What's New in MariaDB Enterprise Server 11.8

An overview of changes, improvements, and what's new in MariaDB Enterprise Server 11.8

The most recent release of MariaDB Enterprise Server 11.8 is:

MariaDB Enterprise Server 11.8.6-3 Download Now

MariaDB Enterprise Server 11.8 introduces a wide range of enhancements spanning developer productivity, security, compatibility, observability, and support for modern workloads like vector search. The MariaDB Enterprise Server 11.8.2-0 Technical Preview adds the innovations from the MariaDB Community Server releases 11.5 to 11.8 to MariaDB Enterprise Server 11.4.

Vector Search

MariaDB Enterprise Server 11.8 continues to expand its native vector search capabilities, positioning MariaDB for AI-powered applications such as semantic search and recommendation systems. Already available in the MariaDB Enterprise Server 11.4 are:

New Data Type: VECTOR(N) to store multi-dimensional embeddings
Vector Indexing: Efficient similarity search with VECTOR INDEX
Distance Functions:
- VEC_DISTANCE_EUCLIDEAN() for L2 distance
- VEC_DISTANCE_COSINE() for angular similarity
Conversion Functions:
- Vec_FromText(json_array) and Vec_ToText(vector_column) transform between the text and binary vector formats
System Variables for Index Tuning:
- mhnsw_max_cache_size, mhnsw_default_distance, mhnsw_default_m, mhnsw_ef_search

Added to MariaDB Enterprise Server 11.8

Distance Functions:
- VEC_DISTANCE() auto-selects the best distance function based on the index configuration.
Optimization that makes vector search 30-50% (depending on the data) faster for the same recall
- Enabled automatically for applicable vectors
- Vectors are applicable if they can be gradually truncated to trade some recall for speed
- For example matryoshka embeddings as produced by OpenAI are applicable
Namespace support was added to HashiCorp Vault in MariaDB
- See the Hashicorp namespaces documentation for details
Audit logging buffer writes are possible now

Backported Features

MariaDB Enterprise Server 11.8.5-2

Log write buffering added to the SERVER_AUDIT and SERVER_AUDIT2 plugins (MENT-2438)
- The buffering is controlled by the added variables:
  - server_audit_file_buffer_size - defines the size of the buffer. The default value is 0 meaning there's no buffering at all. Setting non-zero value enables the buffering with the buffer of the specified size aligned by 8192. The maximum value is 65536.
  - server_audit_sync_log_file - flushes the buffer to the log file. While the log record is in the buffer, it cannot be seen in the log file. And if there are not many events to log, the time before records can be observed can be significant. So user can do SET GLOBAL server_audit_log_file=1 to this variable to force write the buffer to the file, to make sure he doesn't miss the recent records.
Two new fields are available via SHOW REPLICA STATUS (MENT-2129)
1. Connects_Tried, which provides the number of attempts the replica has made to connect to the primary, and
2. Master_Retry_Count, which provides the number of times the replica will attempt to connect to a primary before giving up.
The authentication plugin caching_sha2_password has been added, not loaded by default (MDEV-37600)

Emergency Recovery Across Major Versions

Behavioral change: A MariaDB 11.x server can now start with data files from MariaDB 10.x when innodb_force_recovery is set to 6
- This is a last-resort emergency-recovery path for extracting a logical dump from a corrupted or incompatible database, not a substitute for the normal upgrade procedure
- See Configure the InnoDB Redo Log: Emergency Recovery Across Major Versions (MDEV-39303, available in Enterprise Server 11.8 builds based on MariaDB 11.8.7 and later)

Indexes, SQL Functions, and Query Enhancements

UUID Generation: New functions UUID_v4() and UUID_v7() for modern and time-ordered UUIDs
Multi-Table DELETE Enhancements: Support for ORDER BY and LIMIT
Single-Table DELETE Enhancements: Now it allows index hints
NEW SHOW CREATE SERVER: Recreate server objects similar to SHOW CREATE TABLE
DBMS_OUTPUT: Messages submitted by DBMS_OUTPUT.PUT_LINE() are not sent to the client until the sending subprogram or trigger completes

Performance Improvements

Optimization that makes vector search 30-50% faster (more details in the Vector Search section)
Segmented key cache for Aria storage engine
- aria_pagecache_segments system variable
Add analyze_max_length option to not collect statistics for long char/varchars
- See Skipping Long CHAR/VARCHAR Columns for more information

Data Types and Compatibility

UTF-8 by Default: utf8mb4 replaces latin1 and legacy utf8, ensuring full Unicode support, including emojis
Default Collation: utf8mb4_uca1400_ai_ci is now the standard for Unicode character sets
Extended TIMESTAMP Range: Increased upper bound to 2106 on 64-bit systems
ROW Type Enhancements:
- ROW types are now usable as stored function return values
- The support for %ROWTYPE, TYPE OF, and RECORD(...) declarations for Oracle-like compatibility:
  DECLARE TYPE DeptRecTyp IS RECORD ( dept_id NUMBER(4), dept_name VARCHAR2(30), mgr_id NUMBER(6), loc_id NUMBER(4) );
Triggers:
- BEFORE UPDATE OF col1, col2 limits trigger execution to specific column updates:
  CREATE TRIGGER mytrigger BEFORE UPDATE OF col1, col2 ON t1 FOR EACH ROW …
Use of SIGNAL SQLSTATE '02TRG' allows skipping a row operation

Stored Procedures: Now support default parameter values:

CREATE OR REPLACE PROCEDURE p1(param1 INT, param2 INT DEFAULT 1)

Associative arrays: DECLARE TYPE .. TABLE OF .. INDEX BY
Added caching_sha2_password plugin, see Authentication Plugin - SHA-256 for more information

Enhancements to System Versioned Tables

System Versioned Tables is a powerful feature for auditing changes to data. Enabling System Versioned Tables is as easy as creating a table by using
```
CREATE TABLE contracts (...) WITH SYSTEM VERSIONING;
```
Or enabling the feature for an existing table by using
```
ALTER TABLE contracts ADD SYSTEM VERSIONING;
```
In both cases, invisible fields will be created in the table to track the timestamps and period for which the data is valid. A DBA/DevOps might want these fields to be visible

It is now possible to change such implicit fields to explicit ones by the following types of statements:

SET @@system_versioning_alter_history= keep;
ALTER TABLE contracts ADD COLUMN rs TIMESTAMP(6) AS ROW START, ADD COLUMN re TIMESTAMP(6) AS ROW END, ADD PERIOD FOR SYSTEM_TIME (rs,re);

Security

New Authentication Plugin—PARSEC:

Based on elliptic curve cryptography

Mitigates the replay and man-in-the-middle attacks Example on how to create a user using the new authentication plugin:

CREATE USER 'MariaDBUser'@'%' IDENTIFIED VIA PARSEC USING PASSWORD('MyPassword123!');

This will result in:

SHOW GRANTS FOR MariaDBUser@'%';

Grants for MariaDBUser@%
GRANT USAGE ON *.* TO `MariaDBUser`@`%` IDENTIFIED VIA parsec USING 'P0:lhXyNv1cIxpB8EnTxR7ON7S7:1l3rWRW1/jw45yrvYXB8eh02wzk7lcJcz4CMc
Ww2b+8'

Unix Socket Enhancements:
- Now supports explicit OS user mapping via IDENTIFIED VIA UNIX_SOCKET AS 'user'
- It is also possible to specify more than one OS user with the usual OR syntax:
  CREATE USER dba IDENTIFIED VIA UNIX_SOCKET AS 'jack' OR IDENTIFIED VIA UNIX_SOCKET AS 'jill';

Replication & Clustering

Improved Replication Lag Monitoring:

SHOW REPLICA STATUS now includes:

Master_last_event_time

Slave_last_event_time

Master_Slave_time_diff

Information Schema Table for Replication Status:\

MariaDB [test]> SELECT Master_last_event_time,Slave_last_event_time,Master_Slave_time_diff FROM information_schema.slave_status\G
*************************** 1. row ***************************
Master_last_event_time: 2024-08-13 07:32:38
Slave_last_event_time: 2024-08-13 07:32:37
Master_Slave_time_diff: 1

New Option – --slave-abort-blocking-timeout: Kills blocking non-replication queries after a timeout
Galera SST Automation: SST user is now auto-created and managed internally
Backported MDEV-35304 : Add {{Connects_Tried}} & {{Master_Retry_Count}} from CS-12.0

Key Management

KMS Plugin Enhancement: The file_key_management plugin can now read keys from a Unix socket, not just from files
File Key Management Encryption Plugin: Added key rotation capability; see this page for details. In addition, a new Information Schema table was added, FILE_KEY_MANAGEMENT_KEYS

Observability & Information Schema

Temporary File Disk Space Limits:
- max_tmp_session_space_usage and max_tmp_total_space_usage prevent runaway disk usage
New Status Variables: tmp_space_used, max_tmp_space_used

New Information Schema Views:

SLAVE_STATUS to view replication lag via SQL
USERS for password state and expiration monitoring

SEQUENCES to introspect auto-generated sequences:

SELECT * FROM INFORMATION_SCHEMA.SEQUENCES\G
*************************** 1. row ***************************
       SEQUENCE_CATALOG: def
        SEQUENCE_SCHEMA: test
          SEQUENCE_NAME: s_name
              DATA_TYPE: tinyint
      NUMERIC_PRECISION: 8
NUMERIC_PRECISION_RADIX: 2
          NUMERIC_SCALE: 0
            START_VALUE: 100
          MINIMUM_VALUE: -127
          MAXIMUM_VALUE: 126
              INCREMENT: 10
           CYCLE_OPTION: 0

Enhanced ANALYZE FORMAT=JSON:
- Includes r_index_rows, r_icp_filtered
Thread Naming for Diagnostics: Thread names are now more descriptive
Galera Information Schema: New WSREP_BF_ABORTS Information Schema table
Galera Information Schema: New WSREP_THD_STATE and WSREP_THD_STATE_HISTORY Information Schema tables
Galera Information Schema: New Information Schema table WSREP_CONNECTIONS
Galera Information Schema: New Information Schema WSREP_CERT_KEYS and WSREP_CERT_KEYS_HISTORY tables

PL/SQL

Support for Oracle-stype INDEX BY tables (associative arrays) was backported from MariaDB 12.1 in stored routines and anonymous blocks, with this declaration syntax:

TYPE type_name TABLE OF rec_type_name INDEX BY idx_type_name;

Where:

type_name supports explicit and anchored data types (e.g t1.col1%TYPE)
The INDEX BY clause supports integer and string data types
rec_type_name supports scalar and record data types

The implementation supports the following associative array methods:

FIRST
LAST
NEXT
PRIOR
COUNT
EXISTS
DELETE

Tool Improvements

mariadb-dump:
- The --dir=<path> dumps each database to its subdirectory
- Supports parallel dumps and restores
- New --update-history converts row_end TIMESTAMPs during export
mariadb-import:
- --dir=<path> restores from the dumped directory structure
- Supports --database, --table, --ignore-database, --ignore-table for selective restore
- --innodb-optimize-keys: defers index creation to speed up data loads
mariadb-test (mtr):
- mtr can be started with the --enable_serveroutput option to enable DBMS_OUTPUT messages
  - See this section for details.
mariadb (command-line client)
- New --enable_serveroutput option to enable DBMS_OUTPUT messages at start time
  - See this section for details
- New serveroutput and noserveroutput command to enable and disable those messages at runtime
  - See this section for details.

Userstat Plugin Enhancements

Improved Insights:
- Index usage via QUERIES in INDEX_STATISTICS
- Query operation counts: ROWS_INSERTED, ROWS_UPDATED, KEY_READ_HITS, etc...
- New stats in CLIENT_STATISTICS, and USER_STATISTICS
- Table I/O metrics: PAGES_ACCESSED, and PAGES_READ_FROM_DISK

Security Vulnerabilities (CVE) Fixed in MariaDB Enterprise Server 11.8

For a complete list of security vulnerabilities (CVE) fixed across all versions of MariaDB Enterprise Server, see the Security Vulnerabilities Fixed in MariaDB Enterprise Server page.

CVE ID (with cve.org link)

CVSS base score (v3.1)

Enterprise Server 11.8 Release

10.0

8.0

8.0