start or did not set the variable. (Bug #27523095)

• Under some conditions, an unused-variable warning occurred in the MeCab full-text search code. Thanks to Laurynas Biveinis for the patch. (Bug #27519952, Bug #89598)

• Within stored procedures, it was not possible to write a condition handler to catch foreign key failures if the table involved used the FEDERATED storage engine. (Bug #27509959)

• It was possible to drop a primary key that served as the parent key in a foreign key relationship. In addition, for table creation, the SQL layer now checks whether there is a parent key for the foreign key and reports an error if it is missing. (Bug #27506922, Bug #89570, Bug #11754696, Bug #46337)

• With the NO_ENGINE_SUBSTITUTION SQL mode disabled, an error (rather than substitution) occurred for CREATE TABLE and ALTER TABLE if the desired engine was disabled using the disabled_storage_engines system variable. (Bug #27502530)

• Key rotation using the keyring_encrypted_file keyring plugin could cause a server exit. (Bug #27497018)

• If MySQL Enterprise Firewall was configured with a large number of rules, server shutdown could take a long time. (Bug #27492122)

• SET PERSIST_ONLY var_name = DEFAULT persisted the current global value, not the default value. (Bug #27489026)

• If MySQL is configured with the LINK_RANDOMIZE CMake option enabled, the --gc-sections compiler option is now disabled; the combination causes segmentation faults. (Bug #27484106)

• During server installation, the mysql.infoschema, mysql.session, and mysql.sys reserved accounts were associated with the mysql_native_password authentication plugin even when the default plugin was caching_sha2_password. (Bug #27454299)

• Some window functions used the wrong precision for their aggregated data types. (Bug #27452179)

• For a hexadecimal value H, SELECT with WHERE (-1) - H raised an assertion. (Bug #27452082)

  References: See also: Bug #27041382. This issue is a regression of: Bug #21982792.

• A memory leak in the pfs-t unit test was fixed. Thanks to Yura Sorokin for the patch. (Bug #27440735, Bug #89384)

• Multiple invocations of CHAR() within a stored program could yield different results once an invalid code point was passed as the argument. (Bug #27410088)

• With automatic_sp_privileges enabled, the EXECUTE and ALTER ROUTINE privileges were not correctly granted to routine creators. (Bug #27407480)

• Some queries that used ordering could access an uninitialized column during optimization and cause a server exit. (Bug #27389294)

• For debug builds, an assertion was raised (rather than an error) for ALTER TABLE for an ARCHIVE table with a GEOMETRY NOT NULL column containing empty strings. (Bug #27330634, Bug #89088)

• Accounts that authenticated with the auth_sock authentication plugin could not connect using older clients. (Bug #27306178)

• If system tables could not be read or initialized, the server wrote an incomplete error message. Thanks to Daniël van Eeden for a patch related to the fix. (Bug #27302337, Bug #89001)

• RENAME USER failed when used to rename a role that was not granted to any account or role. (Bug #27284699)

• The keyring_okv plugin did not always fail over to the standby OKV server when the primary server was unavailable. (Bug #27244099)

• The Performance Schema could attempt to access invalid SOURCE column values. (Bug #27231036)

• Mishandling of internal privilege structures could cause a server exit. (Bug #27230925)

• For debug builds, use of a too-small type for calculating the determinant in GIS computations could cause an assertion to be raised. (Bug #27135504)

• For attempts to increase the length of a VARCHAR column of an InnoDB table using ALTER TABLE with the INPLACE algorithm, the attempt failed if the column was indexed.

  If an index size exceeded the InnoDB limit of 767 bytes for COMPACT or REDUNDANT row format, CREATE TABLE and ALTER TABLE did not report an error (in strict SQL mode) or a warning (in nonstrict mode). (Bug #26848813)

• In MySQL 5.7.19, the mysql.gtid_executed table was excluded from dumps. This table is no longer excluded but its data is not dumped. (Bug #26643180, Bug #87455)

  References: This issue is a regression of: Bug #82848, Bug #24590891.

• The slave_rows_search_algorithms system variable controls how rows are searched for matches when preparing batches of rows for row-based logging and replication. Specifying INDEX_SCAN as one of the search algorithms performs an index scan if an index is present. In the situation where a different primary key is used on the master and the slave, and a unique key is present on the slave, a bug in the code meant that the index scan was not being performed as it should be, and a slower table scan was being performed instead. The issue has now been corrected so that an index scan is used. (Bug #26450129, Bug #23311892, Bug #81500, Bug #81501)

• Certain cases of subquery materialization could cause a server exit. These queries now produce an error suggesting that materialization be disabled. (Bug #26402045)

• Spatial operations on geographic GeometryCollection values could return ER_GIS_INVALID_DATA for valid geometries. (Bug #26174808)

• For MyISAM tables, particular sequences of INSERT and DELETE statements could cause table corruption. (Bug #25541037)

• The SHOW CREATE USER statement requires the SELECT privilege for the mysql database, except to see information for the current user. In addition, for the current user, the statement requires the SELECT privilege for the mysql.user system table for display of the password hash in the IDENTIFIED AS clause; otherwise, the hash displays as <secret>. (Bug #24911117)

• CREATE FUNCTION produced an error if there was no default database and the function body called another function. (Bug #24357244, Bug #82350)

• For debug builds, a SELECT statement executed within a stored program could raise an assertion if derived table resolution failed. (Bug #23221336)

• If mysqldump or mysqlpump were used to dump binary data without the --hex-blob option, reloading the dump file could produce spurious warnings (values were inserted correctly regardless of the warnings). Such values are now written preceded by the _binary introducer to silence the warnings. (Bug #22601255, Bug #80150)

• String comparisons involving wildcards could reference uninitialized memory. (Bug #12635103)

• REGEXP_REPLACE() raised a Valgrind warning when invoked with an invalid capture group. (Bug #87842, Bug #27612255)