TaurusDB Kernel Version Release History

2026-07-30

• New features and performance enhancements
  • TaurusDB now supports vector storage and search, meeting enterprise-level requirements for vector search capabilities.
  • TaurusDB allows the optimizer to choose index condition pushdown (ICP) based on cost estimation, improving the accuracy of execution plan selection.
• Fixed issues
  • Restoration to a new instance fails because the backup is too large (over 5 GB).
  • Fixed an issue where temporary files are not removed if a SELECT INTO OBS export operation is interrupted during OBS file import/export.
  • Fixed an issue where LOAD DATA fails when both import and export operations are performed on an instance during OBS file import/export.

2026-04-30

New features and performance enhancements

• SQL tuning
  • LEFT JOIN elimination is supported. For complex queries that meet the prerequisites, the system transforms the SQL queries to eliminate unnecessary LEFT JOIN subqueries, thereby improving query performance.
  • Replacing COUNT(NOT NULL COLUMN) with COUNT(*) helps the optimizer choose a more appropriate index, improving the execution efficiency of the COUNT operation.
• Data can be imported from OBS to TaurusDB and exported from TaurusDB to text files in OBS.

2026-01-04

• New features and performance enhancements
  • All DDL and DCL operations on instances with multi-tenancy enabled can be recorded in binlogs, which ensures normal replication in DRS migration scenarios.
  • The semaphore timeout, which defaults to 600 seconds in the community, can now be adjusted to meet your needs.

• Fixed issues
  • Applied InnoDB and SQL bug fixes from MySQL 8.0.42 and 8.0.43 releases.
  • Optimized index queries for filtering ENUM columns via IN operators.

2025-10-20

• New features and performance enhancements
  • Flashback query. You can use undo log-based flashback query to roll back table-level data to a specified point in time (accurate to 0.1 seconds). You can also flash back just a single table.
  • Dynamic data masking. When an application queries a database, the database can mask sensitive data before returning the results to the application.

• Fixed issues
  • Fixed inaccurate query results in some parallel queries with multi-value indexes.
  • Fixed inaccurate query results in some scenarios where in-list tables are converted to temporary tables.
  • Fixed inaccurate query results in some scenarios where plan cache is enabled.

2025-08-05

Fixed issues:

The following bugs in the open-source community are fixed: #33996054, #33801312, #33781442, #35115909, #35012146, #34801210, #30556595, #36658450, #34574604, and #33691659.

2025-04-30

New features:

TaurusDB is compatible with version 5.7. Certain reserved keywords in TaurusDB 8.0 have been removed or adjusted to partially support MySQL 5.7 syntax. Most new keywords have been deleted, and only EMPTY, JSON_TABLE, LATERAL, and RECURSIVE have been changed from reserved keywords to non-reserved keywords. If a user-defined function has the same name as any of these four keywords, enclose the function name in backticks when calling it, for example, select `EMPTY`();.

If you are migrating data from MySQL 5.7 to TaurusDB 5.7, you are advised to use workload replay to systematically verify their compatibility.

2025-04-23

• New features and performance enhancements
  • Enhanced parallel query: You can use COUNT(DISTINCT column(s)) to speed up parallel query.
  • Database recycle bin: Truncated tables and dropped databases can be moved to the recycle bin. Some third-party binlog collection tools may be incompatible. You are advised to assess whether to upgrade the kernel version based on your specific service requirements.
• Fixed issues
  • Fixed the issue that workloads are occasionally interrupted when a read replica pulls binlogs.
  • Fixed the issue that the Truncate-LSN lag alarm is triggered by LSN stacking in some scenarios.
  • Fixed the issue that the synchronization is occasionally abnormal during an upgrade from 2.0.51.240305 or earlier to a later version.

2025-01-14

• New features and performance enhancements
  • Fast binlog positioning: If you pull binlogs from a TaurusDB instance using auto-positioning, it can be time-consuming to locate the correct binlog position when there are a large number of unread binlogs. For example, if there are 60,000 unread binlogs, it would take about 60 minutes to automatically locate the correct binlog position. With fast binlog positioning enabled, the time required can be reduced to about 2 seconds.
  • Enhanced parallel query: You can use EXPLAIN statements, including EXPLAIN FORMAT=TREE, EXPLAIN ANALYZE (supported in 2.0.60.241200 and later), and EXPLAIN FORMAT=JSON introduced in MySQL 8.0 and traditional EXPLAIN statements, to view parallel execution plans.
• Fixed issues
  • Fixed the issue that workloads are interrupted occasionally when Flink is used to pull binlogs from read replicas in some scenarios.
  • Fixed the issue in the open-source community that the results are incorrect due to duplicate JSON column index data (Bug #33334928).
  • Optimized the traffic distribution to OLAP.
  • Changed the ROW keyword from a reserved keyword to a non-reserved keyword.

2024-10-18

• New features and performance enhancements
  • Partition-level MDL: In MySQL Community Edition, you cannot perform both data manipulation language (DML) operations for accessing data of partitioned tables and data definition language (DDL) operations for maintaining partitions at the same time. This means that DDL operations can only be done during off-peak hours. To resolve such an issue, this version introduces partition-level metadata lock (MDL) to refine the lock granularity of a partitioned table from the table level to the partition level. After partition-level MDL is enabled, DML operations and specific DDL operations (such as adding and deleting partitions) on different partitions can be both performed, greatly improving concurrency between partitions.
  • Table recycle bin: After this function is enabled, the DROP TABLE statement that meets conditions does not directly delete a specified table. Instead, the table is temporarily stored in the recycle bin. When the maximum retention period expires, the table is automatically deleted in the backend. You can change the retention period of a deleted table in the recycle bin. You can also restore or permanently delete a table from the recycle bin at any time.
  • Users without the SUPER privilege are not allowed to grant high-risk dynamic privileges (SYSTEM_VARIABLES_ADMIN and SYSTEM_USER).
• Fixed issues
  • Fixed the issue that CPU resources of each tenant are not strictly allocated based on the configured ratio in resource preemption scenarios.
  • Allowed Statement Outline to support views and EXPLAIN ANALYZE statements.

2024-07-19

• New features and performance enhancements
  • Optimized hot row update: Hot rows are frequently updated for flash sales, concert ticket reservations, and train ticket bookings for popular routes. This version enhances hot row update, which can be enabled automatically or manually. After hot row update is enabled, hot rows can be updated efficiently.
  • Non-blocking DDL: When you perform a DDL operation, if the target table has uncommitted long transactions or large queries, the DDL operation continuously waits for obtaining the MDL-X lock. As a result, service connections are stacked and blocked. This version supports non-blocking DDL, which allows new transactions to enter the target table even if the MDL-X lock cannot be obtained, ensuring the stability of the entire service system.
  • Multi-tenancy: This feature enables a database to serve multiple tenants, maximizing database resource utilization.
  • Binlog pull for read replicas: You can use read replicas as the data source to establish a binlog replication link and synchronize the binlogs in real time, which helps reduce the load on the primary node.
  • Column compression: TaurusDB introduces fine-grained column compression to reduce data page storage and save costs. Two compression algorithms, ZLIB and ZSTD, are provided. You can select either of them to compress infrequently accessed large columns based on the compression ratio and compression and decompression performance.
  • INTERVAL RANGE partitioned tables: In previous versions, if the data to be inserted into an existing RANGE partitioned table exceeds the range of existing partitions, the data cannot be inserted and an error is returned. With the support for INTERVAL RANGE partitioned tables in this version, the database can now create partitions based on rules specified by the INTERVAL clause when new data exceeds the range of existing partitions.
  • LIST DEFAULT HASH partitioned tables: This feature supports two types of partitions at the same level: LIST and HASH. Data is first inserted into LIST partitions. Data that does not comply with the LIST partitioning rules is placed in the DEFAULT partition. If the DEFAULT partition has multiple partitions, HASH rules are used. LIST DEFAULT HASH partitioned tables are usually used in scenarios where LIST VALUES are unevenly distributed and cannot be fully enumerated.
• Fixed issues
  • Optimized the table-level restoration performance.
  • Optimized the execution performance of read replicas of a high-spec instance in high-concurrency scenarios.

2024-03-30

• New features and performance enhancements
  • Added global consistency, which provides strongly consistent reads at the cluster level with low performance loss.
  • Added the SHOW BINARY LOGS NO BLOCK syntax, which prevents transaction commits from being blocked during the execution of SHOW BINARY LOGS.
  • Optimized the UNDO TRUNCATE capability, which solves the issue of undo space expansion caused by a large number of writes.
  • Enhanced the degree of parallelism for full restoration, which optimizes the backup and restoration efficiency.
• Fixed issues
  • The query results of window functions are incorrect, or errors occur when window functions are executed.
  • Database nodes crash when specific PREPARE statements are repeatedly executed after plan cache is enabled.
  • An error is reported due to inconsistent character sets when stored procedures are executed in sequence.
  • Query results do not meet the expectation when an on-disk hash join is performed after PQ is enabled.
  • An error is reported due to duplicate primary keys when a query involves performing a GROUP BY operation on temporary table fields.

2024-01-30

• New features and performance enhancements
  • Enhanced composite partitioning: In addition to RANGE-HASH and LIST-HASH of MySQL Community Edition, added RANGE-RANGE, RANGE-LIST, LIST-RANGE, LIST-LIST, HASH-HASH, HASH-KEY, HASH-RANGE, HASH-LIST, KEY-HASH, KEY-KEY, KEY-RANGE, and KEY-LIST.
  • Added the forward compatibility with GROUP BY implicit/explicit sorting in MySQL 5.7.
  • Added the forward compatibility with the max_length_for_sort_data parameter in MySQL 5.7, which optimizes the file sorting performance in specific scenarios.
  • Optimized the issue that accessing views in information_schema is slow due to incorrect execution plan selection.
  • Added the EXISTS subquery in PQ.
  • Optimized restoration of database tables or instances to a specific point in time.
• Fixed issues
  • OpenSSL is upgraded.
  • The default value SYSTEM of the time_zone parameter impacts the efficiency of concurrent SQL statement execution in some scenarios.
  • SQL query results are incorrect when conditions are partially pushed down to a materialized derived table.
  • Performance suffers after PQ is enabled for on-disk hash joins in some scenarios.
  • The permissions page is not updated accordingly after a user is granted permissions on a database through the console and the database is later deleted in non-console mode.

2023-11-24

• New features and performance enhancements
  • Added forward compatibility of datetime, timestamp, and time field behaviors.
  • Added on-disk hash joins in PQ.
  • Added INSERT and REPLACE SELECT functions in PQ.
  • Added log printing mechanism for connection and disconnection, which helps you locate connection-related issues quickly.
  • Added some useful information in slow query logs, which help you locate slow SQL statements.
  • Allowed you to dynamically enable binlog.
  • Optimized the NDP bloom filter.
  • Allowed you to use the CAST(... AS INT) syntax.
  • Optimized the Nested Loop Join + Distinct performance.
  • Identified slice ID corresponding to the slow I/O quickly.
  • Added the sal_init log, which helps you to locate storage API timeout issues.
• Fixed issues
  • Fixed an issue where the trx_id and cpu_time fields were missing from full SQL logs.
  • Fixed a conversion error in prepared statements when comparing an integer column with a string parameter in the WHERE clause.
  • Fixed an issue where concurrent execution of DDL statements and queries on a read replica had a low probability of causing a crash.
  • Fixed an issue where binlogs were not cleaned up in time after a sudden short-term surge in their count.
  • Fixed an issue where enabling Parallel Query (PQ) for multi-table JOIN SQL statements might result in inconsistent execution results.
  • Fixed an issue where query performance was lower than expected due to incompatibility between Backward Index Scan and Index Condition Pushdown (ICP).
  • Fixed an issue where the weight_string function did not support the LEVEL clause.
  • Fixed an issue where identical SQL statements using different indexes produced inconsistent results in specific scenarios.
  • Fixed an issue where recycle LSN failed to advance for a long time when both Near Data Processing (NDP) and Parallel Query (PQ) features were enabled in certain scenarios.

2023-08-31

• New features and performance enhancements
  • Added support for storing full and incremental backups on read replicas, which reduces the memory and CPU usage of the primary node.
  • Optimized UNDO damage location: When the undo damage occurs during startup, the undo damage log and the corresponding table name are printed.
  • Improved the query performance of read replicas.
  • Added the conversion of IN predicates to subqueries.
  • Supported large-scale commercial use of the NDP feature.
  • Optimized execution plans using statement outline.
  • Supported round functions in PQ.
• Fixed issues
  • The ORDER BY LIMIT and ORDER LIMIT result sets do not overlap when fast sorting and priority queue sorting algorithms are used.
  • Returned results are correct for PQ statements.
  • No errors are reported when PREPARE statements are executed.
  • No PQ assertion errors are reported on UNION queries.
  • The results of full-text index queries are correct after a read replica is promoted to primary while a large amount of data is being inserted into the primary node.
  • When read replicas use the general_log and slow_log tables, warning logs will not be displayed.
  • After the value of the parameter innodb_lock_wait_timeout is changed, the actual timeout wait time is correct.
  • There is a low probability that the error "Failed to find page in slice manager" occurs during the promotion of a read replica to primary, causing the promotion to fail.
  • The percentage for the PWAL scanning progress in the SAL SQL log cannot exceed 100%.
  • When the sqlsmith tool is executed, the error "mysqld coredump" occasionally occurs in the EXPLAIN phase of query statements.
  • In SELECT DISTINCT CAST functions, datetime can be converted to the float type correctly.

2023-05-11

• New features and performance enhancements
  • Supported small-scale instances.
  • Optimized the solution when DDL statements on standby nodes fail.
  • Optimized the capacity calculation of salsql.
  • Supported the restriction on resources of a single SQL statement.
  • Supported the use of per thread for admin port and local socket.
  • Optimized the memory of pwalScanner.
  • Supported the modification of default_collation_for_utf8mb4 parameter.
  • Supported diagnosis on large transactions.
  • Supported the killing of idle transactions.
  • Accelerated incremental restoration.
  • Added database and account descriptions.
  • Supported the acceleration of buffer pool resize.
• Fixed issues
  • PTRC does not lead to inconsistent execution results of Nested-Loop Join.
  • No crash issue occurs when subqueries are sorted using Windows functions.
  • When using rewrites view, tables are not evaluated to turn left joins into inner joins.
  • Execution results are returned from decimal data that meets specified filter criteria.
  • Memory is aligned.
  • Scan_row is correctly recorded in full logs.

2023-05-17

Fixed an issue where queries sorting large JSON columns failed due to exceeding the sort buffer memory limit.

2023-04-02

Character sets are not used in combination in prepared statements.

2023-03-14

• New features

  Reduced primary/standby latency.

• Fixed issues
  • No error occurs when JSON-related functions are used in prepare statements.
  • Query results are returned when filter criteria are specified.
  • No null pointer error is reported after Windows functions generate an ephemeral disk table.
  • The crash issue caused by the use of null pointers in Windows functions is resolved.
  • Prepared statements are executed successfully.

2023-01-11

• New features
  • Supported SQL throttling.
  • Optimized read flow control.
  • Optimized the consistency of primary/standby execution plan.
  • Pre-created slices asynchronously.
• Fixed issues
  • No crash issue occurs when the system variable INNODB_VALIDATE_TABLESPACE_PATHS is disabled and the undo space truncate command is executed.
  • The query of information_schema.innodb_trx is fast.
  • The issue of inconsistent results is resolved: left joins now are turned into inner joins.
  • The crash issue caused by subquery optimization is resolved.
  • Values of the Instant field are correctly obtained under concurrent instant DDL and DML operations.
  • No OOM issue occurs when two InnoDB tables with FTS indexes are loaded.
  • No OOM issue occurs when the data dictionary of millions of tables is being updated.

2022-12-07

Scan errors triggered by Skip Scans are not displayed when a table with virtual columns is updated.

2022-11-16

During a primary/standby switchover, databases will not break down when connecting to the standby instance times out.

2022-09-23

• The If(...) statement in Condition_pushdown::replace_columns_in_cond is modified.
• The database does not break down when:
  • Storage functions are invoked recursively.
  • Multiple tables are deleted or full-text search is performed.
  • SQL query statements of multiple window functions are executed.
• Users with global permission can successfully run SHOW CREATE DATABASE.

2022-08-25

The ptrc crash problem in stored procedure is resolved.

2022-08-12

• New features and performance enhancements
  • Supported SQL throttling.
  • Added a limit to concurrent numbers of Faster DDL.
  • Supported all Faster DDL operations in row format.
  • Extended full SQL fields.
  • Optimized flow control.
  • Supported the quick timeout of ALTER TABLE.
  • Supported the query of plan cache.
  • Optimized statistics on standby nodes.
• Fixed issues
  • Fixed the issue that the standby node crashes after the partition table is renamed on the primary node.
  • Modified the default buffer size of the SQL tracer.
  • Fixed the issue that the standby node fails to be started when the truncated LSN on the standby node is far behind that on the primary node.
  • Fixed the issue that the execution plan is incorrect due to multiple SQL queries with the same range.
  • Fixed the crash issue caused by empty accounts.
  • Fixed the crash issue that may be caused by dropping databases.

2022-07-22

• Databases will not break down due to empty accounts.
• When a temporary table used for aggregation is updated, BLOB points to the latest data.

2022-05-16

• New features
  • You can enable or disable orphaned definer check control.
  • TaurusDB supports transparent transmission of proxy IP addresses.
  • You can set the consistency level of your proxy instances to session consistency.
• Fixed issues
  • The data dictionary on standby nodes is updated if DDL statements on the primary node are not submitted.
  • During a failover, the auto increment of the primary node is not rolled back.
  • The performance issue of standby nodes is resolved.