polardbx-cdc is a core component of PolarDB-X which is responsible for global binary log generation, publication and subscription.
新增binlog多流能力,实时生成多条逻辑日志流,提供更强的分布式扩展能力。 新增MySQL一键导入能力,用户可通过SQL快速导入主MySQL实例的存量结构和数据,建立起主备完全一致的同步链路。
新增多级归并能力,实现DN节点线性增加时性能无明显衰减。 新增定时构建全量元数据Snapshot能力,解决历史DDL打标记录过度膨胀问题。 新增基于Recover TSO的恢复方式,解决本地以及远端存储无Binlog文件场景下的集群恢复问题。 新增透明消费能力,下游通过dump协议可直接消费归档到OSS的binlog文件。 优化事务落盘机制,内存够用的情况下不强制落盘。 优化从OSS下载binlog文件的方式,支持多线程并行下载,提升实例恢复速度。
修复Dumper进程初次启动时,计算的延迟时间有误的问题。 修复CDC集群发生重启后,下游订阅可能发生dump超时的问题。 修复触发binlog event数据整形时的兼容性和正确性问题,进一步提升DDL变更期间CDC链路的稳定性。 修复消费程序在进行binlog dump前,未发送COM_REGISTER_SLAVE引起CDC主动断连的问题。 修复连续增加主键和删除主键,导致CDC链路中断的问题。 元数据增加数据类型校验,解决因物理表元数据不一致导致整形生效,触发binlog数据错误的问题。 优化dumper Master的选举策略,解决当主备同步有延迟时,进度落后的dumper被选举为master的问题。
Added the capability of multiple binlog streams, generating multiple logical log streams in real-time, and providing stronger distributed expansion capabilities. Added one-click MySQL import capability, users can quickly import the structural and data inventory of the main MySQL instance through SQL, and establish a fully consistent synchronization link between the primary and backup.
Added multi-level merging capability to achieve linear increase in performance without significant attenuation when DN nodes increase. Added the ability to schedule the construction of full metadata snapshots to solve the problem of excessive expansion of historical DDL labeling records. Added recovery methods based on Recover TSO to solve cluster recovery problems in scenarios where there are no Binlog files stored locally or remotely. Added transparent consumption capability, downstream can directly consume binlog files archived to OSS through the dump protocol. Optimized the transaction write-to-disk mechanism and will not force disk write if there is enough memory available. Optimized the way binlog files are downloaded from OSS, supporting multi-threaded parallel downloads to improve instance recovery speed.
Fixed the problem of incorrect calculation of delay time when the Dumper process is started for the first time. Fixed the problem of downstream subscription timeout after CDC cluster restart. Fixed compatibility and correctness issues when triggering binlog event data formatting, further improving the stability of the CDC link during DDL changes. Fixed the problem of CDC actively disconnecting due to not sending COM_REGISTER_SLAVE before binlog dump in the consumer program. Fixed the problem of CDC link interruption caused by continuous addition and deletion of primary keys. Added data type validation to metadata to solve the problem of triggering binlog data errors due to inconsistent physical table metadata. Optimized the election strategy of dumper Master to solve the problem of progress-lagging dumper being elected as master when there is a delay in primary-secondary synchronization.
see v5.4.15-alpha