PolarDB:PolarDB for MySQL 8.0.2

Category

Description

Release date

Fixed issue

The following issue is fixed: The auto-increment values of a table revert to previous values when the INSERT ... ON DUPLICATE KEY UPDATE statement is executed on the table after a primary/read-only node switchover.

2024-11-21

Category

Description

Release date

Feature optimization

• The statistical information sampling mechanism is optimized, which increases the sampling frequency for large tables and ensures that the value ranges of each column are covered. This way, the execution cost can be estimated by the optimizer in a more accurate manner.

• The serverless processes of column store nodes are optimized.

  New column store nodes can handle queries only after their indexes are ready. This minimizes the risk of slow SQL queries that may occur when the queries are routed to the nodes and are executed based on row store indexes.

• In-Memory Column Indexes (IMCIs) are used in SQL queries for partitioned tables. This minimizes the overhead associated with managing partition information and improves the query performance for partitioned tables.

2024-10-21

Fixed issue

• The following issue is fixed: The result set is incorrect when nested subqueries use the IN keyword.

• The following issue is fixed: A crash occurs when a subquery uses a window function.

• The following issue is fixed: A migration task crashes because high-frequency instant DDL operations and background compaction tasks simultaneously occur.

• The following issue is fixed: The B-tree crashes when a primary key index in reverse order is created on an auto-increment column.

• The following issue is fixed: After all data in a table is cleared, the AUTO_INCREMENT value for that table may reset to 0

• The following issue is fixed: Row store indexes instead of IMCIs are used in an SQL query whose multiple subqueries use the SUM function.

• The following issue is fixed: The result of an SQL query that uses IMCIs is incorrect when a subquery uses the JSON_ARRAY expression.

• The following issue is fixed: A crash occurs when the number of tables involved in a data query and conversion process exceeds 61.

• The following issue is fixed: An error occurs when the REPLACE keyword is used to reload the same data into a partitioned table whose UNIQUE CHECK IGNORE parameter is set to 1.

• The following issue is fixed: Autoinc rollback occurs on partitioned tables after a node switchover occurs.

• The following issue is fixed: The new-version database and table restoration process is interrupted when the same table is repeatedly restored.

• The following issue is fixed: IMCIs cannot be rebuilt online, which causes the IMCIs to be unavailable.

• The following issue is fixed: An error occasionally occurs due to access to invalid data of the STRING type in queries that use the IMCI feature.

• The following issue is fixed: The performance may be degraded when the index_merge optimization strategy is used for the execution of the UPDATE statement.

• The following issue is fixed: An error occurs when the join conditions of an IMCI join contain a constant.

• The following issue is fixed: Parallel execution fails when the table function is pushed down during parallel execution.

• The following issue is fixed: An error occurs during query execution when an index specified in the outline does not exist after multi-node elastic parallel query (ePQ) is enabled.

• The following issue is fixed: Asynchronous DDL is not supported on a table on which full-text indexes are created when the COMMENT clause is modified to add an IMCI.

• The following issue is fixed: The cluster may crash during the undo initialization phase when force recovery is enabled.

• The following issue is fixed: InnoDB temporary table may cause incorrect results in update scenarios.

• The following issue is fixed: An error occurs on the result set due to the conflict between the advanced transaction splitting feature of PolarProxy and the fast query cache feature.

• The following issue is fixed: Errors may occur in the results of parallel queries when the queried database contains a constant table.

• The following issue is fixed: The JSON_LENGTH function returns an incorrect result when wildcards are used.

• The following issue is fixed: A read-only node may become unresponsive during a High Availability (HA) switchover. As a result, the switchover process has an extended duration.

Category

Description

Release date

Fixed issue

The file system version is upgraded and the process to restore a cluster from a writable snapshot is optimized.

2024-08-13

Category

Description

Release date

Fixed issue

The compatibility issue between file system versions and control systems is fixed.

2024-08-10

Category

Description

Release date

New feature and feature optimization

• The JSON_VALUE() function is added to simplify the index creation operation on JSON columns.

• Updates or changes to business are not interrupted when DDL operations are synchronized.

• Join conditions can be pushed down to corresponding views.

• Hybrid plan-based index join can be used for queries that use In-Memory Column Indexes (IMCIs), which supports InnoDB indexes in join operations.

• AI node data tables can be accessed directly from federated tables.

• The Object Storage Service (OSS) outfile feature is added to support the parallel export of table data to OSS.

• DDL operations can be prioritized to ensure successful execution.

• A new lock method is added to lock archived tables.

• The hybrid row-column optimizer and cost-based row-column routing features are added.

• The query optimizer is enhanced with the ability to eliminate unnecessary inner joins and semi-joins.

  The feature of eliminating unnecessary outer joins is optimized and can be applied to more scenarios.

• Optimized binlog writing for large transactions can be enabled to prevent other transactions from being delayed or blocked by large transactions.

• Queries that involve partitioned tables, virtual columns, or columns of ENUM, SET, or GIS data types can be executed by hybrid plans that use IMCIs.

• Queries that are incorrectly routed to row store nodes can be automatically switched to column store nodes for execution. For more information, see Adaptive execution plan switching.

• OSS data query acceleration settings can be configured during the configuration of archiving table partitions to an OSS foreign table. OSS file filter query acceleration data can be automatically generated when data is manually or automatically archived to OSS.

• PolarDB for MySQL 8.0.1 or later is optimized to display the use of outline hints in the output of the EXPLAIN statement.

• The ORDER BY and LIMIT clauses can be pushed down to join tables in PolarDB for MySQL clusters.

• The number of I/O threads in a cluster can be automatically adjusted without the need to restart the cluster.

• The time required to query a non-partitioned table is reduced.

• The replay performance of DML transactions that use IMCIs is optimized.

• The hash algorithm for bloom filters is optimized for queries that use IMCIs. String prefixes can be used to construct bloom filters, which reduces costs.

• The simulated asynchronous I/O (AIO) performance is improved.

• The performance of multi-zone clusters in low-concurrency scenarios is improved.

• The non-block DDL feature is optimized. More DDL statements can be executed without being blocked by long-running transactions or large queries.

• The memory used by metadata in queries that use IMCIs is reduced.

• The priorities of the transaction processing threads in PolarFS are optimized, which improves the performance of DDL statements.

• The performance stability of the file system when table files are frequently created and deleted during DDL is optimized.

2024-08-08

Fixed issue

• The following issue is fixed: An out-of-memory (OOM) error occurs when a query that has an excessive number of UNION ALL operations is executed by using IMCIs.

• The following issue is fixed: Access of user threads to free pages is delayed when the buffer pool of a read-only node is full.

• The following issue is fixed: Large tables are excessively sampled in data sampling.

• The following issue is fixed: If the innodb_polar_log_rdma_transfer parameter is set to ON, monitoring connections may be blocked for a period of time after you remove a read-only node.

• The following issue is fixed: IMCIs may be inefficiently used in queries that involve CASE statements and implicit data conversions.

• The following issue is fixed: Partitioned tables on which global secondary indexes (GSIs) are created cannot be archived to OSS in the CSV or ORC format. For more information, see Manually archive cold data in the CSV or ORC format.

• The following issue is fixed: The secondary node fails to send or receive Acknowledge packets due to an unrecognized connection error.

• The following issue is fixed: An OOM error occurs when the ANALYZE TABLE statement is executed on a large table.

• The following issue is fixed: Unnecessary subqueries are not removed when views are merged.

• The following issue is fixed: Inconsistent data types in an IN list cause inconsistent result sets when the IN list is converted into a join.

• The following issue is fixed: No data is returned for a single-row subquery that uses the index_merge method.

• The following issue is fixed: A subquery fails to be converted into asemi-join.

• The following issue is fixed: A crash occurs when a scalar subquery with an anti-join is converted into a derived table.

• The following issue is fixed: The precision of floating-point numbers in JSON is insufficient.

• The following issue is fixed: The RDMA-based log shipment solution for global consistency (high-performance mode) is inapplicable to secondary clusters in a GDN.

• The following issue is fixed: An overflow occurs when a large double precision floating-point number (of type double) is converted into an unsigned integer (uint) during parallel computing.

• The following issue is fixed: The cost for queries on a UCI partitioned table by using the REF or MULTI RANGE method on secondary indexes that do not include the partition key is underestimated.

• The following issue is fixed: The double value precision in JSON is lost during the parse phase.

• The following issue is fixed: The database is stuck during a manually triggered checkpoint.

• The following issue is fixed: A crash occurs when derived merging and de-nesting of scalar subqueries are used in tandem or concurrently in a query execution plan.

• The security issues of external eXtended Architecture (XA) transactions during crash recovery are fixed.

• The following issue is fixed: Statistics for a partitioned table that uses IMCIs become outdated or invalid when partition pruning is applied.

• The following issue is fixed: Cold data archiving cannot be performed on InnoDB tables that use function indexes.

• The following issue is fixed: An error rather than a warning is reported when an incorrect separator is used between the optimizer_switch and parallel_query_switch parameters in an SQL statement.

• The following issue is fixed: By default, the index merge intersection access algorithm cannot be used for the execution of DML statements.

• The following issue is fixed: The optimizer fails to select the optimal JOIN order due to cardinality estimation errors.

• The following issue is fixed: An error is reported when the CREATE TEMPORARY TABLE SELECT statement is executed to create a temporary table that uses columnar storage.

• The following issue is fixed: Assertion failures may occur during table optimization.

• The following issue is fixed: Inaccurate query results may occur in the query result cache when time and datetime are compared during a single-node or multi-node elastic parallel query (ePQ).

• The following issue is fixed: Incorrect IP addresses are recorded in the audit logs for connection reuse in a session-level connection pool.

• The following issue is fixed: The costs of point and range queries on a UCI partitioned table based on the primary key are underestimated.

Category

Description

Release date

New feature and feature optimization

• Metadata memory consumption is optimized for queries that use IMCIs.

• The IMCI optimizer's memory consumption is optimized in high-concurrency scenarios.

2024-06-21

Fixed issue

• The following issue is fixed: Creating a global secondary index (GSI) on an empty table using parallel DDL statements results in errors.

• The following issue is fixed: Statistics are not accurately estimated for large tables that contain BLOB columns.

• The following issue is fixed: The number of scanned rows in derived tables is not recorded in audit and slow query logs.

• The following issue is fixed: Space leakage is caused by exceptions that occur during the database shrinking process

• The following issue is fixed: When partitions of a hybrid partitioned table are stored in both the InnoDB storage engine and in Object Storage Service (OSS) as ORC files, parallel query execution on the table results in errors.

• The following issue is fixed: The result set is inaccurate when subqueries are incorrectly folded.

• The following issue is fixed: After the engine minor version is updated from 8.0.1 to 8.0.2, ADD COLUMN operations cannot be executed on partitioned tables which contain subpartitions.

• The following issue is fixed: Clusters may crash during parallel execution of queries with predicates containing DEFAULT expressions.

• The following issue is fixed: Excessive permissions are granted during the execution of the DLM policy.

• The following issue is fixed: Archival of tables fails when the tables contain descending indexes.

• The following issue is fixed: Space taken by old BLOB data cannot be reclaimed after BLOB updates.

• The following issue is fixed: When JOIN clauses use BIGINT columns with large values as conditions, cardinality estimation may be inaccurate and thus IMCIs cannot be used.

• The following issue is fixed: Partitions are not completely sampled during statistics updates for partitioned tables with IMCIs, resulting in inaccurate statistics after update.

• The following issue is fixed: Useless subqueries are not removed during view merging.

• The following issue is fixed: Parallel execution may fail after common table expressions (CTEs) acting as the right table are eliminated in a left join.

• The following issue is fixed: When the process pool is enabled, read-write transaction IDs reported by SQL Explorer and Audit may be 0.

Category

Description

Release date

New feature and feature optimization

The fast query cache feature is optimized to include information about the number of returned rows. For queries that hit the cache, the number of returned rows is displayed in the audit log.

2024-05-20

Fixed issue

• The following issue is fixed: The number of scanned rows in derived tables is not recorded in audit and slow query logs.

• The following issue is fixed: Queries that combine a LIMIT OFFSET clause, an equality condition on a secondary index, and an IN list condition on the primary key may not return data when the conditions are pushed down.

• The following issue is fixed: The comparison result of FLOAT and INT data types may be incorrect.

Category

Description

Release date

New feature and feature optimization

• The limit_orderby_pushdown_mode parameter is added to specify whether to push down the LIMIT and ORDER BY operations to derived tables or each query involved in a UNION ALL operation. If the value specified in the LIMIT clause is smaller than the value of the limit_pushdown_to_union_threshold parameter, the query optimizer pushes down LIMIT and ORDER BY operations to derived tables or each query involved in a UNION ALL operation.

• Query acceleration is supported on CSV data files in Object Storage Service (OSS).

• An INSTANT DROP PARTITION operation can be performed on a partition table for which a global secondary index (GSI) is created. The operation does not cause the GSI on the partition table to become invalid.

• The read performance of read-only (RO) nodes is optimized and improved.

• The LEFT JOIN operations that can be pushed down based on the TopK results are supported on In-Memory Column Indexes (IMCIs).

• Unnecessary LEFT JOIN can be eliminated in an UPDATE statement.

• The SQL Trace feature is optimized. SQL statements and their execution plans can be correctly recorded when built-in stored procedures are called.

• Columns can be added in seconds on a table on which IMCIs are created.

• Inequity predicate inference can be used to push down more predicates.

• The usage of the parameters related to automatic request distribution among row store and column store nodes is simplified. The SET use_imci_engine=FORCED statement can be used to identify the cause when column-oriented storage is not used.

• The query transformation feature is added to convert LEFT JOIN to INNER JOIN based on HAVING conditions and VIEW outer conditions.

• The query optimizer can eliminate redundant DISTINCT operators in query statements to improve query execution speed.

• The subquery execution process is optimized. Multiple subqueries can be folded.

• Temporary tables can be used when the Hybrid Plan feature is used to accelerate wide table queries.

• The speed at which the thread pool processes connection requests in low-concurrency scenarios is improved.

• The rman.import command can be used for IMCIs.

• Sampling statistics can be used to estimate approximate compression ratios for clusters that use X-Engine.

• Tair read-only (RO) nodes can be created to provide cache-based read capabilities.

• The impact of access by system O&M personnel on the slow_queries metric is ignored in database performance monitoring.

• The metadata overhead associated with the allocation and statistics collection of memory for the IMCIs is reduced, which saves memory.

• Parallel queries on X-Engine common tables and partitioned tables are supported.

• The BatchProbe optimization method can be used for IMCIs to improve the performance of JOIN operations.

• The metadata, file modules, and small objects in the resident memory of the IMCIs are optimized to decrease the usage of resident memory.

• The IMCI and global consistency (high-performance mode) features can be used at the same time.

• Histogram collection tasks can be automatically distributed by Stats Manager to read-only nodes or hot standby nodes for execution to implement automatic histogram updates.

• The I/O performance of Polar File System (PolarFS) read-only nodes is optimized to improve the query response speed of read-only nodes.

2024-05-08

Fixed issue

• The following issue is fixed: The buffer pool automatically cancels the preemptive scale-in operation because the pages occupied by large SQL statements are not released. As a result, subsequent scale-out operations cannot be performed.

• The following issue is fixed: Cold data in the CSV and ORC formats cannot be deleted by using dbms_oss.delete_table_file built-in stored procedures.

• The following issue is fixed: Nodes may crash when the buffer pool automatically cancels the preemptive scale-in operation under high concurrency conditions.

• The following issue is fixed: A query may not be distributed to IMCIs for execution when the query contains a subquery that returns an empty result set.

• The following issue is fixed: If an access control list (ACL) lock wait is encountered during the asynchronous authentication of a thread pool, the internal counters of the thread pool may show incorrect results.

• The following issue is fixed: When a multiplication operation is performed on data of the DECIMAL type, the assert information is returned when at least one multiplier is 0 and the scale (number of decimal places) of the result exceeds 16.

• The following issue is fixed: During the upgrade of an ApsaraDB RDS for MySQL instance to a PolarDB for MySQL cluster, the migration links are interrupted if the PolarDB for MySQL cluster is restarted.

• The following issue is fixed: When you query a hybrid partitioned table that contains InnoDB partitions and ORC partitions, the node on which the table resides crashes.

• The following issue is fixed: Data loss occurs when OSS foreign tables are repeatedly archived.

• The following issue is fixed: If the system accesses invalid BLOB references when prefetching BLOB-type data, node crashes occur.

• The following issue is fixed: Nodes occasionally crash when the IMCI and prefetching features are enabled at the same time.

• The following issue is fixed: Hints cannot be used to push down join conditions.

• The following issue is fixed: Hints cannot be used to push down predicates to subqueries.

• The following issue is fixed: The HAVING condition cannot be fully pushed closer to the data source.

• The following issue is fixed: Nodes crashes may occur in a parallel query when the HAVING clause uses projected columns as filters and non-correlated subqueries are executed.

• The following issue is fixed: Excessive amounts of log entries about the initialization of auto-increment columns are generated.

• The following issue is fixed: Optimizer hints cannot be used for subquery merging.

• The following issue is fixed: An error may occur when you perform a sort in descending order on an OSS foreign table.

• The following issue is fixed: A memory overflow may occur when large string packages in IMCIs are compressed.

• The following issue is fixed: Query performance may degrade when query conditions are pushed down to derived tables.

• The following issue is fixed: BIT fields are incorrectly displayed after the GROUP BY operation is performed.

• The following issue is fixed: Stringent concurrency control on background compaction tasks in X-Engine adversely impacts foreground operations, which leads to a degradation in the performance of slow queries.

• The following issue is fixed: Incorrect results may be returned in a parallel query when partition-wise parallel scans across multiple tables and a parallel hash join operation are performed within the same plan slice.

• The following issue is fixed: The column in group statement is ambiguous message appears when some subqueries are pulled up.

Category

Description

Release date

New feature and feature optimization

• The error messages relating to the following case are optimized: Cold data is transferred from OSS back to a database that uses InnoDB.

• The SQL Trace feature can be used to track the execution information of slow queries.

2024-04-17

Fixed issue

• The following issue is fixed: Memory leaks occur when features such as statement outline and concurrency control (CCL) perform Reload operations.

• The following issue is fixed: The result is incorrect when a LEFT JOIN operation is performed on a table in the INFORMATION_SCHEMA database.

• The following issue is fixed: A DLM policy remains in effect after the schema of the table for which the policy is created is changed.

• The following issue is fixed: The Can't find record on <table> error message appears when the INSERT INTO ON DUPLICATE VALUE statement is executed on a partitioned table.

• The following issue is fixed: Crashes may occasionally occur when the TRUNCATE PARTITION statement is executed on a partitioned table that contains an auto-increment column.

• The following issue is fixed: The number of parallel shards may be excessively small or the data of concurrent workers may be unbalanced when parallel queries are performed on a large table whose B-tree index is high or has data skew.

Category

Description

Release date

New feature and feature optimization

• The global dynamic permissions to execute the SHOW_ROUTINE, FLUSH_USER_RESOURCES, FLUSH_TABLES, FLUSH_STATUS, and FLUSH_OPTIMIZER_COSTS statements can be granted to users.

• Small partial updates of BLOB fields are not allowed. This prevents incorrect reading of BLOB data.

• When the SET_VAR hint is used to specify a value for the loose_optimizer_switch parameter, a comma (,) is not required at the end.

2024-04-07

Fixed issue

• The following issue is fixed: Indexes are corrupted due to different index orders. As a result, table data cannot be accessed when the EXCHANGE PARTITION statement is executed.

• The following issue is fixed: A deadlock exists due to DDL read-aheads in parallel DDL scenarios.

• The following issue is fixed: When a checkpointing operation is performed, the process of recovery from a failure cannot run as expected after the cluster is unexpectedly restarted due to the unexpected log sequence number (LSN) of the checkpoint.

• The following issue is fixed: Nodes must be restarted when auto-increment columns experience a rollback.

• The following issue is fixed: When an SQL statement that contains an ORDER BY clause is executed based on a hybrid execution plan and the output contains NULL values, the table definition has changed error message appears.

• The following issue is fixed: When an incorrect number of variables is entered for some built-in stored procedures, errors are not correctly reported.

• The following issue is fixed: The returned result set contains duplicate groups when the GROUP BY operation is pushed down to the workers for parallel execution in a database that has partitioned tables.

• The following issue is fixed: When a table that contains columns of the DECIMAL type is created, 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 message is mistakenly reported.

• The following issue is fixed: Crashes may occasionally occur if you execute undo cleanup when flashback query is enabled and temporary tables are used.

• The following issue is fixed: The query optimizer selects the ordering index but data sorting is not omitted during actual execution.

• The following issue is fixed: Remaining table columns are not updated after a JOIN operation is eliminated. As a result, the table cannot be accessed by using the optimal method.

• The following issue is fixed: If materialization is specified as the semi-join strategy, the query results may be incorrect.

• The following issue is fixed: If the binlog_row_image parameter is set to MINIMAL for a partitioned table whose primary key does not include a partition key, binary logs may not have sufficient information to uniquely identify individual row changes.

• The following issue is fixed: When the primary key does not include the partition key, unique constraint indexes are not unique. The following items describe the scenarios in which such an issue occurs:

  • Scenarios in which semi-join operations are pulled up to inner tables.

  • Scenarios in which GROUP BY or DISTINCT elimination is performed.

  • Scenarios in which unique key dependencies are added for JOIN elimination.

  • Scenarios in which correlated scalar subqueries are decorrelated by transforming subqueries from using window functions into using a derived table.

  • Scenarios in which GroupJoin queries are performed based on IMCIs.

Category

Description

Release date

New feature and feature optimization

• The speed at which the thread pool processes connection requests in low-concurrency scenarios is enhanced.

• If the current SQL statement that is executed does not require the global consistency (high-performance mode) feature after the feature is enabled, hints can be used in the statement to disable the feature.

• Correlated subqueries can be used and complex correlated subqueries can be eliminated in the GROUPING clause.

• The concurrency control and statement outline features can be enabled for a cluster of Multi-master Cluster (Database/Table) Edition. If the features are added to or removed from a primary node of the cluster, the addition or removal operation is automatically synchronized to other primary nodes of the cluster.

• Tair cache nodes can be added to a PolarDB for MySQL cluster to read data from the PolarDB for MySQL cluster.

• The process for enabling the IMCI feature is optimized to reduce the time required before the feature can be used.

• The read performance of read-only nodes in scenarios where secondary indexes return the primary key of a table is improved.

• The process for calculating transaction mask in IMCIs is optimized.

2024-01-17

Fixed issue

• The following issue is fixed: When you upgrade a cluster, a concurrency control (CCL) rule that matches the DIGEST value may become invalid.

• The following issue is fixed: Nodes crash if backticks are included in the input parameters for calling built-in stored procedures.

• The following issue is fixed: When the IMCI feature is used to query vectors, invalid sorting results are returned for descending sorting due to excessive NULL values in tables specified by the query statement.

• The following issue is fixed: When read-only column store nodes start during the execution of DDL statements on tables for which IMCIs are created, column-oriented data failed to be restored.

• The following issue is fixed: In some special scenarios, the elastic parallel query (ePQ) feature cannot be used in query statements that contain Recursive CTEs.

• The following issue is fixed: When a database or table restoration task is initiated, the read-only node fails to be added because the redo log on the primary node is cleared.

• The following issue is fixed: When a database or table restoration task is initiated, the thread fails to exit due to exhausted disk space.

• The following issue is fixed: When data structures are concurrently modified during the physical replication in X-Engine, read-only nodes are abnormal.

• The following issue is fixed: A node in a cluster of X-Engine Edition cannot be upgraded from a read-only node to a synchronous secondary node.

• The following issue is fixed: When the character set of the partition key in a partitioned table is modified, data is not repartitioned.

• The following issue is fixed: When the IMCI feature is used to query a view that contains data of the BIT type, invalid query results are returned.

• The following issue is fixed: A high latency occurs when data is written to tables in X-Engine in low concurrency scenarios.

• The following issue is fixed: After you pull up a single-row subquery, the GROUP BY clause cannot identify the column involved in the subquery.

• The following issue is fixed: After the subquery_to_derived feature is enabled in MySQL Community Edition, invalid results are returned.

• The following issue is fixed: Memory corruption occurs when the polar_abbrev_sort_key_size parameter is used to control sort operations.

• The following issue is fixed: When an equivalent query is performed on a partitioned table whose primary key does not contain the partition key and the unique key (UK) is used for an index merge operation, only the data of the first partition is returned.

Category

Description

Release date

New feature and feature optimization

• For a cluster for which the statement outline feature is disabled, if the feature is enabled for the cluster after an upgrade, the cluster is not affected by character sets when it is matched against outlines.

• The process for enabling the global consistency (high-performance mode) feature is optimized.

• The INFORMATION_SCHEMA views of all read-only column store nodes can be accessed. By default, information about the nodes is aggregated for display.

• The database and table restoration feature is optimized. For example, databases and tables can be restored more quickly, and failed tasks can be rolled back.

2023-12-20

Fixed issue

• The following issue is fixed: If a ROLLUP calculation is performed on a derived table, nodes crash due to the parallel scanning of the table.

• The following issue is fixed: When you create a table in OSS, nodes crash if you specify an invalid CONNECTION format.

• The following issue is fixed: When a physical machine hosting a read-only node fails, the primary node cannot flush dirty pages in an efficient manner due to the continuity of its replication relationship with the read-only node in special cases.

• The following issue is fixed: After a high-availability (HA) switchover between the primary and read-only nodes is performed, Auto-increment is rolled back due to table cache evictions caused by the excessive number of tables.

• The following issue is fixed: When a table or an index in X-Engine is deleted, an extreme redo log time series is generated due to the concurrency with an occasional SwitchMemtable operation that is performed on the background, which results in the breakdown of read-only nodes. and therefore the replica node crashes.

• The following issue is fixed: When a remote connection is forcibly closed, the internal connection object is leaked due to the abnormal execution of an ePQ.

• The following issue is fixed: When ROLLUP statements are executed in parallel, invalid query results are returned due to the presence of the same constant column in the GROUP BY clause.

• The following issue is fixed: ASSERT CRASH (assertion error) is triggered when all partitions of a partitioned table are pruned during partition pruning and RANGE INDEX is used to access global secondary indexes (GSIs) in the table.

• The following issue is fixed: When cold data is archived in the ORC format, the archiving operation fails due to the presence of the fields of the BLOB type in the table.

• The following issue is fixed: Limits are imposed on GSIs in parallel queries in scenarios such as Multi-Range Read (MRR) operations.

Category

Description

Release date

Fixed issue

To fix deadlock issues caused by a large amount of BLOB data, an official patch from MySQL 8.0.16 is introduced.

2023-11-13

Category

Description

Release date

New feature and feature optimization

• The logic for switching read-only nodes to primary nodes is optimized.

• The logic of the thread pool in processing new connections is optimized.

• When a subquery is used as a UNION clause, the conditions can be pushed down to a derived table.

• More equivalent conditions can be pushed down to a derived table.

• Predicates that meet a specified condition can be pushed down to subqueries by the optimizer.

• The cost model of accessing non-covered indexes for equality condition queries is optimized.

• The opt_non_covering_key_cost parameter is added to specify whether to consider the initial positioning cost of secondary indexes.

• The worst_seek_records_threshold parameter is added to specify whether to limit the maximum I/O cost of accessing non-covered indexes for equality condition queries during cost estimation.

• The amount of memory consumed by the table object cache and the table definition cache is reduced.

• Data lifecycle management (DLM) can be used to automatically archive the cold data of all tables partitioned by range.

• The optimizer can select global secondary indexes of the partitioned table during a query that involves partition pruning.

2023-10-30

Fixed issue

• The following issue is fixed: When the AccessKey pair of the Object Storage Server (OSS) server is corrupted, clusters cannot be started.

• The following issue is fixed: In some scenarios, queries fail due to excessive memory usage of the GroupJoin operator.

• The following issue is fixed: Some queries that contain OR predicates fail to be executed because the query plan cannot be bound to the correct columns.

• The following issue is fixed: When a join operation is performed on the unsigned column and a signed column, the optimizer cannot accurately estimate the join selection rate.

• The following issue is fixed: When a query is terminated by the scheduler, a database may be restarted.

• The following issue is fixed: Read-only nodes that access the X-Engine cannot report version reference information, resulting in space reclamation failures.

• The following issue is fixed: When a check point is loaded, the row ID fails to be rolled back in advance.

• The following issue is fixed: The I/O performance of In-memory Column Indexes (IMCIs) is degraded in deferred materialization scenarios.

• The following issue is fixed: If the optimizer fails to locate the appropriate partitions during partition pruning, an infinite loop may occur.

Category

Description

Release date

Fixed issue

The following issue is fixed: Nodes may crash if only keywords are specified but not the table schema and table name in the last rule when multiple CCL rules are configured to delete the same type of SQL statements.

2023-09-17

Category

Description

Release date

New feature and feature optimization

The speed of archiving cold data in the CSV format to OSS is improved.

2023-08-23

Fixed issue

• The following issue is fixed: The Table definition has changed error is reported when the global consistency (high-performance mode) and multi-node elastic parallel query (ePQ) features are enabled.

• The following issue is fixed: The number of scanned rows in an SQL statement may be inaccurate in scenarios where index condition pushdown is involved.

• The following issue is fixed: Nodes crash when the phrase search mode is used for a full-text index.

• The following issue is fixed: After the UNDO TRUNCATE statement is executed on the read/write node, if you use a read-only node to connect to the database and retrieve data, the read-only node may crash.

Category

Description

Release date

Fixed issue

• The following issue is fixed: Background threads consume a large number of CPU resources.

• The following issue is fixed: When the AccessKey pair of the OSS server is corrupted, clusters cannot be started.

2023-07-25

Category

Description

Release date

New feature and feature optimization

• An IMCI can be used to query data in a partition of a partitioned table.

• The primary index of the InnoDB engine can be used to accelerate queries in IMCIs.

• An IN subquery can be converted to a semi-join subquery when UPDATE or DELETE is performed on a table.

• Scalar subqueries can be converted to derived tables in JOIN queries.

• The Fast Query Cache feature is supported for partitioned tables that run the InnoDB engine.

• An implicit auto-increment primary key is not automatically created when an OSS foreign table is being created.

• A global secondary index can be created on a partitioned table whose UNIQUE CHECK IGNORE option is set to 1.

• The innodb_polar_import_tablespace_strict_mode parameter is added. The default value is ON, which specifies that an error is returned if the table schema of a database or table that you want to restore contains full-text index.

2023-07-23

Fixed issue

• The following issue is fixed: The JPPD optimization feature fails when a common table expression (CTE) contains a nested CTE and has multiple references.

• The following issue is fixed: When you restore a database or table for a hybrid partitioned table, the restore operation fails.

• The following issue is fixed: When the primary node is disabled and restarted, the secondary node does not trigger the reconnection mechanism.

• The following issue is fixed: The fault recovery is interrupted at a low probability.

• The following issue is fixed: Data is inserted into the wrong partition of tables that include the UNIQUE CHECK IGNORE attribute.

Category

Description

Release date

New feature and feature optimization

• Log transmission over the RDMA network is supported between the primary node and read-only nodes.

• Partitioned tables and non-partitioned tables that contain auto-increment columns can be archived to OSS.

• The federated query engine can push down the Order clause to the remote database for execution. Use the Order clause together with the Limit Offset clause to reduce network costs.

• IMCIs can be created on a hybrid partitioned table that uses InnoDB and ORC.

• The ALTER TABLE table_name CHANGE PARTITION partition_name ENGINE=CSV/ORC FORCE; syntax is added.

• The performance of the thread pool in processing short-lived connections is improved.

• The DLM feature can be used to archive the InnoDB partitions of a partitioned table to OSS.

• You can add and remove InnoDB partitions in hybrid partitioned tables.

2023-06-28

Fixed issue

• The following issue is fixed: When the disk is filled by temporary files generated by an IMCI query, the system cannot interrupt the query and clear the files in time.

• The following issue is fixed: Parameters do not take effect when the stored procedure is repeatedly executed.

• The following issue is fixed: Standard accounts can be used to delete sys.

• The following issue is fixed: When slow query logs are concurrently recorded, the failure of internal counter in the thread pool may cause accumulated SQL requests.

• The following issue is fixed: A node restarts unexpectedly when you restore a database or table for a partitioned table after high availability switchover is enabled.

• The following issue is fixed: Checkpoint LSNs are not pushed down as expected.

• The following issue is fixed: An error may occur when you execute the ALTER TABLE ... IMPORT TABLESPACE statement on a read/write node and perform a query operation on a read-only node at the same time.

Category

Description

Release date

New feature and feature optimization

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

• You can create IMCIs on virtual columns.

• You can use the recycle bin feature for tables with IMCIs.

• The ALTER TABLE table_name CHANGE PARITION part_name ENGINE = 'csv'; command is added to archive partitions.

• The row-level compression feature is added.

• The OSS query capability is optimized. After the elastic parallel query feature is enabled, the system performs parallel queries when the value of loose_csv_max_oss_threads is greater than 1.

• The error message that is reported when the format of an OSS foreign table file is invalid is optimized.

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

• The error message that is reported when misspelling happens.

2023-05-24

Fixed issue

• The following issue is fixed: Replication delay occurs when the redo log of a wide table is returned from a read-only node.

• The following issue is fixed: Nodes crash if you use the ORDER index to optimize a query in a hybrid partitioned table.

• The following issue is fixed: Nodes crash if you truncate hybrid partitions.

• The following issue is fixed: If you use the LIMIT OFFSET pushdown feature to optimize a query statement that contains window functions, the query result is incorrect.

• The following issue is fixed: OSS cannot make parallel execution.

• The following issue is fixed: OSS cannot read JSON files.

• The following issue is fixed: The word "engine" is misspelled.

• The following issue is fixed: When you read cold OSS data in the CSV format, you may read only the first object.

• The following issue is fixed: Data that contains \n cannot be read after being archived by using DLM.

• The following issue is fixed: After you archive cold data, some temporary files may remain on OSS.

Category

Description

Release date

New feature and feature optimization

• The memory usage is optimized when you scan tables in the recycle bin.

• You can query global transaction identifiers (GTIDs) on read-only nodes.

• You can add columns in seconds by creating IMCI data again.

• The REGEXPR function is supported in IMCIs.

• The IS NULL and IS NOT NULL conditions can be used to determine whether to use the IMCI pruner feature.

• The parallel DDL feature can be used to create GSIs in parallel.

• The Instant ADD COLUMN feature is supported on partitioned tables on which GSIs are created.

• The BNL, BKA, and MRR types are added for hint plans.

• The by_expr_ignore_charset variable is added in SQL sharing. Differences in digests generated by the ORDER BY or GROUP BY clause in different character sets are eliminated after you set the by_expr_ignore_charset parameter to ON. However, when the parameter is set to ON, all digests generated in this mode are different from the digests generated before the parameter is set to ON.

• The dbms_outln.add_optimizer_outline('schema', 'hints', 'query') API is provided in the statement outline feature to add optimizer hints.

• You can switch a table that has global secondary indexes created to an interval range partitioned table, and create global secondary indexes on an interval range partitioned table.

• The ALTER TABLE table_name CHANGE PARTITION partition_name ENGINE = 'csv'; command is added to archive all data of a specified partition in a partitioned table to OSS.

• If the statement outline is not added, the digest value of the SQL statement is not calculated when the SQL statement is executed.

• The non-blocking DDL feature supports the OPTIMIZE TABLE and TRUNCATE TABLE statements.

2023-04-21

Fixed issue

• The following issue is fixed: A read timeout error occurs and the database cannot be connected when physical replication is established on a read-only node.

• The following issue is fixed: When the engine_condition_pushdown parameter is set to true, tables that contain NULL fields are not pushed down correctly.

• The following issue is fixed: The source file cannot be deleted from OSS when the ALTER TABLE engine = csv storage oss, force; command is executed.

• The following issue is fixed: When you perform a check operation on an empty OSS foreign table, a database node crashes.

Category

Description

Release date

New feature and feature optimization

• The non-blocking DDL feature is added.

• The SQL Trace feature is added.

• The SQL detail feature is added.

• The auto plan cache feature is added.

• The IMCI feature is added.

• If the index hints specified for a statement outline do not exist, only an alert is reported instead of an error.

• Execution plans are displayed in the extended data of slow query logs.

• The Innodb_scc_timeout_count parameter is added for the global consistency (high-performance mode) feature to record the number of timeouts that occur during a query.

• The Innodb_scc_timeout_degrade_count parameter is added for the global consistency (high-performance mode) feature to record the number of events in which queries time out and are converted to asynchronous queries.

• The adaptive digest length is supported. The memory space calculated by the statement digest function is adaptive to the length of query statements.

• The ccl_mode parameter is added for the CCL feature. When the access traffic of the PolarDB for MySQL database reaches a specified concurrency control threshold, other access requests to the current cluster are denied.

• The variables that start with an at sign (@) can be used as input values in built-in stored procedures.

• The lock granularity of PolarFS is optimized to improve the performance in some high-load scenarios.

2023-03-20

Fixed issue

• The following issue is fixed: No error is reported when you add or remove a CCL rule or statement outline on a hot standby or read-only node.

• The following issue is fixed: When a SHOW SLAVE HOSTS statement is executed on the primary node, the cluster crashes.

Category

Description

Release date

New feature and feature optimization

• If the value of the Schema_name parameter is empty in an SQL statement, and the digest value of the statement matches the digest value specified by a statement outline, the statement outline takes effect.

• The loose_group_by_compatible_sorting parameter is added to specify whether to enable the GROUP BY sorting compatibility mode.

2023-02-21

Fixed issue

The following issue is fixed: When you execute DDL statements on a partitioned table, an "illegal memory access" error message is returned.

Category

Description

Release date

New feature and feature optimization

• If a partition expression contains an auto-increment column, you can specify the value of the column and trigger partition pruning based on the specified value.

• The without validation keyword can be added to the ADD PARTITION statement to create a LIST DEFAULT HASH partitioned table and add list partitions.

2023-02-20

Category

Description

Release date

New feature and feature optimization

• The without validation keyword can be added to switch a non-partitioned table to a partitioned table. Syntax: ALTER TABLE t1 partition by range(a) (partition p0 values less than (100), partition p1 values less than (200)) without validation.

• Cost-based optimization allows the outer join conditions of a materialized derived table to be pushed down inside the derived table.

• You can convert a predicate with an IN-List clause to a temporary table. This table can be semi-joined with a query in a FROM clause and uses the zero-copy method to avoid repeated materialization overheads.

• HAVING conditions can be pushed down to WHERE conditions.

• When the partition-level MDL feature is enabled, DDL statements can be executed on subpartitions and lock granularity is reduced. DDL statements only affect the involved partitions. DML statements executed on other partitions are not affected.

2023-01-09

Fixed issue

• When TDE is enabled, the encryption operation on system tables is disabled. You are not allowed to encrypt system tables.

• The timeout parameter related to the physical replication connection is modified to fix the issue of stuttering when the physical server crashes.

Category

Description

Release date

New feature and feature optimization

• A semicolon (;) at the end of an SQL statement in the query parameter of the dbms_ccl.add_ccl_digest_rule, dbms_outln.add_index_outline, or dbms_outln.add_optimizer_outline built-in stored procedure can be ignored.

• The MEMBER keyword is changed to a non-reserved keyword.

• The partial result cache feature is supported.

• The join elimination feature of the optimizer can eliminate inner tables of left joins, including inner tables with multi-layer nested joins and semi-joins.

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

• The federated query engine allows you to push down the Limit or Offset clause to the remote database for execution.

2022-12-23

Fixed issue

The following issue is fixed: IF NOT EXISTS cannot be included in the statement that is used to create servers.

Category

Description

Release date

New feature and feature optimization

• Indexes can be created on some partitions of a partitioned table. You can create, delete, and rebuild indexes on the partitions of a partitioned table.

• The REBUILD PARTITION and REORGANIZE PARTITION operations can be performed after partition-level MDL locks are created. This ensures that DDL operations performed on a partition do not affect DML operations on other partitions.

• The usernames and passwords in the mysql.servers table are no longer stored and displayed in plaintext.

  Note

  The existing username and password of a server are not encrypted after the upgrade. You must manually delete the existing username and password and then recreate a username and password for the server.

• The SELECT INTO OSSOUTFILE statement is added to export local tables on PolarFS to OSS.

• The failover with hot standby feature is supported. For more information, see Failover with hot replica.

• The dynamic_partition_prune_enabled parameter is added to control partition pruning for dynamic partitions. After partition pruning is enabled in various scenarios in which a join query is performed on multiple partitioned tables, the system does not scan unnecessary partitions if the connection between partitions meets specified partition pruning conditions. This improves the query performance.

• The partition-wise join feature can be used between partitioned tables and non-partitioned tables. This improves the efficiency of joins between partitioned tables and non-partitioned tables in queries that involve large volumes of data.

• All UNION clauses are supported for statement outline.

• The LIMIT OFFSET pushdown feature can be used when you access an independent partition in a partitioned table.

2022-11-30

Fixed issue

• The following issue is fixed: The number of unique values of the GROUP BY field in the parallel optimizer is inaccurately estimated.

• The following issue is fixed: Binary logs are generated when SQL statements are executed during hot upgrade.

Category

Description

Release date

New feature and feature optimization

• The ePQ feature supports cross-node parallel execution for data in OSS foreign tables.

• Global secondary indexes are supported on partitioned tables.

• Multi-stage ePQ is supported in queries that contain semi-join materialization (SJM) strategies.

2022-10-25

Fixed issue

• The following issue is fixed: The ERROR 1028 (HY000): Sort aborted: Query execution was interrupted error is reported when a query execution is halted prematurely due to the use of the LIMIT 1 clause.

• The following issue is fixed: Nodes do not respond when data is inserted into tables with implicit primary keys by using the Event Scheduler.

• The following issue is fixed: Incorrect result sets of count(distinct ...) are returned for ePQ.

• The following issue is fixed: If query conditions contain the DATE and DATETIME types, the result set of a parallel query may be incorrect.

Category

Description

Release date

Fixed issue

The following issue is fixed: Read-only nodes do not respond when the hash values of file names conflict.

2022-09-17

Category

Description

Release date

New feature and feature optimization

• Thread pool optimization: When database transactions wait for metadata locks or row locks, additional threads can be generated in the thread pool to process new requests.

• CCL is enhanced to support DDL throttling.

• Temporary files can be asynchronously written to accelerate the process of reconstructing data in DDL statements.

• The performance of the thread pool in high-concurrency environments is enhanced.

• The oss_sts_token parameter is added to the statement for creating an OSS server to configure temporary credentials for OSS.

2022-08-19

Fixed issue

The following issue is fixed: Errors are returned when both LIMIT OFFSET pushdown and MRR optimization are enabled.

Category

Description

Release date

New feature and feature optimization

• Hybrid partitioned tables can be created. For more information, see Create a hybrid partitioned table.

• In-memory temporary tables can be customized on read-only nodes. Only the MEMORY and MyISAM engines are supported.

• The ReadableProto feature is supported. The Protobuf protocol is used to read the information serialized and stored in the BLOB-typed field on the database.

• Optimized DDL sorting: Two-way merging is changed to multiple-way merging to reduce I/O operations and improve the performance of index creation by using DDL statements.

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

2022-07-22

Fixed issue

• The following issue is fixed: group_by_implicit_sorting concurrent execution results are unordered when you select indexes for group by columns.

• The following issue is fixed: cost and rows errors occur when you execute the EXPLAIN FORMAT=TREE statement to view the Hash Join results.

• The issue that causes high latency when physical replication is performed during a time period of low workloads in the cluster is fixed.

• The following issue is fixed: When a large number of dirty pages exist in the buffer pool and page buffer pool flushing is slow, the least used page cannot be evicted.

• The following issue is fixed: A database fails to respond due to an excessive number of log records that are generated after you use the Instant ADD COLUMN feature and then roll back an update operation.

Category

Description

Release date

New feature and feature optimization

The FEDERATED storage engine is supported. You can create and use FEDERATED tables.

2022-05-31

Fixed issue

• The following issue is fixed: When a skip scan is performed on an index that is prefixed with multiple sets of data, the results of the skip scan may be incorrect.

• The following issue is fixed: When a skip scan is performed and specific data in the database is marked as deleted, the results of the skip scan may be incorrect.

• The following issue is fixed: The fast query cache feature blocks the synchronization of redo logs when the read-only node obtains the metadata lock (MDL).

Category

Description

Release date

Fixed issue

The issue that causes incorrect results when an index is used for GROUP BY operations is fixed.

2022-04-19

Category

Description

Release date

New feature and feature optimization

• The parallel query feature supports parallel SQL execution on multiple nodes. In a single query, some or all of the workers can be allocated to and executed on remote nodes to effectively use the computing resources of multiple nodes in the cluster. The following descriptions provide details:

  • The parallel query feature can overcome the performance bottleneck of a single node and accelerate analytical queries such as big queries.

  • Queries are allocated to idle nodes to accelerate executions and improve resource usage.

  • You can create a computing resource pool that contains multiple cluster groups to achieve high elasticity.

• The restrict_on_limit_without_order switch that is used to control the limit parallel query without the ORDER BY clause is added to the parallel_query_switch parallel query parameter.

• Conditions can be pushed down to materialized derived tables.

2022-03-31

Category

Description

Release date

New feature and feature optimization

• The SET INTERVAL statement is added to modify the INTERVAL data type and value of an interval range partitioned table and implement conversion between range partitioned tables and interval range partitioned tables. For more information, see Modify an interval range partition and Switch an interval range partitioned table.

• OSS foreign tables are supported. For more information, see Use OSS foreign tables to access OSS data.

2022-03-15

Fixed issue

• The following issue is fixed: A large number of undo logs are accumulated because the history list cannot be removed.

• The following issue is fixed: A partitioned table cannot be created by executing the CREATE TABLE .. LIKE.. LIST DEFAULT HASH statement.

• The following issue is fixed: A partitioned table that includes the UNIQUE CHECK IGNORE attribute cannot be displayed in the system table mysql.tables.

Category

Description

Release date

New feature and feature optimization

• The parallel query capability is improved. The linear acceleration capability is significantly optimized and the multi-phase parallel computing capability is introduced. For more information, see Overview.

• The capabilities for partitioned tables are enhanced. More partitioned table types and interval partitioning are supported. For more information, see Partitioned tables.

• Subquery decorrelation is supported.

• When PolarProxy is connected, you can use the stored procedure to define sensitive fields and tag the metadata of sensitive field values.

• The status variable Online_altered_partition is added to count the number of online ALTER TABLE ADD PARTITION and ALTER TABLE DROP PARTITION operations.

• Transaction resumable uploads are supported for the database kernel. Uncommitted transactions can be read by read-only nodes for the database kernel.

2022-01-12

Fixed issue

• The SELECT FOR UPDATE WAIT N and SELECT FOR SHARE WAIT N statements are supported. N specifies the number of seconds during which a row remains locked. The timeout applies only to a single row. To lock multiple rows for a query, the system detects the timeout only based on a single row instead of multiple rows. If the lockout time exceeds the specified N value, the system returns a similar lockout timeout error: Lock wait timeout exceeded; try restarting transaction.

• The following issue is fixed: The optimizer fails to choose a better path to the range of an index for some GROUP BY statements.

• The following issue is fixed: When a large value is specified for group_concat_max_len, an arithmetic overflow occurs, which causes an incorrect result of the GROUP_CONCAT function.

• The size of database cache can be modified.

• This prevents unnecessary traversal of partitioned table partitions and improve query efficiency.

• The issue of failed DDL operations on encrypted tables is fixed.

• master_key_id_mutex is optimized to ensure that DDL operations can be executed in parallel.

• The following issue is fixed: A read-only node fails to respond when the node replays logs in high concurrency scenarios.

• The following issue is fixed: A read-only node stops responding due to m_recv_bits.is_set(fold) when physical replication is being performed on the node in high concurrency scenarios.

Category

Description

Release date

New feature and feature optimization

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

2021-10-26

Fixed issue

The size of database cache can be modified.

Category

Description

Release date

New feature and feature optimization

The Index Hints feature of MySQL 8.0.20 is supported.

2020-12-01

Fixed issue

• The stability issue of parallel hash joins is fixed.

• The following issue is fixed: The estimated number of rows for each index key in statistics is miscalculated.

• The following issue is fixed: An error occurs in processing empty data when shared temporary tables are scanned in parallel.

• The following issue is fixed: An error occurs when you use the REGEXP_REPLACE function provided by PolarDB to process data.

• The following issue is fixed: An unexpected exit occurs when a complex query contains subquery constants.

• The following issue is fixed: A filesort operation returns invalid data.

Category

Description

Release date

New feature and feature optimization

• The cost estimation model for hash joins is improved.

• Parallel hash joins are supported. The build phase and the probe phase can be executed in parallel.

2020-10-26

Fixed issue

• The following issue is fixed: The displayed number of parallel partitions in a parallel execution plan is inaccurate.

• The following issue is fixed: An unexpected exit occurs when parallel subqueries are run.

• The following issue is fixed: Multiple worker threads cannot return random results when the RAND() function is used in parallel queries.

Category

Description

Release date

New feature and feature optimization

The latest volcano iterator executor for parallel queries is supported.

2020-09-27

Category

Description

Release date

New feature and feature optimization

The write performance of clusters is optimized.

2020-08-26

Fixed issue

The memory leak issue is fixed.

Category

Description

Release date

New feature and feature optimization

• The EXPLAIN statement can be executed to query the information about a parallel execution plan. After you specify FORMAT=TREE in the EXPLAIN statement, the execution plan is displayed in the tree format.

• Hash joins are supported in parallel queries. The performance of join operations that do not use indexes is improved. For more information, see Hash joins in parallel queries.

• The Resource Manager feature is added to monitor CPU and memory resources. For more information, see Resource Manager.

• The Performance Agent feature is added to collect statistics about the performance metrics of PolarDB for MySQL clusters. To collect performance metrics, you can query the PERF_STATISTICS in-memory table. For more information, see Performance Agent.

• The warm buffer pool feature is supported. When an instance restarts after an unexpected exit or an upgrade, the data in the buffer pool is not cleared. This significantly accelerates the restart process and allows you to maintain the same level of system performance after the restart.

2020-07-22