Release Notes for MariaDB Enterprise Server 10.4.25-16

When the parser converts a string from the binary character set to a multi-byte character set (such as utf32), an invalid string could be produced. (MDEV-23210)

When rows are inserted into an intermediate temporary table via the LOAD DATA INFILE statement ,and then the rows are copied from the temporary table to a persistent table, the rows are not written to binary log if binlog_format=MIXED is set, which prevents the operation from properly replicating to replica servers. (MDEV-24667)

When innodb_disallow_writes=ON is set, mariadb-admin shutdown can hang. (MDEV-25975)

• Starting with this release, the innodb_disallow_writes system variable has been removed.

When a replica server's I/O thread receives an incomplete event group from the primary server, the replica server continues writing events to the relay log and does not raise an error. (MDEV-27697)

Semisync-replica server recovery fails to rollback a prepared transaction. (MENT-1534)

If a primary is shutdown during an active semi-sync connection during the period when the primary is awaiting an ACK, the primary hard kills the active communication thread and does not ensure the transaction was received by a replica. This can lead to an inconsistent replication state. (MDEV-11853)

With MariaDB Enterprise Cluster, powered by Galera, when wsrep_sst_method is set to rsync or mariabackup, the donor node does not transfer the correct binary logs to the joiner node with some configurations. (MDEV-27524)

When a window function is used in the global ORDER BY clause of a SELECT statement with a UNION, the statement should be rejected, but the server executes the statement and crashes with a segmentation fault. (MDEV-15208)

When a stored procedure queries a view and uses a for loop, the server can crash with a segmentation fault when the stored procedure is called twice in the same session. (MDEV-26009)

When innochecksum is executed on an encrypted tablespace file using the --page-type-summary or -S option, innochecksum crashes with a segmentation fault unless the --page-type-dump or -D option is also specified. (MDEV-27835)

When an invalid CREATE SEQUENCE ... RESTART statement is used inside of a CREATE PROCEDURE or CREATE FUNCTION statement, the server can crash while parsing the statement. (MDEV-28220)

When a table contains a virtual generated column that is defined using the IF() and DATE_FORMAT() functions, inserting a row into the table can cause the server to crash with a segmentation fault. (MDEV-24176)

When a non-updateable view is defined with ALGORITHM=TEMPTABLE, selecting data from the view can cause the server to crash with a segmentation fault. (MDEV-21028)

When a UNION of decimal types is performed in an ORDER BY clause, the server can crash with a segmentation fault. (MDEV-25994)

When a stored procedure executes a query that results in a mergeable derived table, the server can crash with a segmentation fault when the stored procedure is called twice in the same session. (MDEV-27212)

• Querying views can result in mergeable derived tables.

• Using subqueries with outer references can result mergeable derived tables.

When ANALYZE FORMAT=JSON is executed with a query that contains a subquery, the server can crash. (MDEV-28268)

When a query using a window function with an OVER (PARTITION BY ..) or OVER (ORDER BY ..) clause is executed, the server can crash with a segmentation fault. (MDEV-19398)

• If the query also uses WITH ROLLUP, this crash is more likely.

• If the query's OVER (PARTITION BY ..) or OVER (ORDER BY ..) clause uses an aggregate function, this crash is more likely.

When a query contains an outer join expression and a non-correlated subquery that the optimizer determines is low cost, executing the query can cause the server to crash if the optimizer also determines that the inner join expression can be eliminated. (MDEV-28437)

With MariaDB Enterprise Cluster, a joiner node's binary logs could be deleted during an Incremental State Transfer (IST), which causes the node to fail to start, because it can not read the binary logs. (MDEV-28583)

• This issue is confirmed to happen with wsrep_sst_method set to rsync, but it might also occur with other SST/IST methods.

ALTER TABLE on a large InnoDB table can hang. (MDEV-28415)

Online DDL fails while checking for instant alter condition. (MDEV-28060)

With MariaDB Enterprise Cluster (Galera), parallel async replication hangs on a Galera node when slave-parallel-threads greater than 1 is used. (MDEV-27568)

With MariaDB Enterprise Cluster (Galera), possible crash after a conflict of the applier thread with a stored procedure call triggered by the event scheduler. (MDEV-27713)

Server crashes when executing a shutdown statement after starting an XA transaction. (MDEV-26575)

Possible server crash during mariadbd initialization if the number of GTIDs added since the last purge of the mysql.gtid_slave_pos tables is greater than or equal to the --gtid-cleanup-batch-size value. (MDEV-26473)

Possible server crash if an INSERT .. SELECT or REPLACE .. SELECT statement contains an ON expression in the top-level select and this expression used a subquery with a column reference. (MDEV-28578)

When OPTIMIZE TABLE is executed on a sequence, the server raises an ER_BINLOG_UNSAFE_STATEMENT warning, even if binlog_format is set to ROW or MIXED. (MDEV-24617)

• The warning can appear in the MariaDB Error Log as the following:

  [Warning] Unsafe statement written to the binary log using statement format since BINLOG_FORMAT = STATEMENT. Statement is unsafe because it uses a system function that may return a different value on the slave Statement: OPTIMIZE TABLE SEQUENCE_NAME
  

When an UPDATE or DELETE statement uses an OR clause, the server raises an ER_UPDATE_WITHOUT_KEY_IN_SAFE_MODE error. (MDEV-18304)

• The error can appear on the client as the following:

  You are using safe update mode and you tried to update a table without a WHERE that uses a KEY column
  

When an UPDATE statement uses a subquery inside an IN() clause, the optimizer could incorrectly choose to perform a full table scan (with type: ALL) when a range scan (with type: range) would be more efficient. (MDEV-22377)

When a table contains multiple ENUM columns with identical values, the values could become corrupt if multi-byte character sets are used. (MDEV-28078)

When DECIMAL columns are read from information_schema.PROCESSLIST, the server could incorrectly raise an ER_BAD_DATA warning. (MDEV-28131)

• The warning can appear on the client as the following after executing SHOW WARNINGS:

  Encountered illegal value '' when converting to DECIMAL
  

With sql_mode=ORACLE, the parser allows functions to be called using PACKAGE_NAME.FUNCTION_NAME(), but the parser raises a ER_PARSE_ERROR error if the function is called using DATABASE_NAME.PACKAGE_NAME.FUNCTION_NAME(). (MDEV-28166)

When a view is used to update multiple rows of a temporal table in a single UPDATE statement, the server incorrectly raises a ER_WARN_VIEW_WITHOUT_KEY warning. (MDEV-22973)

• The warning can appear on the client as the following after executing SHOW WARNINGS:

  View being updated does not have complete key of underlying table in it
  

When using mariadb-binlog with --raw and --stop-never, events from the primary server's currently active log file are not written to their respective log file specified by --result-file. (MDEV-14608)

When a prepared statement is used to execute EXPLAIN, a different query plan can be returned on the second execution. (MDEV-19631)

When slave_compressed_protocol is enabled with semi-synchronous replication, the numbering on packet headers can become out of sync between the primary and replica servers, but the inconsistency does not have any negative impact with release builds of MariaDB Enterprise Server. (MDEV-25580)

The server and MariaDB Enterprise Backup can raise unnecessary warnings about tablespace IDs in some scenarios. (MDEV-27343)

• The warning can appear as the following:

  InnoDB: Allocated tablespace ID TABLESPACE_ID for DATABASE_NAME/TABLE_NAME, old maximum was 0
  

• This warning can be written to standard output (stdout) when preparing a backup with MariaDB Enterprise Backup.

• This warning can be written to the MariaDB Error Log when InnoDB performs crash recovery at startup.

With MariaDB Enterprise Cluster, the joiner node fails to complete an SST when innodb_log_group_home_dir specifies a directory different than datadir. (MDEV-27740)

When a non-partitioned table using the DATA DIRECTORY clause is converted to a partitioned table, the DATA DIRECTORY clause is silently ignored, and the partitioned table is moved to the default directory. (MDEV-27065)

Starting with this release, the server will raise the WARN_OPTION_IGNORED warning.

When a table contains an invisible column, mariadb-dump does not produce the correct CREATE TABLE statement in the backup file. (MDEV-28253)

When a Spider table is queried using IF(COUNT() ..), the server can raise an ER_BAD_FIELD_ERROR error. (MDEV-25116)

With MariaDB Enterprise Cluster, State Snapshot Transfers (SSTs) can fail on the donor node when binary logs are enabled. (MDEV-26201)

With MariaDB Enterprise Cluster, when wsrep_node_address contains an IPv6 address and wsrep_sst_receive_address is set to the default value of AUTO, State Snapshot Transfers (SST) can fail. (MDEV-26171)

With temporal tables, replication can break if a system versioned table has been created on the replica via mariadb-dump, if the original table was altered before. (MDEV-28254)

mariabackup does not detect multi-source replication primary. (MDEV-21037)

Master_SSL_Crl shows Master_SSL_CA value in SHOW SLAVE STATUS output. (MDEV-28428)

MariaDB Audit plugin produces corrupted log entries for CONNECT events. (MENT-1438)

mariadb-dump does not create a dump where the sql_mode is set correctly for SQL/PL packages. (MDEV-27816)

Missing binlog data for INSERT .. ON DUPLICATE KEY UPDATE (MDEV-28310)

mariabackup prepare fails for incremental backups if a new schema is created after full backup. (MDEV-28446)

Optimizer uses all partitions during an UPDATE and ignores partitioning filters. (MDEV-28246)

When INSERT .. ON DUPLICATE KEY UPDATE is executed and binlog_format is set to MIXED, the server raises an ER_BINLOG_UNSAFE_STATEMENT warning even though the statement is written to the binary log in row-based format. (MDEV-21810)

• The warning can appear in the MariaDB Error Log as the following:

  [Warning] Unsafe statement written to the binary log using statement format since BINLOG_FORMAT = STATEMENT. INSERT... ON DUPLICATE KEY UPDATE  on a table with more than one UNIQUE KEY is unsafe Statement: INSERT INTO TABLE_NAME VALUES (..) ON DUPLICATE KEY UPDATE KEY_NAME = KEY_VALUE
  

When INSERT .. ON DUPLICATE KEY UPDATE is executed and binlog_row_image is set to FULL, the server does not write unchanged columns to the binary log. (MDEV-21810)

Crash recovery fails if the configured server ID does not match the server ID in the crashed data directory. (MENT-1535)

PAM v2 plugin (auth_pam) produces zombie processes. (MENT-1443)

With MariaDB Enterprise Cluster (Galera), wsrep_incoming_addresses does not include address details. (MENT-1527)

With HashiCorp key management (hashicorp_key_management), dynamic changes to hashicorp_key_management_cache_timeout and hashicorp_key_management_cache_version_timeout system variables are ignored. (MENT-1512)

Galera snapshot transfer fails to upgrade between some major versions. (MDEV-27437)

On Windows, MSI installer does not install client shared libraries. (MDEV-28581)

After upgrade, mysql.plugin table has an entry for Semi-sync Replication, though this former plugin functionality is now built-in. (MDEV-21873)

On RPM-based distributions, the Spider storage engine is distributed as a separate RPM package. (MENT-1092)

• In prior releases, Spider's components were installed with the server's RPM package.

• Starting with this release, Spider adds unixODBC as a dependency, so Spider has been moved to a separate RPM package to avoid adding new dependencies to the server's RPM package. For users of RPM-based distributions who loaded the Spider plugin with previous ES 10.4 releases, Spider's new RPM package and dependencies must be installed after upgrading to this release.

• To install Spider's new package and dependencies on CentOS, Red Hat Enterprise Linux (RHEL), or Rocky Linux:

  $ sudo yum install MariaDB-spider-engine
  

• To install Spider's new package and dependencies on SUSE Linux Enterprise Server (SLES):

  $ sudo zypper install MariaDB-spider-engine
  

When the hashicorp_key_management encryption plugin is loaded, mariadb-dump should be used to take a logical backup before all minor and major upgrades due to changes that are not backward-compatible:

• Starting with this release, the hashicorp_key_management encryption plugin rejects encryption keys that contain non-digit characters in the hex notation.

• If a database is encrypted with encryption keys that only contain digit characters in the hex notation, the encryption keys will continue to work with the new version.

• If a database is encrypted with encryption keys that are no longer valid, it is recommended to upgrade by installing the new version to a clean system and restoring a logical backup of the data. However, another alternative is to migrate the system to a new encryption key that only contains digit characters in the hex notation prior to the upgrade. If the alternative route is taken, extra care must be taken to ensure that all encrypted data uses the new encryption key. For help with this process, contact MariaDB Support.

• When performing minor and major upgrades on Debian and Ubuntu, a logical backup must be taken before upgrading the server and plugin packages. When the old packages are upgraded, a message about plugin incompatibility is printed, but the server and plugin packages are upgraded anyway.

• When performing minor upgrades on CentOS, RHEL, Rocky Linux, and SUSE, a logical backup can be taken before or after upgrading the server package. When the old packages are upgraded, a message about plugin incompatibility is printed, and the server package is upgraded, but the plugin package is not. The old plugin package must be manually removed, and then the new plugin package can be installed.

• When performing major upgrades on CentOS, RHEL, Rocky Linux, and SUSE, a logical backup must be taken before upgrading the server and plugin packages. The old server and plugin packages must be manually removed, and then the new server and plugin packages can be installed.

• When the plugin package is manually removed, the plugin configuration file can also be removed, so it is recommended to backup the file.