PolarDB:Release notes for PolarDB-X data nodes

Version

Category

Description

5.4.19 (MySQL 5.7-compatible)

New feature

Faster DDL is supported.

Fixed issue

The following issue is fixed: Data nodes do not run when the operating system is AliOS 6.

The following issue is fixed: After switching over to the new primary node, the data nodes cannot be accessed when the old primary node is being recovered.

Version

Category

Description

8.4.19 (MySQL 8.0-compatible)

New feature and feature optimization

Imperceptible primary node switch is supported.

Implicit primary keys are supported for WRITESET.

WRITESET-related metrics are supported for O&M and monitoring.

The default value of BINLOG_TRANSACTION_DEPENDENCY_HISTORY_SIZE is changed to 200k.

The page weight for accessing historical data in some scenarios is optimized to minimize the impact on buffer pool.

Fixed issue

The following issue is fixed: The WRITESET calculation result is incorrect when federated indexes are used.

The following issue is fixed: When you modify the value of BINLOG_TRANSACTION_DEPENDENCY_TRACKING of a data node when it is generating traffic, the system crashes.

The following issue is fixed: Rollback operations may trigger PURGE operations.

The following issue is fixed: After you switch the primary node of the three-node storage, the shared sessions may not be released.

Version

Category

Description

5.4.19 (MySQL 5.7-compatible)

New feature

The primary/secondary replication feature provided by MySQL is supported for data nodes.

Version

Category

Description

8.4.19 (MySQL 8.0-compatible)

New feature

The GUESS PRIMARY PAGE Cleanout feature is supported to improve the hit ratio of table access by index primary key.

The primary/secondary replication feature provided by MySQL is supported for data nodes.

Fixed issue

The following issue is fixed: For flashback queries, the index visibility is not as expected across DDL operations.

The following issue is fixed: The delete operations cannot be performed based on the System Change Numbers (SCNs).

The following issue is fixed: The table property FLASHBACK AREA in the data dictionary is different from that in the engine.

The following issue is fixed: The space used by redo logs is incorrectly calculated.

Version

Category

Description

8.4.19 (MySQL 8.0-compatible)

Fixed issue

The following issue is fixed: In high-concurrency scenarios, the GUESS PRIMARY PAGE mechanism may cause deadlocks.

The following issue is fixed: Data inconsistencies may occur in the backup set when tablespaces are simultaneously imported during the XtraBackup backup process.

The following issue is fixed: The system may crash if you use XPLAN to query the execution plans of native partitioned tables of MySQL that are created in the DRDS or AUTO mode and have local partitions specified.

The following issue is fixed: Audit logs are not generated for XRPC.

Version

Category

Description

8.4.19 (MySQL 8.0-compatible)

New feature and feature optimization

The dbms_consensus.force_learner_node command is supported.

In manual recovery from backups, O&M configuration is optimized to provide all required options.

The performance of a query that uses clustered indexes and involves scans, purge, and locking is optimized.

Fixed issue

The following issue is fixed: When optimizer trace is enabled, sampling causes database crashes.

The following issue is fixed: Secondary databases crash during the execution of XA transactions or the RENAME TABLE statement.

The following issue is fixed: Too many debug logs are printed in certain scenarios.

Version

Category

Description

5.4.19 (MySQL 5.7-compatible)

Fixed issue

The following issue is fixed: The gap lock on the supremum record is held during XA transactions in the PREPARE state, which causes performance impact in the secondary database.

Version

Category

Description

8.4.19 (MySQL 8.0-compatible)

New feature and feature optimization

Historical versions of a table in the recent period can be saved for user query. The length of the recent period can be specified.

The innodb_pfs_data_locks_max_locks_per_batch parameter can be used to specify the maximum number of locks returned in a query on the pfs.data_locks table.

The performance of space reclamation for the flashback area is improved.

The performance of update operations on the scn_history system table is improved.

Fixed issue

The following issue is fixed: The apply thread fails to start when the instance is restored from specific backup sets.

The following issue is fixed: An extended period of time is required to create an instance.

The following issue is fixed: The changset memory is incorrectly calculated.

The following issue is fixed: Concurrency control (CCL) causes thundering herd problems in high-concurrency conditions, which degrades the database performance.

The following issue is fixed: An instance crashes when a query is performed on the SEQUENCE table to which a write lock is applied.

The following issue is fixed: An instance fails to flush dirty pages to the disks when the instance is forcibly shut down by executing the SHUTDOWN statement during parallel DDL operations.

The following issue is fixed: During the upgrade of a MySQL database from a lower version to a PolarDB-X instance 5.7, the instance crashes when the INSTANT DDL statement is executed.

The following issue is fixed: Incorrect information is recorded in the redo logs when the ALTER STORED COLUMN ORDER and INSTALL DDL statements are executed on a table at the same time.

Version

Category

Description

5.4.19 (MySQL 5.7-compatible)

New feature

The XRPC.CMD() stored procedure is added to clear the database cache generated by the execution of SQL statements. This feature addresses the cache pollution issues introduced by compute nodes of earlier versions.

Fixed issue

The following issue is fixed: An instance may become unavailable when a PURGE operation is performed on temporary tables in the instance.

The following issue is fixed: Data inconsistencies may occur in the backup set when tablespaces are simultaneously imported during the XtraBackup backup process.

The following issue is fixed: The CPU usage is high when a non-primary key index is used to query a partitioned table.

The following issue is fixed: A directory conflict occurs when an RPM Package Manager (RPM) installation package is restored.

Version

Category

Description

5.4.19 (MySQL 5.7-compatible)

New feature

The concurrency control (CCL) feature that allows you to control the concurrency of statements based on syntax rules is supported.

Fixed issue

The following issue is fixed: An Out of Memory (OOM) kill may occur when the write traffic is high during scaling operations.

The following issue is fixed: A full table scan may mistakenly be performed in a query on a MySQL partitioned table when data is sorted by the partition key in reverse order.

The following issue is fixed: Sorting results of a query on a MySQL partitioned table may be incorrect if the query uses a composite index with the partition key as the prefix.

The following issue is fixed: When the user thread stack is exhausted, the check_stack_overrun check fails to detect the condition, resulting in a core dump.

Version

Category

Description

8.4.19 (MySQL 8.0-compatible)

New feature

Key Management Service (KMS)-based transparent data encryption (TDE) is supported.

The inventory hint feature is supported.

The IMPORT TABLESPACE IF NOT EXISTS clause is supported.

The removal of useless files during redo file creation is optimized.

The global transaction identifiers (GTIDs) that are used to track modifications on non-transactional tables are optimized.

Error log notifications are optimized.

Fixed issue

The following issue is fixed: The running status of the slave coordinator is incorrectly displayed.

The following issue is fixed: The apply thread starts running prematurely, which leads to an interruption in the replication process on the secondary database.

The following issue is fixed: The timeout duration for purging large files is excessively long.

The following issue is fixed: Dumps may be generated during member changes when the system runs in debug mode.

The following issue is fixed: Deadlocks may occur when flush and rotate operations are simultaneously performed.

The following issue is fixed: A crash may occur when the metadata_locks table in the Performance Schema (PFS) is queried.

The following issue is fixed: The binlog is empty during the full backup of the secondary database.

Version

Category

Description

5.4.19 (MySQL 5.7-compatible)

New feature and feature optimization

A new packaging method is added. The package names are in the t-polardbx-engine-57-yyyymmdd- format.

The polardbx_release_date and polardbx_engine_version system variables are added to record the current version information.

The returning feature for data backfilling is added.

Fixed issue

The following issue is fixed: The return values of file system calls are not checked.

The following issue is fixed: A crash occurs during the strong read consistency check that follows a restart of the secondary instance.

The following issue is fixed: Data nodes occasionally crash during data migration.

The following issue is fixed: The replication between the primary and secondary instances is interrupted after an insert statement applies a Next-Key lock to the subsequent record.

8.4.19 (MySQL 8.0-compatible)

New feature

The service is compatible with MySQL 8.0.32.

jemalloc 5.3.0 is statically compiled into the service.

OpenSSL 1.1.1 is statically compiled into the service.

The retention duration of a transaction slot in the Lizard transaction system can be controlled separately.

Forced check for the uniqueness of server_id is supported.

A SQL statement is added, which allows you to check whether undo logs are purged.

Fixed issue

Several defects in large transaction scenarios are fixed.

Several defects in the Sampling Scan scenario are fixed.

Category

Description

New feature

The import tablespace if not exists syntax is supported.

Fixed issue

The following issue is fixed: The read inconsistency issue occurs when a secondary database is started in abnormal scenarios.

The following issue is fixed: The changeset is closed when DROP TABLE is executed.

The following issue is fixed: The execution status of data definition language (DDL) statements in multiple phases is disordered if the names of databases and tables are long or contain non-ASCII special characters.

The following issue is fixed: Empty set data of the ENUM and SET data types is incorrectly encoded in XRPC.

The following issue is fixed: When the Transmission Control Protocol (TCP) connection is disconnected, the corresponding processes in XRPC exit occasionally and unexpectedly.

Category

Description

New feature

Timestamps are added to standalone transactions and extended architecture (XA) transactions.

Two-phase DDL operations are supported.

A defensive code is added to handle Machine Protection Exceptions (MPE) in an XRPC cache when the instance unexpectedly exits.

Fixed issue

Issues related to binlogs and synchronization protocols in large transactions are fixed.

The following issue is fixed: When a process in XRPC exits, the TCP listener is not released in a proper manner.

The following issue is fixed: The encoder in XRPC is incompatible with PolarDB-X in the protocol buffers (protobuf) of a later version.

The issue related to the compatibility of auto savepoints with binlogs is fixed.

Category

Description

Fixed issue

The following issue is fixed: Pointer leakage may exist during session attachment and detachment.

The following issue is fixed: When a separated XA transaction is recommitted or rolled back, released transactions may be accessed, resulting in an instance exception.

Category

Description

Feature optimization

The code is optimized to align with MySQL 8.0.32.

The initialization of the context environment of a request plan is optimized.

Fixed issue

The following issue is fixed: Databases in XRPC occasionally fail to be switched in scenarios that require streaming early stopping.

Category

Description

Feature optimization

The memory ordering of SpinLock SpinRWLock used by XRPC is optimized. This optimization is mainly suitable for CPUs with the weak memory ordering such as ARM processors.

Fixed issue

The sample crash issue is fixed.

The following issue is fixed: Resources such as locks fail to be released when a killed session is not scheduled.

Category

Description

New feature and feature optimization

The replication and migration feature is supported for physical tables.

The logging feature is optimized for XRPC network-related events by deleting logs concerning TCP and session disconnections.

XRPC code is optimized to align with open-source MySQL 8.0.

Fixed issue

The following issue is fixed: The names of auto savepoints are truncated.

Category

Description

Fixed issue

The following issue is fixed: The secondary database fails to be rebuilt due to the leakage of changeset resources.

The following issue is fixed: The configurations of the primary and secondary instances are inconsistent in the production environment when XRPC is enabled to replace the old proprietary protocol.

Category

Description

Fixed issue

The following issue is fixed: XPLAN unexpectedly exits when created columns are queried.

Issues related to foreign key constraints are fixed.

Category

Description

New feature and feature optimization

The useAffectedRows option can be added to the frontend connection string of the PolarDB-X proprietary protocol.

More character set encodings are available in the PolarDB-X proprietary protocol.

The metadata locks (MDLs) detection mechanism is supported for XRPC.

Fixed issue

The following issue is fixed: The number of affected rows is incorrectly recorded in the audit logs of XPLAN.

The following issue is fixed: The data nodes of the instance fail to exit.

The following issue is fixed: A scheduling latency occurs on the threads of XRPC when a large number of tasks are run.

Issues related to version display are fixed.

Category

Description

New feature and feature optimization

The CPU affinity of threads can be automatically corrected after XRPC allows the controller to dynamically modify the CPU quota of a data node.

The auto savepoint feature is supported in XRPC.

The hostname of XRPC is fixed to enhance the detection of Thread Descriptor (THD) attach and detach operations.

Fixed issue

The potential THD leak issue is fixed.

The following issue is fixed: Inaccurate numbers of scanned rows and returned rows are displayed in XPLAN audit logs and slow query logs.

The following issue is fixed: Out-of-bound memory accesses may occur when null bitmap is specified for the chunk encoder proprietary protocol.

Category

Description

New feature

XPLAN audit logs and slow query logs can be generated.

The number of XPLAN requests can be taken into account when the number of queries per second (QPS) and the number of executed SELECT statements are calculated.

Fixed issue

The issue that occurs when throttling is implemented based on the number of lines being transmitted is fixed. The throttling mechanism is changed from restricting the traffic of specific proprietary protocols to limiting the number of transmitted bytes.

The following issue is fixed: A crash occurs due to the leakage of a mark in an async commit.

Category

Description

New feature

The proprietary protocol server is rebuilt.

Fixed issue

The following issue is fixed: Inconsistency occurs between single-shard reads and single-shard writes.

The following issue is fixed: Data migration fails when the primary key is of the integer type and contains negative values.

The issue of out-of-memory in Sequence Lex is fixed.

Category

Description

New feature

The xcluster standalone feature is supported.

Category

Description

Fixed issue

The following issue is fixed: No error message is returned if an ultra-large package is sent over the PolarDB-X proprietary protocol.

The following issue is fixed: Excessive logs are generated.

Category

Description

New feature

The changeset feature is supported.

Fixed issue

The following issue is fixed: XPLAN unexpectedly exits when a table that has virtual columns is queried.

Category

Description

Fixed issue

The following issue is fixed: Backup operations fail.

The following issue is fixed: PolarDB-X unexpectedly exits due to the array index out of bounds error that is returned when a hint is executed.

Category

Description

Fixed issue

The consistent read in secondary databases for rollback is optimized to ensure that data can be restored from the backup.

Category

Description

Feature optimization

The efficiency of consistent read in secondary databases is optimized.

Fixed issue

The following issue is fixed: Incorrect filtered results are returned because incorrect data is prefetched in MySQL.

Category

Description

New feature

The function that is used to change column types is supported for Online Modify Column.

Fixed issue

The following issue is fixed: An error may be returned for flashback queries after a data node is restarted.

The following issue is fixed: An error occurs when an ALTER TABLE statement in which an ADD COLUMN clause is used and INSTANT is specified as the algorithm is executed.

Category

Description

New feature

The flashback query feature is supported.

Category

Description

Fixed issue

The following issue is fixed: The sample feature exits unexpectedly in some scenarios.

The following issue is fixed: An error occurs when m_token is asserted over the PolarDB-X proprietary protocol.

Category

Description

Fixed issue

The following issue is fixed: Read operations and write operations cannot be performed when the thd->get_protocol command is run.

The following issue is fixed: Errors occur during the execution of prepared statements that include sequences of the New Sequence type.

The following issue is fixed: Large files cannot be purged efficiently.

The following issue is fixed: Specific characters that are included in JSON-format data cannot be correctly escaped.

The following issue is fixed: The memory leak occurs in schema stats.

Category

Description

New feature

A new sequence engine is supported.

Category

Description

New feature

The port check feature is provided. When a data node starts, the system checks whether the port that is used to send requests over the PolarDB-X proprietary protocol is being used by another process. If the port is being used by another process, the current process automatically exits.

Fixed issue

The following issue is fixed: A crash occurs due to data leaks when chunks are transmitted over a proprietary protocol.

Category

Description

Fixed issue

The following issue is fixed: When a Bloom filter is used, data boundary errors occur if the remainder that is obtained by dividing the length of an element that is calculated by using the xxHash algorithm by 8 is equal to 4.

The following issue is fixed: Compatibility errors occur after the value of innodb_thread_concurrency is changed.

Category

Description

Fixed issue

Redundant rotate events are removed from binlogs.

The following issue is fixed: The previous version of MySQL plugins may be used after the storage engine is updated and the deployment script is run to deploy data nodes.

Category

Description

New feature

The xxHash algorithm is supported for Bloom filters.

Raw strings are supported, and the issue in which precision loss occurs for decimal values is fixed in PolarDB-X kernel V5.4.13 and later versions.

The purge sequence feature is provided. The system can automatically increase the value of the lower threshold for data purging to support long-term transactions.

Fixed issue

The following issue is fixed: Faster Checker returns incorrect hash values if Faster Checker does not obtain data.

The following issue is fixed: Data nodes fail when a graceful shutdown is performed for a least recently used (LRU) cache.

The following issue is fixed: Followers can obtain locks.

The following issue is fixed: The corresponding data nodes fail when a graceful shutdown is performed for the Global Timestamp Service (GTS).

The following issue is fixed: Results of de-parameterization for parameters that are represented in binary form are incorrect.

Category

Description

New feature

INSTANT ADD COLUMN is available.

The user defined function (UDF) fast hash checker component is provided.

Fixed issue

The following issue is fixed: Token errors occur when requests are sent over the PolarDB-X proprietary protocol.

The following issue is fixed: The network layer may reject requests that are sent over the PolarDB-X proprietary protocol.

Category

Description

New feature

The returning syntax is supported.

A lock-free algorithm is provided for Timestamp Oracle (TSO) to generate timestamps.

Category

Description

Fixed issue

The following issue is fixed: The corresponding data nodes crash when the system performs data sampling.

Category

Description

Feature optimization

Log content is optimized.

Fixed issue

The following issue is fixed: The virtual input and output (VIO) for sessions that are established over the PolarDB-X proprietary protocol fail.

The following issue is fixed: Table leakage occurs when a GTS lease does not exist.

Category

Description

Fixed issue

The following issue is fixed: Sessions that are established over the PolarDB-X proprietary protocol cannot be terminated.

Category

Description

Feature optimization

The process of accessing transactions that are executed on a single shard is optimized.

Fixed issue

The following issue is fixed: TSO returns incorrect results in batch processing mode. The algorithm that is used to distribute timestamps is modified.

Category

Description

Fixed issue

The following issue is fixed: When the PolarDB-X proprietary protocol is used to transmit execution plans to data nodes, the data nodes crash if operations that are not supported by the data nodes are included in execution plans. After this issue is fixed, data nodes report errors and do not crash in this scenario.

Category

Description

New feature

The auto savepoint feature is provided.

The random sampling feature provided by InnoDB is supported.

By default, the polarx_udf plugin is enabled.

The feedback feature is provided to return the statistics of indexes and the number of rows.

Fixed issue

The following issue is fixed: The transaction sequence leaks over the PolarDB-X proprietary protocol.

The following issue is fixed: When the PolarDB-X proprietary protocol is used, the length of decimal values exceeds the upper limit if specific values are specified in double precision scenarios.

The following issue is fixed: Statistics of specific physical data are missing.

The following issue is fixed: The HyperLogLog (HLL) service may exit unexpectedly.

The following issues are fixed: The batch timestamp generation feature of TSO does not take effect, and TSO may generate duplicate timestamps.

The following issue is fixed: The corresponding data nodes crash when utf8mb4 is used as a comparative parameter value in a filter expression that is executed over the PolarDB-X proprietary protocol.

Category

Description

New feature

The cache feature is provided to cache SQL statements and execution plans of SQL queries that are executed over the PolarDB-X proprietary protocol.

The chunk transmission feature is provided to transmit data chunks over the PolarDB-X proprietary protocol.

Fixed issue

The following issue is fixed: The system unexpectedly exits when TSO transactions and DDL operations are performed in parallel.

Category

Description

New feature

The SQL flashback feature is provided.

Fixed issue

The following issue is fixed: Unexpected prepare wait timeout issues occur when specific operations are performed.

The following issue is fixed: Data nodes crash in high concurrency scenarios.

Category

Description

New feature

The dump feature of mysqldump is supported. The system can back up data based on specified snapshot sequences.

Fixed issue

The following issue is fixed: When enable_gts is enabled, binlog events are replayed from the earliest offset after a follower node is restarted.

The following issue is fixed: The leader node of an instance cannot be restored after the leader node is restarted.

By default, the lease period of the GTS is set to 2 seconds.

Commit timestamps (CTSs) can be included in messages that are transmitted over the PolarDB-X proprietary protocol, and the cache feature is provided to cache SQL statements and execution plans of SQL queries.

Category

Description

New feature

TSO timestamps can be transmitted by using messages.

Common transactions can be committed in one-phase commit mode.

Fixed issue

The following issue is fixed: Data nodes fail when the data types of parameters that are returned by UDFs of Bloom filters are incorrect.

The following issue is fixed: Null values and decimal values are not correctly processed by UDFs of Bloom filters.

The following issue is fixed: When CTS transactions are performed based on indexes, the isolation levels of the transactions are changed.

The issue in which an error occurs when one-phase commit transactions are performed is fixed.

Issues related to the XA protocol are fixed.

Category

Description

New feature

The CTS persistence feature is provided.

Bloom filters that are created by using UDFs are supported.

UDFs are supported in HLL.

By default, the UDF module polarx_udf.so is loaded.

Feature optimization

The authentication process for requests that are sent over the PolarDB-X proprietary protocol is optimized.

Fixed issue

The following issue is fixed: Assert operations cause data nodes to fail.

Specific content is removed from logs that are recorded when the PolarDB-X proprietary protocol is used.

Category

Description

New feature

A proprietary protocol is provided by PolarDB-X.

The TSO service is supported.

Timestamp-based distributed transactions are supported.

The primary-secondary consistent read feature is provided.

Fixed issue

The following issue is fixed: After a server fails during binlog rotation in an XA transaction, the XA transaction cannot be restored after the server is restarted.

The following issue is fixed: XA transactions cannot be replayed on follower nodes due to internal locks.

The following issue is fixed: When an XA_PREPARE_EVENT log event is performed, logs are incorrectly replayed and the data that is stored in the disk is incorrect.