PolarDB:PolarDB for MySQL 8.0.2 release notes

Description

• Added support for adding read-only column store nodes to a secondary cluster in a Global Database Network.

• Added support for REORGANIZE PARTITION and LIST DEFAULT HASH to merge a LIST partition into a HASH partition, and split a VALUE into a new LIST partition.

  Added support for using the prefix of a HASH partition as the entire HASH partition name.

• Added support for large wide tables by extending the maximum number of columns for tables in X-Engine to 10,000.

• Added the DISTANCE expression for calculating cosine similarity, dot product similarity, and Euclidean distance.

• Added support for using IMCI to accelerate vector retrieval queries.

• Added support for the table value constructor syntax. You can now use the VALUES ROW syntax in DML statements.

• Added support for using a stored procedure to add columnstore indexes to all tables in a database in batches.

• Added support for sharded tables in Outline. A single Outline can now match statements with the same template across multiple sharded tables.

• Added support for collecting and using statistics for non-indexed columns, including a unified management method for extended statistics.

• Added support for dynamically adjusting Order Index selection results at runtime based on actual conditions.

Description

• Optimized the startup speed of read-only column store nodes. The column store engine can now start independently of the column store snapshot offset.

• Optimized the startup time of column store nodes in scenarios with a large number of tables.

• Enhanced the Orca feature to support specifying a username and password by using the username:password syntax in the Redis AUTH and HELLO commands.

• Optimized DDL scenarios for the X-Engine storage engine. The system now automatically enables the pre-commit feature for X-Engine.

• Adjusted the lock granularity for the sorted set data type in Orca. Only flags are used for lock checks on secondary indexes.

• Optimized the thread startup and teardown logic for logical pre-reading and simplified the memory check logic.

Description

• Fixed an issue where DDL operations on a table with a view could report an error or not take effect after a high-availability (HA) switchover.

• Fixed an issue where the parallel query parameter pq_enable_xengine_mpp did not take effect when set to OFF.

• Fixed an issue where the auto-increment value of a partitioned table occasionally rolled back, causing a duplicate key error on INSERT.

• Fixed the session-level binlog_rows_query_log_events permission to support the execution of SQL backups and rollbacks in DMS.

• Fixed an issue where an error was not reported for duplicate items when setting a flagset-type variable, such as optimizer_switch.

• Fixed an issue where enabling the win magic switch caused a subquery rewrite overflow due to insufficient base_ref_items space, resulting in a process crash.

• Fixed an expression evaluation error when converting an INLIST to a JOIN result set if the INLIST VALUES contains a null value.

• Fixed an issue where a HASH JOIN in the plan could cause incorrect parallel query results.

• Fixed an issue where queries that included Lateral Derived Tables (LDTs) could return incorrect results.

• Fixed the implementation of the Redis protocol transaction model in Orca, including the WATCH, UNWATCH, MULTI, EXEC, and DISCARD commands.

• Fixed an issue where the results returned by the BITOP, ZRANGEBYLEX, ZRANGEBYSCORE, ZREVRANGEBYSCORE, and ZREVRANGEBYLEX commands in Orca were inconsistent with native Redis behavior.

• Fixed an issue that could cause a process to crash when the SUBSTRING_INDEX expression processes an empty string from GROUP_CONCAT and the separator's length is greater than 1.

• Fixed an issue where the CAST(REGEXP_SUBSTR(...)) AS DECIMAL statement caused a process to crash.

• Fixed an issue where data validation was not performed when changing the dimension of a VECTOR type with a DDL statement.

• Fixed a data dictionary inconsistency that occurred during a minor version upgrade from a version that did not support vector retrieval to one that did.

• Fixed a rare issue of duplicate data on the primary key index when partition MDL was used.

• Fixed compatibility, scaling, and other issues between IMCI Serverless and internal memory elasticity.

• Fixed an issue where using the ALTER TABLE DROP statement to simultaneously delete multiple full-text indexes (FTS) could cause residual metadata.

• Fixed an issue where the uniqueness of a unique index was not used to accelerate query execution.

• Fixed a cluster crash that occurred in extreme cases when performing DML operations on a table with an auto-increment column after online partition maintenance (MDL) was enabled.

• Fixed a potential cluster crash when a multi-value index was selected in a Multi-Range Read (MRR) optimization.

• Fixed an issue where Orca did not lock the database when executing certain management commands, such as the ORCA_OPTIMIZE, AUTH, and ACL commands.

• Fixed an issue where using a parallel query could lead to incorrect results when querying data that includes BLOBs.

• Fixed an issue where certain types of queries were not correctly routed in adaptive execution.

• Fixed an issue in cold data scenarios where the metadata cache on a read-only node was not updated after archiving to CSV or ORC format.

• Fixed a cluster crash due to a memory issue during DDL operations on a table containing a BLOB field.

• Fixed an issue where IMCI did not support online rebuilds, causing columnstore indexes to become unavailable.

Description

An INSERT ... ON DUPLICATE KEY UPDATE statement no longer causes a table's auto-increment to roll back after an RO node is switched to an RW node.

Description

• Optimized the statistics sampling mechanism by increasing the sampling frequency for large tables and ensuring coverage of the value range for each column. This improves the accuracy of the optimizer's cost estimation.

• Optimized the serverless process for column store nodes.

  New column store nodes now serve requests only after their indexes are built. This prevents slow queries caused by SQL statements being routed to the new node to run row store plans.

• Reduced the performance overhead of saving partition information during SQL execution in IMCI, improving query performance for partitioned tables.

Description

• Fixed incorrect result sets that occurred when merging IN clauses in a nested subquery.

• Fixed a crash that occurred when a correlated subquery used a nested window function.

• Fixed a crash during migration caused by a concurrency issue between frequent INSTANT DDL operations and background compaction tasks.

• Fixed B-tree corruption caused by adding an auto-increment column and creating a reverse-order primary key index on that column.

• Fixed an issue where the Autoinc value rolled back to 0 after clearing all data from a table.

• Fixed an issue that caused IMCI to fall back to the row store when processing an expression with multiple subqueries containing SUM.

• Fixed incorrect results that could occur when using a JSON_ARRAY expression in a subquery with IMCI enabled.

• Fixed a system crash that could occur during query transformation if the number of tables exceeded 61.

• Fixed an error that occurred when re-importing data into a UNIQUE CHECK IGNORE partitioned table by using LOAD DATA with REPLACE.

• Fixed an Autoinc rollback issue in a partitioned table after an HA switchover.

• Fixed an issue that interrupted the database and table restoration process when the same table was restored multiple times.

• Fixed an issue where an IMCI could not be rebuilt online, making it unavailable.

• Fixed a rare query failure caused by IMCI accessing invalid string data.

• Fixed an issue where using Index_merge optimization for UPDATE statements could cause performance degradation.

• Fixed an issue triggered when an index join in IMCI used a constant as a join condition.

• Fixed an issue with condition pushdown for table function to prevent failures during parallel execution.

• Fixed an issue where specifying a non-existent index in an Outline caused an error after multi-node ePQ was enabled.

• Fixed an issue that prevented the use of asynchronous DDL when adding an IMCI by modifying the COMMENT on a table with a full-text index.

• Fixed a potential cluster crash during the undo initialization phase when Force Recovery was enabled.

• Fixed an issue that could cause incorrect results when updating an InnoDB temporary table.

• Fixed a result set conflict between advanced transaction splitting in the proxy and the Fast Query Cache.

• Fixed an issue where parallel query results could be incorrect if the query contained a constant table.

• Fixed an issue where JSON_Length returned incorrect results when a wildcard character was used.

• Fixed an issue where an RO node became stuck during the HA phase, causing prolonged HA times.

Description

Upgraded the file system version and optimized the process of restoring a cluster from a writable snapshot.

Description

Fixed a compatibility issue between the file system version and the control system.

Description

• Added the JSON_VALUE() function, which simplifies creating an index on a JSON column.

• Supports continuous application updates during Copy DDL execution.

• Supports pushing down JOIN conditions to the corresponding views.

• In-Memory Column Index (IMCI) now supports a Hybrid Plan-based index join, which uses InnoDB indexes for JOIN operations.

• Supports direct access to data tables on AI nodes through federated external tables.

• The OSS OUTFILE feature now supports parallel export to the OSS engine.

• Introduced a high-priority DDL capability to improve the success rate of DDL execution.

• Added support for locking archived tables.

• Added a hybrid row-column optimizer and cost-based row-column routing.

• The optimizer now supports inner join and semi-join elimination.

  Additionally, outer join elimination is optimized for more scenarios.

• Optimized large transactions in binary logging to avoid blocking other transaction commits.

• The Hybrid Plan in IMCI now supports partitioned tables, virtual columns, and the ENUM, SET, and GIS column types.

• Added support for dynamically switching to the column store during the in-row execution process.

• The OSS query acceleration feature now supports archiving partitioned tables to OSS external tables. Manual or automatic partition archiving now synchronously generates OSS_FILE_FILTER query acceleration data.

Description

• PolarDB for MySQL 8.0.1 and later now displays outline call information in EXPLAIN.

• Improved the ORDER BY LIMIT pushdown feature for PolarDB for MySQL to support pushdown to JOIN tables.

• PolarDB IO threads can now be adjusted dynamically without a restart.

• Reduced the time required for query optimization on single-table queries.

• Improved the performance of DML transaction replay in IMCI.

• Optimized the In-Memory Column Index (IMCI) bloom filter hash algorithm and reduced build overhead by using string prefixes to construct the bloom filter.

• Optimized the performance of simulated AIO.

• Improved the performance of 3AZ clusters under low concurrency.

• Enhanced the nonblocking DDL feature to support more DDL statements.

• Optimized the memory usage of metadata used in IMCI queries.

• Adjusted the priority of PolarFS transaction processing threads to improve the DDL performance of PolarDB for MySQL.

• Improved the performance stability of the file system during DDL processes that frequently create and delete table files.

Description

• Fixed a crash caused by a hash join between the same common table expression (CTE) in a parallel query.

• Fixed an out-of-memory (OOM) error during IMCI execution when a query contained too many UNION ALL clauses.

• Fixed an issue where user threads on an RO node experienced long waits for a free page when the buffer pool was full.

• Fixed an issue where the statistics sampling mechanism sampled large tables too frequently.

• Fixed an issue where removing an RO node with innodb_polar_log_rdma_transfer enabled could block monitoring connections for a period of time.

• Fixed an issue where certain implicit conversions in IMCI were incompatible with the MySQL CASE statement.

• Added support for global secondary indexes (GSIs) when archiving partitioned tables to OSS external tables.

• Fixed an issue where connection errors in acknowledgment (ACK) packets from a secondary node were not recognized.

• Fixed an out-of-memory (OOM) error that occurred during an analyze table operation on a large table.

• Fixed an issue where unneeded subqueries were not removed during the view merging process.

• Fixed an issue that caused inconsistent result sets when an inlist was converted to a JOIN and the inlist type was inconsistent.

• Fixed an issue where data could not be read when a query rewrite used index_merge to access a single-row subquery.

• Fixed an issue that caused a semi join crash during subquery conversion.

• Fixed a crash in query rewrite when converting a scalar subquery with an anti join to a derived table.

• Fixed a precision loss issue with floating-point numbers in JSON.

• Fixed an incompatibility between the Log RDMA feature of global consistency high-performance mode (SCC) and GDN secondary clusters.

• Fixed an overflow issue caused by converting a large double value to uint during the parallel computation planning process.

• Fixed an issue where the REF or MULTI RANGE access cost was underestimated for secondary indexes that did not include the partition key in a UCI partitioned table.

• Fixed an issue where double values in JSON would incorrectly lose precision during parsing.

• Fixed a bug that caused the database to hang during a manually triggered checkpoint.

• Fixed a crash caused by an interaction between query rewrite's derived merge and the scalar subquery unnesting rule.

• Fixed a security vulnerability in the crash recovery process of MySQL external XA transactions.

• Fixed an issue where statistics on an IMCI partitioned table became invalid during partition pruning.

• Fixed an issue where InnoDB tables with a function index failed cold data archiving.

• Fixed an issue where setting the values for optimizer_switch and parallel_query_switch with an incorrect separator caused an error instead of a warning.

• Fixed an issue by disabling the index merge intersection access path for DML statements by default to improve performance and prevent deadlocks.

• Fixed an issue where the optimizer failed to select the optimal JOIN ORDER due to cardinality estimation errors.

• Fixed an error that occurred when a column-store column was included in a CREATE TEMPORARY TABLE SELECT statement.

• Fixed a potential assertion failure during table optimization.

• Fixed an issue where the query result cache could be inaccurate when time and datetime are compared during single- and multi-machine parallel operations.

• Fixed an incorrect IP address in the audit log for connection reuse under the session-level connection pool.

• Fixed an issue where the cost of point queries and range queries based on the primary key was underestimated for UCI partitioned tables.

Description

• Optimized the memory usage of metadata used during queries in IMCI.

• Optimized the memory footprint of the IMCI optimizer for high-concurrency queries.

Description

• Fixed an error when creating a GSI on an empty table using parallel DDL.

• Fixed incorrect statistics estimation for large tables with BLOB fields.

• Fixed an issue where the number of rows scanned by a derived table was not recorded in the audit log and slow query log.

• Fixed a space leak caused by an exception during the Shrink process.

• Fixed an error during a parallel query on a hybrid partition containing both InnoDB and ORC partitions.

• Fixed incorrect result sets caused by incorrectly collapsing a subquery.

• Fixed an issue that prevented an ADD COLUMN operation on a partitioned table with subpartitions after upgrading the kernel version from 8.0.1 to 8.0.2.

• Fixed a potential cluster crash during a parallel query where the predicate condition contained a DEFAULT expression.

• Fixed excessive permissions for DLM policy execution.

• Fixed an issue where archiving failed for an archived table with a descending index.

• Fixed an issue that prevented reclaiming BLOB space from an original record when its BLOB column was updated.

• Fixed inaccurate cardinality estimation that prevented the use of IMCI for JOIN conditions on BIGINT columns with large values.

• Fixed inaccurate statistics for IMCI partitioned tables caused by sampling only a subset of partitions during statistics updates.

• Fixed an issue where useless subquerys were not deleted during view merging.

• Fixed a potential parallel query crash caused by eliminating a CTE used as the right table of a LEFT JOIN.

• Fixed an issue that caused the transaction ID for read-write transactions in SQL Explorer to be 0 when the thread pool was enabled.

Description

The Fast Query Cache now logs the number of returned rows. This count is now available in the audit log for each cache hit.

Description

• The number of rows scanned by a derived table is now correctly recorded in the audit log and slow query log.

• Queries that combine a Limit Offset pushdown, an equality condition on a secondary index, and an IN list on the primary key now return the correct results.

• Comparisons between FLOAT and INT types now return the correct result.

Description

• The new limit_orderby_pushdown_mode parameter controls whether to push down LIMIT and ORDER BY clauses to derived tables or all branches of a UNION ALL statement. When the LIMIT value is less than the value of the limit_pushdown_to_union_threshold parameter, the LIMIT and ORDER BY clauses are pushed down to all branches of the UNION ALL statement.

• Added support for LEFT JOIN operations based on Top-K pushdown in IMCI.

• The optimizer can now convert a LEFT JOIN to an INNER JOIN based on HAVING conditions and the outer conditions of a VIEW.

• The optimizer now eliminates redundant DISTINCT operators in queries to improve execution speed.

Description

• Query acceleration is now supported for CSV-formatted data files on OSS.

• You can now run the INSTANT DROP PARTITION operation on a partitioned table that has a global secondary index (GSI) without invalidating the GSI.

• Improved the read performance of read-only nodes.

• Added support for LEFT JOIN elimination in UPDATE statements.

• Optimized the SQL Trace feature to correctly record SQL statements and their execution plans when a built-in stored procedure is called.

• Added support for the instant ADD COLUMN feature on tables with a columnstore index.

• Added support for non-equi predicate derivation to push down more predicates.

• Simplified the configuration for automatic request distribution between row and column stores. You can now execute SET use_imci_engine=FORCED to determine why a query was not routed to the column store.

• Optimized subquery execution to collapse and eliminate multiple subqueries whenever possible.

• Added support for using temporary tables when using the Hybrid Plan to accelerate wide-table queries.

• Improved the connection handling speed of the thread pool in low-concurrency scenarios.

• Added support for using the rman.import command on columnstore indexes.

• Improved X-Engine sampling statistics to provide more accurate compression ratio estimates.

• Added support for creating Tair read-only nodes to provide cache read capabilities.

• Access by system O&M engineers is now excluded from the slow_queries status metric during database performance monitoring.

• Reduced the metadata overhead for memory allocation and memory statistics in IMCI to save memory.

• Added support for parallel queries on single tables and partitioned tables in X-Engine.

• IMCI now supports the BatchProbe optimization to improve the performance of JOIN operations.

• Optimized the memory usage of metadata, file modules, and small in-memory objects in IMCI to reduce resident memory usage.

• Added support for using the columnstore index and global consistency (high-performance mode) features at the same time.

• Stats Manager can now automatically dispatch histogram collection tasks to read-only nodes or hot standby nodes for automatic histogram updates.

• Improved the I/O performance of PolarFS read-only nodes to reduce query latency.

Description

• Fixed an issue where subsequent scale-out requests could not be executed because a large SQL query holding pages without releasing them blocked the automatic cancellation of a previous scale-in operation in the Buffer Pool.

• Fixed an issue where cold data in CSV and ORC formats could not be deleted by using the dbms_oss.delete_table_file built-in stored procedure.

• Fixed a potential node crash during the automatic cancellation of a scale-in operation in the Buffer Pool under high concurrency.

• Fixed an issue where a query could not be dispatched to the columnstore index if it contained a subquery that returned an empty result set.

• Fixed an internal thread pool counter error that could occur when an ACL lock wait was encountered during asynchronous authentication.

• Fixed an issue that caused an assert message to be returned when multiplying Decimal data, if at least one multiplier was 0 and the result's scale exceeded 16.

• Fixed an issue in the one-click upgrade from RDS for MySQL to PolarDB for MySQL where a PolarDB for MySQL cluster restart interrupted the migration connection.

• Fixed a node crash when querying a hybrid partitioned table that contains both InnoDB and ORC partitions.

• Fixed data loss that occurred when repeatedly archiving data to an OSS external table.

• Fixed a node crash caused by accessing an invalid BLOB reference when pre-reading BLOB data.

• Fixed an issue where enabling the pre-read feature while using the columnstore index could cause intermittent node crashes.

• Fixed an issue where the HINT syntax was not supported by the join condition pushdown feature.

• Fixed an issue where the HINT syntax was not supported by the predicate pushdown subquery feature.

• Fixed incomplete HAVING condition pushdown.

• Fixed a potential node crash in a parallel query when referencing a projection column in a HAVING clause and executing a non-correlated subquery.

• Fixed excessive Autoinc init logging.

• Fixed an issue where the subquery merging feature did not support Optimizer HINT.

• Fixed a potential error with OSS external tables in ORDER BY DESC scenarios.

• Fixed an out-of-bounds memory access issue that occurred when compressing oversized strpack values in IMCI.

• Fixed a potential performance regression when pushing down conditions to a derived table.

• Fixed an issue where BIT fields were displayed incorrectly after a GROUP BY operation.

• Fixed a slow query performance degradation caused by strict concurrency control for background compaction tasks in X-Engine, which affected foreground operations.

• Fixed an issue that could produce incorrect results in a parallel query if a plan segment had multiple tables performing a partition-wise parallel scan and a parallel hash join operation was also run in the segment.

• Fixed an issue where the pull-up of some subqueries caused the column in group statement is ambiguous warning.

Description

• Improved the error message for importing cold data from OSS back to the InnoDB engine.

• Added support for tracking slow query execution in SQL Trace.

Description

• Fixed a memory leak during a Reload operation in features such as Statement Outline and Concurrency Control.

• Fixed an issue where a LEFT JOIN on tables in the INFORMATION_SCHEMA database returned incorrect results.

• Fixed an issue where a DLM policy remained active after the associated table structure was changed.

• Fixed an issue where running the INSERT INTO ON DUPLICATE VALUE command on a partitioned table resulted in a Can't find record on <table> error.

• Fixed a rare crash when executing the TRUNCATE PARTITION command on a partitioned table with an auto-increment column.

• Fixed an issue that caused parallel queries on large tables with tall or skewed B-trees to generate an insufficient number of parallel shards or distribute data unevenly among concurrent workers.

Description

• Added support for granting users global dynamic permissions to execute the SHOW_ROUTINE, FLUSH_USER_RESOURCES, FLUSH_TABLES, FLUSH_STATUS, and FLUSH_OPTIMIZER_COSTS commands.

• Disabled the small partial update feature for BLOBs to prevent potential data read errors when reading BLOB fields.

• When you use the SET_VAR hint to set the loose_optimizer_switch parameter, you no longer need to add a trailing comma.

Description

• Fixed an issue where an index was corrupted and table data became inaccessible due to differing index orders during an EXCHANGE PARTITION operation.

• Fixed a deadlock in DDL read-ahead during parallel DDL operations.

• Fixed an issue where an incorrect LSN offset during a checkpoint operation caused crash recovery to fail after an abnormal cluster restart.

• Fixed an issue where a node restart was required to recover from an Autoinc rollback.

• Fixed an issue where a table definition has changed error occurred when a Hybrid Plan was applied to an SQL statement containing an ORDER BY clause and returning NULL values.

• Fixed an issue where some built-in stored procedures did not correctly report an error when provided with an incorrect number of input variables.

• Fixed an issue where duplicate groups appeared in the result set when a parallel query pushed down GROUP BY to a worker for parallel execution on a partitioned table.

• Fixed an issue where the "Row size too large. The maximum row size for the used table type, not counting BLOBs, is 65535. This includes storage overhead, check the manual. You have to change some columns to TEXT or BLOBs" error was incorrectly reported when creating a table with a DECIMAL column.

• Fixed a rare crash during the undo cleanup process that occurred when using a temporary table with the flashback query feature enabled on a backquery machine.

• Fixed an issue where the optimizer selected an ordering index but failed to omit the sort operation during execution.

• Fixed an issue where the JOIN elimination optimization feature did not update the column information for the remaining tables, which could prevent the selection of the optimal access method for those tables.

• Fixed an issue where using the Materialization policy for a Semi-Join could produce an incorrect query result.

• Fixed an issue where a binlog record failed to uniquely identify a record in a partitioned table whose primary key does not include the partition key when binlog_row_image=MINIMAL.

• Fixed an issue where a UCI index might not be unique when the primary key does not include the partition key in the following scenarios:

  • Pulling up the inner table of a semi-join.

  • Eliminating GROUP BY or DISTINCT.

  • Adding a unique key dependency for JOIN elimination.

  • Decorrelating a correlated scalar subquery by converting it into a derived table through a window function.

  • Processing GroupJoin queries with the columnstore index.

Description

• Added support for using a hint to disable global consistency (high-performance mode) for individual SQL statements.

• Added support for using correlated subqueries in the GROUP BY clause and for eliminating them when they are complex.

• Added support for Concurrency Control and Statement Outline in a multi-master cluster (database/table). When Concurrency Control or Statement Outline rules are added or deleted on a primary node, the changes automatically synchronize to all other primary nodes in the cluster.

• You can add Tair cache nodes to a PolarDB for MySQL cluster to read data from the PolarDB for MySQL database.

Description

• Accelerated the thread pool's handling of connection requests in low-concurrency scenarios.

• Optimized the IMCI startup process, reducing the preparation time for the columnstore index.

• Improved data read performance on read-only (RO) nodes for primary key lookups that use a secondary index.

• Optimized the transaction mask calculation process in IMCI.

Description

• Fixed an issue where CCL rules matched by a DIGEST value could become invalid after a cluster upgrade.

• Fixed a node crash caused by adding backticks to the input parameters when calling some built-in stored procedures.

• Fixed an issue where querying TopK data with IMCI returned incorrect results when sorting in descending order on tables with a large number of NULL values.

• Fixed an issue where a read-only columnstore node could not recover columnstore data when started during a DDL execution on a table with a columnstore index.

• Fixed an issue where elastic parallel query (ePQ) could not be used for queries that include a recursive CTE in certain scenarios.

• Fixed an issue during the new database and table restoration process where adding a new read-only (RO) node failed due to a low registration offset.

• Fixed an issue in the new database and table restoration workflow where threads could not exit if the disk became full.

• Fixed an issue where concurrent data structure changes during physical replication in X-Engine caused anomalies on a read-only node.

• Fixed an upgrade failure when upgrading an X-Engine from a read replica to a standby.

• Fixed an issue where data was not repartitioned after changing the character set of a partition key in a partitioned table.

• Fixed an issue where querying a view that contains BIT-type data with IMCI returned incorrect results.

• Fixed high write latency for X-Engine tables in low-concurrency scenarios.

• Fixed an issue where GROUP BY could not find the corresponding column after a single-row subquery was pulled up.

• Fixed an issue where enabling the subquery_to_derived option in MySQL Community Edition returned an incorrect result set.

• Fixed memory corruption that occurred when using the polar_abbrev_sort_key_size parameter to control sorting operations.

• Fixed an issue where an index merge plan using a unique key (UK) returned data from only the first partition. This occurred during equality queries on a partitioned table whose primary key did not include the partition key.

Description

• For clusters without a Statement Outline, outline matching is no longer affected by the character set if the outline is added after a version upgrade.

• Optimized the process of enabling global consistency (high-performance mode).

• Added support for accessing the INFORMATION_SCHEMA views of the column store from any node. By default, the system aggregates and displays information from all column store nodes.

• Optimized the database and table restoration feature, improving restoration speed and adding support for task failure rollbacks.

Description

• Fixed a potential node crash that could occur during a parallel scan on a derived table with a ROLLUP calculation.

• Fixed a node crash caused by an incorrectly formatted user-defined CONNECTION when creating an OSS external table.

• Fixed an issue where dirty page flushing on an RW node would be blocked because the node failed to interrupt its replication relationship after a physical machine failure on an RO node.

• Fixed an issue where an excessive number of tables caused table cache eviction after an HA switchover, resulting in the rollback of Autoinc values.

• Fixed a replica node crash caused by an extreme Redo log sequence. The sequence was generated by a concurrency issue between deleting a table or index in X-Engine and an occasional background SwitchMemtable operation.

• Fixed an internal connection object leak caused by abnormal execution of elastic parallel query (ePQ) when a remote connection was forcibly closed.

• Fixed an issue where executing a ROLLUP statement in parallel could return incorrect results because the GROUP BY clause contained multiple identical constant columns.

• Fixed an assertion error triggered when accessing a global secondary index (GSI) on a partitioned table by using a RANGE INDEX, if all partitions were pruned during the partition pruning phase.

• Fixed an issue where tables with a BLOB field failed to archive cold data to the ORC format.

• Fixed a parallel query restriction on a global secondary index (GSI) in MRR and other scenarios.

Description

Applied a patch from MySQL 8.0.16 to fix a deadlock caused by an oversized blob.

Description

• Added the opt_non_covering_key_cost parameter to control whether to include the initial seek cost of a secondary index.

• Added the worst_seek_records_threshold parameter to control whether to cap the maximum I/O cost of equality lookups on a non-covering index during cost estimation.

Description

• Optimized the logic for switching a read-only node to a primary node.

• Optimized the thread pool logic for handling new connections.

• The optimizer can now push down conditions to a materialized table when the subquery contains a UNION clause.

• The optimizer can now push down more eligible conditions to a materialized table based on the transitivity of equality predicates.

• The optimizer can now push down eligible predicates into a subquery.

• Optimized the cost model for equality lookups on a non-covering index.

• Optimized the table object cache and table definition cache to reduce memory consumption.

• You can now use DLM to automatically archive cold data for all RANGE-partitioned tables.

• The optimizer can now select a global secondary index on a partitioned table in queries that use partition pruning.

Description

• Fixed an issue where the cluster could fail to start when the AccessKey information for the OSS SERVER was corrupted.

• Fixed an issue where the GroupJoin operator consumed excessive memory, causing query failures in some scenarios.

• Fixed an issue that prevented queries with OR predicates from executing due to a query plan binding failure.

• Fixed an issue where the optimizer could not accurately estimate the join selectivity when joining an unsigned column with a signed column.

• Fixed an issue where terminating a query via the scheduler could cause a database restart.

• Fixed an issue where a read-only node of the X-Engine failed to report version reference information, preventing space reclamation.

• Fixed an issue where RowID rollback could fail when loading an ahead-of-time checkpoint.

• Fixed an I/O performance regression for the columnstore index in late materialization scenarios.

• Fixed a potential infinite loop during partition pruning when a partition could not be located.

Description

Fixed a potential node crash that could occur when deleting the last CCL rule for a given SQL statement type, where the rule specified only a keyword but not a schema or table.

Description

Improved performance for archiving cold data to OSS in CSV format.

Description

• Fixed a Table definition has changed error that could be triggered when the global consistency (high-performance mode) and elastic parallel query (ePQ) features were enabled at the same time.

• Fixed inaccurate scan row count statistics for SQL statements when using index condition pushdown.

• Fixed a node crash when using the phrase search mode in a full-text index.

• Fixed an issue that caused a newly connected RO node to crash after the RW node completed an undo truncate operation.

Description

• Fixed an issue where a background thread caused high CPU usage.

• Fixed an issue where a corrupted OSS Server AccessKey could prevent a cluster from starting.

Description

Added the innodb_polar_import_tablespace_strict_mode parameter. The default value is ON. When this parameter is enabled, the system returns an error when restoring a database or table if its structure contains a full-text index.

Description

• You can now use a columnstore index to query data from specific partitions of a partitioned table.

• You can now use the primary index in the InnoDB engine to accelerate columnstore index queries.

• The system now converts an IN subquery to a Semi-Join subquery for UPDATE or DELETE operations on a single table.

• The system now converts a scalar subquery to a materialized table in a join query.

• The Fast Query Cache feature now supports partitioned tables in the InnoDB engine.

• The system no longer automatically creates an implicit auto-increment primary key when you create an OSS external table.

• You can now create a global secondary index on a partitioned table with UNIQUE CHECK IGNORE=1.

Description

• Fixed an issue where the JPPD optimization feature would crash when a CTE contained a nested CTE with multiple references.

• Fixed an issue where database and table restoration would fail for hybrid partitioned tables.

• Fixed an issue where a secondary node failed to trigger the reconnection mechanism when the primary node was shut down and restarted.

• Fixed a rare issue where a process could be interrupted again during fault recovery.

• Fixed an issue where data was inserted into the wrong partition of a UNIQUE CHECK IGNORE partitioned table.

Description

Added the ALTER TABLE table_name CHANGE PARTITION partition_name ENGINE=CSV/ORC FORCE; syntax.

Description

• Enabled log transmission between read-write (RW) and read-only (RO) nodes over an RDMA network.

• You can now archive partitioned tables and standard tables with auto-increment columns to OSS.

• The federated query engine can now push down the ORDER syntax to a remote database, which can be combined with LIMIT OFFSET to reduce network overhead.

• You can now create a columnstore index on a hybrid partitioned table (InnoDB+ORC).

• Optimized the thread pool for short-lived connections.

• The DLM feature now supports archiving InnoDB partitions from partitioned tables to OSS.

• You can now delete and add InnoDB partitions in a hybrid partitioned table.

Description

• Fixed an issue where queries using a columnstore index were not interrupted and temporary files were not cleaned up when the temporary files filled the disk.

• Fixed an issue where parameters were ignored during repeated executions of a stored procedure.

• Fixed an issue where a regular user could delete sys.

• Fixed a faulty internal counter in the thread pool that could cause a backlog of SQL requests during concurrent recording of the slow query log.

• Fixed an unexpected node restart that occurred during a database and table restoration on a partitioned table after an HA switchover.

• Fixed an issue where the checkpoint LSN did not advance as expected.

• Fixed an issue where queries on a read-only (RO) node could fail while an ALTER TABLE ... IMPORT TABLESPACE operation was running on the read-write (RW) node.

Description

• Added the ALTER TABLE table_name CHANGE PARTITION part_name ENGINE = 'csv'; command to archive partitions.

• Introduced the row-level compression feature.

• Introduced the primary key data slicing function call dbms_dts.show_parallel_ranges (<schema_name>, <table_name>, [<suggestion_ranges>]).

Description

• You can now create a columnstore index on a virtual column.

• You can now use the recycle bin feature for tables with a columnstore index.

• Optimized OSS query performance. When elastic parallel query is enabled, queries run in parallel if the value of the loose_csv_max_oss_threads parameter is greater than 1.

• Improved the error messages for incorrect file formats of OSS external tables.

• You can now read data in ORC format from OSS and write it to OSS.

• Improved error messages for spelling errors.

Description

• Fixed a replication delay that occurred when replaying redo logs of a wide table on an RO node.

• Fixed a node crash that could occur when using order index optimization to query a hybrid partitioned table.

• Fixed a node crash that could occur when truncating a hybrid partition.

• Fixed an issue where the Limit Offset pushdown optimization returned incorrect results for queries that contain a window function.

• Fixed an issue that prevented OSS from performing parallel execution.

• Fixed an issue that prevented OSS from reading JSON files.

• Fixed a spelling error in the engine.

• Fixed an issue where only the first file was read when reading cold data in CSV format from OSS.

• Fixed an issue where data containing \n was unreadable after being archived by DLM.

• Fixed an issue where temporary files remained on OSS after cold data was archived.

Description

• Added the BNL, BKA, and MRR hint types to Hint Plan.

• Added the by_expr_ignore_charset variable to SQL Sharing. Setting by_expr_ignore_charset to ON eliminates digest discrepancies for statements that contain ORDER BY or GROUP BY clauses across different character sets. However, when this parameter is set to ON, all newly generated digest values will differ from those created before setting the parameter to ON.

• The Statement Outline feature now supports the dbms_outln.add_optimizer_outline('schema', 'hints', 'query') interface for adding Optimizer Hints.

• Added the ALTER TABLE table_name CHANGE PARTITION partition_name ENGINE = 'csv'; command to archive all data from a specified partition of a partitioned table to OSS.

Description

• Optimized memory consumption when scanning tables in the recycle bin.

• Added support for querying GTID information on a read-only node.

• The columnstore index now supports instant add column by rebuilding the index data.

• The columnstore index now supports the REGEXPR function.

• The columnstore index now supports pruning optimization based on IS NULL and IS NOT NULL conditions.

• Added support for using parallel DDL to create a global secondary index.

• Added support for using the instant add column feature on a partitioned table with a global secondary index.

• Added support for converting a table with a global secondary index to an interval range partitioned table. You can also create a global secondary index on an interval range partitioned table.

• The system no longer calculates digests for SQL statements if no Statement Outline is added.

• Nonblocking DDL now supports OPTIMIZE TABLE and TRUNCATE TABLE statements.

Description

• Fixed database connection failures caused by a read timeout when establishing physical replication on a read-only node.

• Fixed an issue where condition pushdown was incorrect for tables with NULL fields when the engine_condition_pushdown parameter was set to true.

• Fixed an issue where the ALTER TABLE engine = csv storage oss, force; command failed to delete the source file on OSS.

• Fixed a database node crash when performing a check operation on an empty OSS external table.

Description

• Added support for nonblocking DDL.

• Added SQL Trace.

• Added SQL detail.

• Added Auto Plan Cache.

• Added support for columnstore indexes.

• Global consistency (high-performance mode) now includes the Innodb_scc_timeout_count parameter to track the number of query timeouts.

• Global consistency (high-performance mode) now includes the Innodb_scc_timeout_degrade_count parameter to track the number of timed-out queries that are degraded to asynchronous requests.

Description

• Statement Outline now reports a warning instead of an error if a specified index hint does not exist.

• Execution plans are now included in the extended data of the slow query log.

• Added support for adaptive digest length, which automatically adjusts the memory allocated for statement digest calculation based on query length.

• Concurrency control (CCL) now includes the ccl_mode parameter. When database traffic reaches the concurrency control limit, subsequent requests to the cluster are rejected.

• Built-in stored procedures now accept variables starting with the at sign (@) as input parameters.

• Optimized the internal lock granularity of PolarFS, significantly improving performance in some high-load scenarios.

Description

• Fixed a bug that prevented an error from being reported when adding or deleting CCL rules or Statement Outlines on a hot standby or read-only node.

• Fixed a bug that caused a cluster to crash when executing the SHOW SLAVE HOSTS statement on the primary node.

Description

Added the loose_group_by_compatible_sorting parameter to enable or disable the Group By compatible sorting mode.

Description

A Statement Outline with an empty Schema_name now matches all SQL statements with the same Digest value.

Description

Fixed a potential illegal memory access when using a DDL statement on a partitioned table.

Description

• You can now specify a value for an AUTO INCREMENT column in a partition expression to enable partition pruning based on the specified value.

• You can now use the without validation option in the add partition syntax to add a LIST partition to a LIST DEFAULT HASH partitioned table.

Description

The syntax for converting a standard table to a partitioned table now supports the without validation option: Alter table t1 partition by range(a) (partition p0 values less than (100), partition p1 values less than (200)) without validation.

Description

• Cost-based computation supports pushing the outer join condition of a materialized derived table down into the derived table.

• Supports converting a predicate with an IN list to a temporary table, performing a semi-join with the query in the FROM clause, and avoiding overhead from repeated materialization by using a zero-copy approach.

• Supports pushing down conditions from the HAVING clause to the WHERE clause.

• Subpartitions now support executing DDL operations under a partition-level MDL, which reduces lock granularity. During a DDL execution, the operation affects only the modified partition and does not affect DML operations on other partitions.

Description

• Enabling TDE disables system table encryption.

• Adjusted timeout parameters for physical replication connections to prevent interruptions during a physical machine failure.

Description

• The dbms_ccl.add_ccl_digest_rule, dbms_outln.add_index_outline, and dbms_outln.add_optimizer_outline built-in stored procedures now ignore the semicolon at the end of an SQL statement in the query parameter.

• The MEMBER keyword is now a non-reserved keyword.

• Added support for the Partial Result Cache feature.

• The optimizer's join elimination feature now supports eliminating the inner table of a left join, including inner tables of multi-level nested joins and Semi-Joins.

• The federated query engine now pushes down all compatible conditions to the remote server for execution and returns only the columns required by the query.

• The federated query engine now pushes down the LIMIT OFFSET syntax to the remote database.

Description

The CREATE SERVER syntax now supports IF NOT EXISTS.

Description

• Added the SELECT INTO OSSOUTFILE statement to export a local table (a table on PolarFS) to OSS.

• Added support for failover with hot standby.

• Added the dynamic_partition_prune_enabled parameter to control the dynamic partition pruning feature. When this feature is enabled for multi-table join queries that involve a partitioned table, it reduces unnecessary partition scans if the join condition meets the pruning criteria, which improves query performance.

Description

• Added support for creating, deleting, and rebuilding an index on specific partitions of a partitioned table.

• Added support for REBUILD PARTITION and REORGANIZE PARTITION operations with a partition-level MDL lock. This ensures that a DDL operation on one partition does not affect DML operations on other partitions.

• Usernames and passwords in the mysql.servers table are no longer stored or displayed in plaintext.

  Note

  The usernames and passwords of existing servers are not encrypted after the upgrade. You must manually delete and recreate them.

• Added support for partition-wise joins between a partitioned table and a standard table, significantly improving join efficiency for large datasets.

• Statement Outline now supports all UNION clauses.

• Added support for LIMIT OFFSET pushdown when accessing a specific partition of a partitioned table.

Description

• The parallel optimizer no longer incorrectly estimates the number of unique values for a GROUP BY field.

• SQL statements no longer write to the Binlog during a hot upgrade.

Description

• Multi-node elastic parallel query (ePQ) now supports cross-node querying of external table data on OSS.

• Adds support for using a global secondary index on a partitioned table.

• Adds support for using multi-stage multi-node parallel query in queries that use SJM (Semi-join Materialization).

Description

• Fixed an issue where a LIMIT 1 operation could terminate early, causing an ERROR 1028 (HY000): Sort aborted: Query execution was interrupted.

• Fixed an issue where writing an implicit primary key via a scheduled task could cause a node crash.

• Fixed an issue where COUNT(DISTINCT ...) returned an incorrect result set in multi-node elastic parallel query (ePQ).

• Fixed an issue where a parallel query could return an incorrect result set when the query conditions included DATE and DATETIME data types.

Description

Fixed an RO node crash caused by a file name hash collision.

Description

• Thread pool optimization: When a database transaction waits for a metadata lock or row lock, the thread pool now generates additional threads to process new requests.

• Enhanced CCL to support DDL throttling.

• Accelerated the DDL data rebuild process by asynchronously writing to a temporary file.

• Improved thread pool performance under high concurrency.

• Added the oss_sts_token parameter to the syntax for creating an OSS server to configure a temporary access credential.

Description

Fixed an issue where queries that used both Limit offset pushdown and MRR optimization returned incorrect results.

Description

• Added support for creating a hybrid partitioned table. For more information, see Create a hybrid partition.

• Read-only nodes now support custom memory temporary tables. Currently, only the MEMORY and MyISAM engines are supported.

• Introduced the ReadableProto feature. This feature allows you to directly read serialized information from a BLOB field by specifying the Protobuf protocol.

• Optimized DDL sorting by replacing the two-way merge with a multi-way merge. This change reduces I/O operations and improves performance when adding an index via DDL.

• The oss_prefix parameter is added to the syntax for creating an OSS server to configure the OSS file path.

Description

• Fixed an issue where concurrent execution results were unordered when group_by_implicit_sorting selected an index for a GROUP BY column.

• Fixed an issue where using an EXPLAIN FORMAT=TREE statement to display Hash Join results produced incorrect cost and rows values.

• Fixed excessive physical replication latency under low cluster load.

• Fixed an issue that blocked LRU List cleanup and prevented it from releasing a page. This issue occurred when the Buffer Pool had too many dirty pages and flushing was slow.

• Fixed a database crash caused by an excessively long record generated when rolling back an Update operation after an instant add column operation.

Description

Added support for creating and using federated engine tables.

Description

• Fixed a bug that caused SKIP_SCAN to return an incorrect result set when an index prefix contained multiple data groups.

• Fixed a bug that caused SKIP_SCAN to return an incorrect result set for data marked for deletion.

• Fixed a bug that could cause the fast query cache to block redo log synchronization when a read-only node acquires a metadata lock (MDL).

Description

Fixed an issue that caused incorrect results when using index for group by.

Description

• The PolarDB parallel query feature now uses a multi-node execution engine, allowing you to schedule workers to remote nodes and use computing resources from multiple nodes in the cluster. Key benefits include:

  • The degree of parallelism can now exceed the limits of a single node, accelerating large analytical queries.

  • The system schedules queries to idle nodes based on global resource load to accelerate execution and improve resource utilization.

  • You can create multiple cluster groups to form a computing resource pool, providing greater elasticity.

• The parallel query parameter parallel_query_switch now includes the restrict_on_limit_without_order switch, which controls parallel queries that have a LIMIT clause but no ORDER BY clause.

• This release adds support for condition pushdown to materialized derived tables.

Description

The SET INTERVAL syntax is added to modify the INTERVAL type and value of interval range partitions and to enable conversion between range-partitioned tables and interval range-partitioned tables. For more information, see Modify interval range partitions and Convert partitioned tables.

Description

• Added support for OSS external tables. For more information, see Access OSS data through an OSS external table.

Description

• Fixed an issue where the history list could not be deleted, which caused the undo log to bloat.

• Fixed an issue where creating a partitioned table by using CREATE TABLE .. LIKE.. LIST DEFAULT HASH failed.

• Fixed a display issue with UNIQUE CHECK IGNORE in the system table mysql.tables.

Description

• Enhanced the Elastic Parallel Query feature by improving linear acceleration and introducing a multi-stage parallel computing capability.

• Expanded support for partitioned tables. This update adds support for more table types and Interval partitioning. For more information, see Partitioned table.

• Added support for subquery decorrelation.

• In conjunction with PolarProxy, you can now use stored procedures to define sensitive fields and tag the metadata of their values.

• A STATUS variable, Online_altered_partition, is added to count the number of online ALTER TABLE ADD PARTITION and ALTER TABLE DROP PARTITION operations.

• The database kernel now supports transaction breakpoint resume and allows read-only nodes to read uncommitted transactions.

Description

• The SELECT FOR UPDATE WAIT N and SELECT FOR SHARE WAIT N syntax is supported. In this syntax, N specifies the timeout period in seconds for a single row lock. If a query needs to lock multiple rows, the time is not cumulative, and the timeout is checked for each row individually. If the wait time exceeds N seconds, a lock wait timeout error Lock wait timeout exceeded; try restarting transaction is returned.

• Fixed an issue where the optimizer did not select a more optimal index range path for some GROUP BY statements.

• Fixed an issue where large values for the group_concat_max_len parameter caused an arithmetic overflow, causing the GROUP_CONCAT function to return an incorrect result.

• Enabled modification of the database cache size.

• Improved query efficiency by avoiding unnecessary traversal of partitions in partitioned tables.

• Fixed an issue that caused DDL operations to fail on encrypted tables.

• Optimized master_key_id_mutex to enable DDL operations to run in parallel.

• Fixed a replay log crash on a read-only node under high concurrency.

• Fixed an issue where read-only nodes would crash at m_recv_bits.is_set(fold) during physical replication under high concurrency pressure.

Description

The default value of the innodb_adaptive_hash_index parameter is changed from ON to OFF.

Description

The database cache size can now be modified.

Description

Added support for MySQL 8.0.20 Index Hints.

Note

• Fixed a stability issue in parallel Hash Join.

• Fixed an incorrect row count estimation for each index key in statistics.

• Fixed an error that occurred when processing empty data during a parallel scan of a shared temporary table.

• Fixed a data processing error in the REGEXP_REPLACE function.

• Fixed a crash that occurred in complex queries that contain subquery constants.

• Fixed an issue where filesort returned incorrect data.

Description

• Optimized the cost estimation model for Hash Join.

• Added support for parallel Hash Join, including parallel Build and Probe phases.

Description

• Corrected the displayed number of parallel partitions in a parallel execution plan.

• Fixed an issue where parallel subqueries would crash.

• Fixed an issue where multiple worker threads failed to produce random results when using the RAND() function in a parallel query.

Description

PolarDB parallel processing now fully supports the latest volcano model iterative executor.

Description

Optimized cluster write performance.

Description

Fixed a memory leak.

Description

Adds the Warm Buffer Pool feature. Data in the Buffer Pool persists across crash restarts and upgrades, which significantly speeds up restarts and prevents subsequent performance degradation.

Description

• EXPLAIN supports displaying parallel plans. Use FORMAT=TREE to view the parallel execution plan tree.

• Parallel query supports hash join, enhancing JOIN performance for queries without indexes. For more information, see Parallel acceleration of Hash Join.

• Adds Resource Manager, which lets you monitor CPU and memory resources. For more information, see Resource Manager.

• Adds Performance Agent, which collects and aggregates performance data from nodes in a PolarDB for MySQL cluster. You can query the PERF_STATISTICS memory table to obtain specific performance metrics. For more information, see Performance Agent.