Oceanbase Versions Save

Version information

Information	Description
Release date	April 24, 2024
Version	V4.2.1_CE_BP5
Commit number	ec03c25
OBServer RPM version	oceanbase-ce-4.2.1.5-105000032024041915

Enhanced features

• Compatibility with MySQL

  • A SET statement that contains the SET NAMES statement can also set other variables.
  • The information_schema.events table of MySQL is supported. #1194

• Other optimizations

  • The [G]V$OB_SESSION view is provided to display session information and help reduce the time required in collecting processlist statistics.
  • The NETWORK_WAIT_TIME column is added to the GV$OB_SQL_AUDIT view to indicate the total amount of time spent on events of the Network class.
  • A switch is provided for controlling parallel execution of DDL statements.
  • The log_storage_compress_all parameter is provided to specify whether to compress clogs for storage, and the log_storage_compress_func parameter is provided to specify the clog compression algorithm.
  • The slog and sstable directories now can be stored in different paths on the disk.
  • The OUTROW storage performance of large object (LOB) data is improved.
  • The issue of PL cache failure is resolved.
  • Adaptive major compaction is optimized for buffer tables.
  • The statistics collection performance is improved.

Product behavioral changes

To ensure stability, the batch rescan optimization feature was disabled by default for the NESTED-LOOP JOIN (NLJ) and SUBPLAN FILTER (SPF) operators during cluster creation in OceanBase Database since V4.2.1_CE_BP2_HF1. This feature is optimized in V4.2.1_CE_BP5 and is now enabled by default. In other words, the batch rescan optimization feature is enabled by default for new tenants. In OceanBase Database of a version earlier than V4.2.1_CE_BP5, the feature is still disabled by default. You can manually enable the feature as needed.

Bug fixes

• Fixed the issue where the query result of an SQL statement in a standalone database is inconsistent with that in a multi-node cluster. #1845
• Improved the performance of hotspot functions in an ARM environment.
• Fixed the issue where requests are accumulated in the queue of the sys tenant because the slog disk is used up.
• Fixed the issue where an additional record is displayed in some views such as information_schema.tables and all_tables after a DDL operation is performed to convert a non-partitioned table into a partitioned table.
• Fixed the issue where in the case of frequent connection and disconnection attempts, Error 4016 is returned during connection establishment when the total number of established connections reaches the specified value, because the reference count is leaked.
• Fixed the issue where the OBServer node can be hung if the obstack command is automatically executed when requests are accumulated in the queue.
• Fixed the issue where Error 4016 is returned when you execute the SHOW TRACE statement for a distributed execution plan after end-to-end tracing is enabled.
• Fixed the issue where the performance deteriorates after SQL plan management (SPM) is enabled.
• Fixed the issue where a core dump occurs if the offset exceeds INT64_MAX when the LIMIT OFFSET syntax is used in the .NET driver.
• Fixed the issue where the memory of the CommonArray memory module leaks due to bypass import.
• Fixed the issue where the memory of the PlJit module leaks due to exceptions.
• Fixed the issue where a core dump occurs during parallel execution when memory allocation fails.
• Fixed the issue of memory bloat during parallel index creation in specific scenarios.
• Fixed the issue where Error 4344 may be returned when you back up a standby database in specific scenarios.
• Fixed the issue where creating a standby tenant in the standby cluster will fail if case sensitivity is enabled for table names of tenants in the primary cluster.
• Fixed the issue where the execution time is abnormal if you execute the create table if not exists statement to create an existing table immediately after the cluster is restarted.

Considerations

OceanBase Database V4.2.1_CE_BP5 fixes the issue where connection establishment fails because the reference count of session connections is leaked. We recommend that you upgrade OceanBase Database to V4.2.1_CE_BP5 as soon as possible.

版本信息

项目	描述
发布日期	2024-04-24
版本号	V4.2.1_CE_BP5
Commit 号	ec03c25
OBServer RPM 版本号	oceanbase-ce-4.2.1.5-105000032024041915

特性增强

• MySQL 兼容性

  • 支持在同一个 SET 语句中同时包含 SET NAMES 语句和使用 SET 设置其他变量。
  • 支持兼容 MySQL 的 information_schema.events 表结构。 #1194

• 其他优化

  • 新增 [G]V$OB_SESSION 视图展示会话信息，优化 Processlist 统计时间。
  • GV$OB_SQL_AUDIT 视图新增 NETWORK_WAIT_TIME 字段，用于展示所有 Network 类事件的总时间。
  • 新增控制 DDL 并行开关。
  • 支持通过 log_storage_compress_all 控制是否开启 Clog 存储压缩功能以及通过 log_storage_compress_func 设置 Clog 存储压缩的压缩算法。
  • 支持将 slog 和 sstable 的目录放在不同的磁盘目录下。
  • 优化 LOB 类型数据外联存储（OUTROW）性能。
  • 优化 PL Cache 缓存失效的问题。
  • 优化 Buffer 表的自适应合并。
  • 优化统计信息收集的性能。

产品行为变更

出于稳定性因素的考虑，从 V4.2.1_CE_BP2_HF1 版本开始，新建集群时，Nested Loop join（NLJ）和 SUBPLAN FILTER（SPF）算子的 BATCH RESCAN 优化默认会被关闭。我们针对该功能进行了专项提升，在 V4.2.1_CE_BP5 版本重新打开该默认开关。V4.2.1_CE_BP5 版本开始，新建的租户默认保持 BATCH RESCAN 优化开启。升级前已存在的老租户维持原来配置，需要手动进行开启。

缺陷修复

• 修复单节点与多节点查询结果不一致的问题。#1845
• 优化 ARM 环境下的性能热点问题。
• 修复 slog 盘满导致系统租户队列积压的问题。
• 修复执行非分区表转分区表 DDL，会导致一些视图（如 information_schema.TABLES/ALL_TABLES）多显示一条记录的问题。
• 修复频繁建连接/断连接场景下，引用计数泄漏导致建连接总数达到一定量后，建连接报错 4016 的问题。
• 修复队列积压情况下自动 obstack 可能导致 observer hang 住的问题。
• 修复开启全链路追踪后分布式执行计划 show trace 报错 4016 的问题。
• 修复打开 SPM 后性能下降的问题。
• 修复 .NET 驱动中使用 limit offset 时遇到 int64_max 溢出导致的 core 问题。
• 修复旁路导入产生 CommonArray 内存模块泄漏的问题。
• 修复异常场景下 PlJit 模块内存泄漏问题。
• 修复分配内存失败场景下，并行执行 core 的问题。
• 修复并行执行创建索引特定场景下内存膨胀的问题。
• 修复特定场景下备库备份可能出现 4344 报错的问题。
• 修复主集群租户设置了区分表名大小写，备集群同步创建备租户会失败的问题。
• 修复集群重启后，立即执行 create table if not exists 创建已经存在的表，执行时间异常的问题。

注意事项

V4.2.1_CE_BP5 版本修复了集群 session 连接数引用计数泄漏导致不能新建连接的问题，请尽快升级到 V4.2.1_CE_BP5 版本。

Version information

Information	Description
Release date	March 13, 2024
Version	V4.2.2_CE_BP1
Commit number	083a68a
OBServer RPM version	oceanbase-ce-4.2.2.1-101000012024030709

Overview

• This release addressed several internal code compatibility issues. To upgrade a cluster from V4.2.2_CE or V4.2.2_CE_HF1 to a version higher than V4.2.2_CE_BP1, you need to upgrade it to V4.2.2_CE_BP1 first.

Bug fixes

• Fixed the issue where during the current execution of transfer, medium compaction, and backup, if an exception occurs, such as OSS write failure, Error 4016 might be reported during physical restore.
• Fixed the issue that could cause core dump on OBServer nodes when creating and switching sessions frequently.
• Fixed the issue that could cause core dump on OBServer nodes due to memory allocation problems in parallel column deletion scenarios.

版本信息

项目	描述
发布日期	2024-03-13
版本号	V4.2.2_CE_BP1
Commit 号	083a68a
OBServer RPM 版本号	oceanbase-ce-4.2.2.1-101000012024030709

发版目的

• 修复若干内部代码兼容性问题，已经部署 V4.2.2_CE 或 V4.2.2_CE_HF1 版本的集群，升级更高版本时，需经停 V4.2.2_CE_BP1 版本。

缺陷修复

• 修复在 Transfer、Medium Compaction 和备份操作并发执行期间，遇到写入 OSS 失败等可重试错误的异常场景时，可能会导致物理恢复数据报错 4016 的问题。
• 修复多次创建并切换 Session 时，可能导致 OBServer Core 的问题。
• 修复在并行删除列场景下，可能由于内存分配问题导致 OBServer Core 的问题。

Version information

Information	Description
Release date	March 5, 2024
Version	V4.2.1_CE_BP4
Commit number	3246b00
OBServer RPM version	oceanbase-ce-4.2.1.4-104000052024022918

Enhanced features

• The overwrite feature is now supported for OBKV Batch Put. To use this feature, you must first upgrade your client to the latest version.
• The manual partition transfer now can be cancelled.
• The query performance is now improved for query statements that contain multiple MIN or MAX functions.
• The index status now can be determined based on the COMMENT field in the information_schema.STATISTICS view.

Product behavioral changes

• The TENANT keyword is changed from a required parameter to an optional one in the table-level restore statement.
• The max_syslog_file_count parameter is changed to control the total number of log files of all types.

Bug fixes

• Fixed the issue where the return result for a UNION ALL operation on TIMESTAMP data may be inaccurate.
• Fixed the issue where Error 4016 is returned for a rewrite failure that occurs due to a parsing error when SELECT and JOIN statements are both used in PS mode.
• Fixed the issue where the performance of the HBase PUT operation of OBKV is affected by the system variable binlog_row_image.
• Fixed the issue where the upgrade_post.py script fails to execute when an upgrade is performed after the enable_rebalance parameter is manually disabled.
• Fixed the issue where memory leaks may occur in the SeArray module when an SQL statement accesses multiple sequences.
• Fixed the issue where SQL execution is stuck due to a deadlock caused by insufficient RPC threads when multiple SQL statements are executed in distributed mode and resource isolation is enabled.
• Fixed the issue where some parameters may fail to be synchronized an ALTER SYSTEM SET statement is used to modify multiple parameters.
• Fixed the issue where a cluster with a large number of tablets becomes unavailable when it is restarted. This is because all tablets in virtual tables are scanned during the restart, which occupies threads and leads to accumulation in the queue.
• Fixed the issue where CPU utilization remains high in a no-load scenario due to conflicts between automatic cleanup tasks when there exists a very large amount of historical statistics. Such conflicts occur when tasks on the previous day are not completed within 24 hours but new tasks are triggered on the current day.
• Fixed the issue where Error 4013 may be returned due to insufficient memory of the sys tenant when the DROP DATABASE statement is executed in a cluster with small specifications.
• Fixed the issue where the import task times out and cannot normally exit due to incorrect stop judgment logic when a large amount of data is imported in bypass mode.
• Fixed the issue where a major compaction cannot be started due to frequent transfer retries during RootService scheduling.
• Fixed the issue where TPS jitter occurs during unit migration and replication in weak-consistency read scenarios.
• Fixed the issue where the query result is inconsistent with that in MySQL when a query statement contains a field whose name starts with a digit.

版本信息

项目	描述
发布日期	2024-03-05
版本号	V4.2.1_CE_BP4
Commit 号	3246b00
OBServer RPM 版本号	oceanbase-ce-4.2.1.4-104000052024022918

特性增强

• OBKV Batch Put 现支持覆盖写功能，使用该功能前需要升级到最新版本的客户端。
• 手动迁移分区功能支持 Cancel。
• 优化查询语句中包含多个 MIN 或 MAX 函数时的查询性能。
• 支持通过 information_schema.STATISTICS 的 COMMENT 字段来判断索引状态。

产品行为变更

• 表级恢复命令调整 TENANT 关键字为可选。
• max_syslog_file_count 调整为控制所有类型的 Log 总量。

缺陷修复

• 修复对 TIMESTAMP 类型的数据进行 Union All 操作时，返回结果可能不正确的问题。
• 修复 PS 模式下包含 Select 和 Join 语句时，可能存在解析问题导致改写无法正确处理报错 4016 的问题。
• 修复 OBKV HBase Put 接口性能问题，修复后性能不受 binlog_row_image 系统变量的影响。
• 修复用户手动关闭 enable_rebalance 参数后升级，执行 upgrade_post.py 脚本失败的问题。
• 修复 SQL 访问多个 Sequence 场景下，可能存在的 SeArray 模块内存泄漏的问题。
• 修复使用资源隔离的场景下，当存在多条 SQL 有分布式执行场景下，可能存在 RPC 线程不足引发死锁，导致 SQL 执行卡住的问题。
• 修复通过一条 ALTER SYSTEM SET 语句修改多个配置项时，可能出现部分配置项同步失败的问题。
• 修复 Tablet 数量过多场景下，重启集群时虚表扫描所有 Tablet 占住线程引发队列积压，导致集群不可用的问题。
• 修复当历史统计信息数据量非常大的情况下，当天的清理任务 24 小时没有执行完，第二天又发起了新的自动清理任务，出现并发冲突，导致空载场景下，CPU 利用率一直很高的问题。
• 修复小规格场景下，执行 DROP DATABASE 可能由于 sys 租户内存过小报错 4013 的问题。
• 修复旁路导入大量数据时，可能由于 Stop 判断逻辑问题导致超时无法正常退出的问题。
• 修复因 RS 调度中 Transfer 重试过于频繁，导致无法启动合并操作的问题。
• 修复弱读业务场景下，Unit 迁移复制过程中 TPS 抖动的问题。
• 修复查询中包含数字开头的字段名时，与 MySQL 表现不一致的问题。

Version information

Information	Description
Release date	February 28, 2024
Version	V4.2.2_CE_HF1
Commit number	c984fe7
OBServer RPM version	oceanbase-ce-4.2.2.0-100010012024022719

Bug fixes

Fixed the issue where DDL execution was stuck due to a schema refresh failure caused by system view changes. This issue occurred when the cluster was either directly upgraded from V4.1.0_CE_BETA or V4.1.0_CE to V4.2.2_CE, or when the cluster was first upgraded to V4.1.0_CE_BETA or V4.1.0_CE and then further upgraded to V4.2.2_CE.

版本信息

项目	描述
发布日期	2024-2-28
版本号	V4.2.2_CE_HF1
Commit 号	c984fe7
OBServer RPM 版本号	oceanbase-ce-4.2.2.0-100010012024022719

缺陷修复

修复部署或在升级过程中经停过 V4.1.0_CE_BETA 或 V4.1.0_CE 版本的集群，在升级到 V4.2.2_CE 版本后，由于系统视图变更引发 Schema 刷新失败，导致 DDL 执行卡住的问题。

Version information

Information	Description
Release date	February 5, 2024
Version	V4.2.1_CE_BP3_HF2
Commit number	73d0496
OBServer RPM version	oceanbase-ce-4.2.1.3-103020042024020317

Bug fixes and improvements

• Supports importing LOB data larger than 2 MB using the Java client.
• Supports cluster maintenance and management through the OBShell command-line tool, allowing you to start or stop OceanBase Database using systemd service.
• Fixed the issue where OBServer core dump could occur due to memory allocation problems in parallel column deletion scenarios.

Considerations

• When using the bypass import feature to import LOB data that exceeds 4 KB, a transfer event could result in import failure. Therefore, we recommend that you temporarily disable the transfer feature.

• When using the bypass import feature to import LOB data that exceeds 4 KB, if a leader switchover occurs during the import process, the system will attempt to perform the import operation again.

版本信息

项目	描述
发布日期	2024-02-05
版本号	V4.2.1_CE_BP3_HF2
Commit 号	73d0496
OBServer RPM 版本号	oceanbase-ce-4.2.1.3-103020042024020317

发版目的

• 支持 Java 客户端导入超过 2M 的 LOB 数据。
• 支持通过 OBShell 命令行工具对集群进行运维管理，支持通过 systemd service 的方式启停 OceanBase 数据库。
• 修复在并行删除列场景下，可能由于内存分配问题导致 OBServer Core 的问题。

注意事项

• 使用旁路导入功能导入超过 4 KB 的 LOB 数据时，如果发生 Transfer，则会导致导入失败，建议使用时临时关闭 Transfer 功能。

• 使用旁路导入功能导入超过 4 KB 的 LOB 数据时，如果导入过程中发生切主，则系统会进行重试。

Version information

Information	Description
Release date	January 25, 2024
Version	V4.2.1_CE_BP3_HF1
Commit number	37a4c62
OBServer RPM version	oceanbase-ce-4.2.1.3-103010052024011916

Bug fixes

• Fixed the issue where calling the DBMS_MONITOR package with decimal type data would result in an Incorrect arguments to control info error.
• Fixed the issue where time-to-live (TTL) tasks in OBKV failed when the table contains LONGTEXT columns.
• Fixed the issue where the insertOrUpdate operation on the OBKV client failed to write data when an auto-increment column of the table was used as the primary key and secondary indexes existed.
• Fixed the issue where a core dump occurred on an OBServer node due to the lack of null pointer check when the OBServer node processed the driver code of Alibaba Cloud Object Storage Service (OSS).
• Fixed the issue where the paging query performance was low due to an inappropriate execution plan in a scenario where the ORDER BY and LIMIT clauses were used.
• Fixed the issue where the major compaction for the standby tenant could be stuck when a backup task and a medium-priority compaction task were concurrently executed.
• Fixed the issue where the query performance was downgraded when a large number of SSTables were involved in major compactions during a transfer operation.

版本信息

项目	描述
发布日期	2024-1-25
版本号	V4.2.1_CE_BP3_HF1
Commit 号	37a4c62
OBServer RPM 版本号	oceanbase-ce-4.2.1.3-103010052024011916

缺陷修复

• 修复调用 DBMS_MONITOR 包传入小数类型数据报错 Incorrect arguments to control info 的问题。
• 修复表中包含 LONGTEXT 列时，OBKV 的 TTL 任务执行失败的问题。
• 修复当表中自增列作为主键且存在二级索引时，使用 OBKV 客户端的 insertOrUpdate 接口写入失败的问题。
• 修复 OBServer 内部在处理 OSS 驱动代码时，缺少的空指针判断，导致的 OBServer Core 的问题。
• 修复 order by + limit 分页场景下，由于执行计划不优导致查询性能慢的问题。
• 修复备份和 medium compaction 并发时，可能导致备租户合并卡住问题。
• 修复 Transfer 期间转储 SSTable 个数比较多时，查询性能下降的问题。

Version information

Information	Description
Release date	January 2, 2024
Version	V4.2.1_CE_BP3
Commit number	8fe69c2
OBServer RPM version	oceanbase-ce-4.2.1.3-103000032023122818

Enhanced features

• Data compression at the communication protocol layer is supported for OBServer nodes and OceanBase Database Proxy (ODP) to reduce the bandwidth costs in data transmission.

• Global indexes are supported in OBKV.

• The INSERT ... ON DUPLICATE KEY UPDATE statement is supported.

• The max_partition_num parameter is added to control the maximum number of partitions that can be created or modified in the MySQL mode.

• Statistics information can be copied by using DBMS_STATS.copy_table_stats.

• Permission-based filtering is implemented in the result set of the information_schema.SCHEMATA table, allowing users to query only schemas for which they have permissions.

• The data synchronization performance is improved for the primary/standby architecture in direct connection mode.

• The system load is reduced in scenarios where SQL statements are throttled using max_concurrent.

• The DATA_PROGRESS column is added to the CDB_OB_BACKUP_TASKS view to show the data backup progress.

Bug fixes and improvements

• Fixed the issue where ERROR 4016 was returned after executing the SHOW CREATE TABLE statement when no length was specified for type inference on the input and output of the JSON_UNQUOTE function.

• Fixed the issue where an error occurred when importing data using the LOAD DATA statement for columns containing JSON or large object (LOB) data exceeding 2 MB.

• Fixed the issue where an exception occurred when connecting to an OBServer node using a .NET driver in Visual Studio for Windows.

• Fixed the issue where an exception occurred when querying data using the ST_Contains function with spatial index.

• Fixed the issue where the SHOW CREATE TABLE statement would return a CHARSET_INVALID error when character_set_results was not specified.

• Fixed the issue where the REPLACE system function did not correctly handle input parameters of the LOB type, resulting in execution errors.

• Fixed the issue where ERROR 4013 was returned when starting the OBServer node due to excessive tenant numbers.

• Fixed the issue of delayed response to business requests and increased tenant CPU utilization during concurrent SQL execution, particularly when a hash join processing a small amount of data is on the right of the SUBPLAN FILTER operator in the execution plan.

• Fixed the issue where a statement may time out if a large number of logs were written in a short period of time in a parallel DML (PDML) scenario.

• Fixed the issue of tenant memory overload caused by memory bloat within the EncodePivot module in small-specification scenarios.

• Fixed the issue of tenant memory overload caused by memory bloat within the SqlExecutor module in multi-table join queries during data transfer.

• Fixed the issue where ERROR 4038 was returned when using the INSERT INTO SELECT statement for bypass import.

• Improved the performance of the OBKV insertup API, which can be enhanced by setting binlog_row_image to MINIMAL.

• Fixed the issue where cluster connection exceptions occurred due to high CPU utilization of background threads in an environment with a large number of tenants.

Considerations

Starting from V4.2.1_CE_BP2_HF1, batch rescan is disabled by default for the NESTED-LOOP JOIN (NLJ) and SUBPLAN FILTER (SPF) operators during cluster creation. However, for clusters upgrading from earlier versions, batch rescan is enabled by default. If stability issues are found in batch rescan for the NLJ or SPF operator in an environment with large amounts data, you can disable batch rescan for the operator in the entire cluster by executing the SET GLOBAL _nlj_batching_enabled = false; statement. If the performance is not as expected after batch rescan is disabled, you can evaluate whether to enable batch rescan again based on the actual situation.

版本信息

项目	描述
发布日期	2024-01-02
版本号	V4.2.1_CE_BP3
Commit 号	8fe69c2
OBServer RPM 版本号	oceanbase-ce-4.2.1.3-103000032023122818

特性增强

• 支持 OBServer 与 obproxy 通信协议层的压缩功能，降低数据传输带宽成本。
• OBKV 支持全局索引的能力。
• 支持 INSERT ... ON DUPLICATE KEY UPDATE 语法。
• 新增配置项 max_partition_num 用于控制 MySQL 模式下允许创建或者更改的最大分区数。
• 支持通过 DBMS_STATS.copy_table_stats 拷贝统计信息。
• 对 information_schema.schemata 表的结果集增加权限过滤，用户只能查询权限范围内的 Schema。
• 优化网络直连主备库架构下的数据同步性能。
• 优化通过 max_concurrent 对 SQL 进行限流场景下系统的负载。
• CDB_OB_BACKUP_TASKS 视图新增 DATA_PROGRESS 字段，用于展示数据备份进度。

缺陷修复

• 修复 JSON_UNQUOTE 类型推导时没有设置 Length，导致 SHOW CREATE TABLE 报错 4016 的问题。
• 修复 Json 或 Lob 类型数据的列数据超过 2M 时，LOAD DATA 导入报错的问题。
• 修复在 Visual Studio for Windows 下，使用 .NET 驱动连接 OBServer 异常的问题。
• 修复使用 ST_Contains 函数时，通过空间索引查询数据异常的问题。
• 修复 character_set_results 设置为空的场景下，SHOW CREATE TABLE 报错 CHARSET_INVALID 的问题。
• 修复输入参数为 Lob 类型时，REPLACE 系统函数没有正确处理，导致执行时报错的问题。
• 修复租户数量过多时，可能导致 OBServer 启动时报错 4013 内存不足的问题。
• 修复 SQL 并发执行时，当执行计划中 SUBPLAN FILTER 右侧是 Hash Join，且 Hash Join 处理数据量比较小时，可能触发业务请求变慢, 租户 CPU 升高的问题。
• 修复 PDML 场景下，短时间内大量日志写入可能导致语句超时的问题。
• 修复小规格场景下，EncodePivot 模块内存膨胀导致租户内存爆的问题。
• 修复 Transfer 过程中，多表 Join 查询时，SqlExecutor 模块内存膨胀导致租户内存爆满的问题。
• 修复使用 INSERT INTO SELECT 语句进行旁路导入数据时，报错 4038 的问题。
• 修复 OBKV insertup 接口性能问题，支持通过设置 binlog_row_image 为 MINIMAL 优化 insertup 接口性能。
• 修复大量租户环境下，后台线程占用 CPU 过高，导致集群连接异常的问题。

注意事项

从 V4.2.1_CE_BP2_HF1 版本开始，新建集群时，Nested Loop join（NLJ）和 SUBPLAN FILTER（SPF）算子的 BATCH RESCAN 优化默认会被关闭。然而，对于从老版本升级上来的集群，它们会默认保持原来的 BATCH RESCAN 打开的状态。对于存量的环境，如果发现 NLJ/SPF BATCH RESCAN 的稳定性问题，可以通过执行 SET GLOBAL _nlj_batching_enabled = false; 命令来关闭该优化，这将在整个集群范围内禁用 NLJ/SPF 的 BATCH RESCAN 优化。如果在关闭优化后性能不符合预期，可以根据具体情况重新评估并选择是否重新开启。