event source - Scale based on events while preserve rich connection & processing semantics with event source (in order processing, retries, deadletter, checkpointing) - Some main logic - No message: to HPA to drive scale out - each replica of deployment is actively processing messages, or each replica is processing a batch of messages in a distributed manner - ScaledObject ​is used to define messge concurrently - Schedule example: send scheduled message and register message listener - Batch example : send messages in batches, split into lists - Filter example: RocketMQ offer some SQL

rebalance ，降低消费延迟概率  集群中某些消费端假死不影响整体消费进度  客户端更加轻量，多语言友好 流批一体  在Streaming场景下，单一消费者消费保证顺 序  在 batch 场景下，无需保证顺序，可以多个 consumer 加快数据读取速度 你集群是正常的，但我消费就是出问题了，怎么办！？ 无损弹性扩缩 固定分区面临的挑战 • 无切换架构中，主节点宕机，备节点不可写，分区数量减少

5.0 34 Apache RocketMQ 5.0 35 Apache RocketMQ 5.0 On-Line Analytical Processing Streaming Processing Offline Processing Metrics Logs IoT Data Finance Business Game Implementation 36 Aliware

Notification Service Specification 1.1 http://www.omg.org/spec/NOT/1.1/PDF  Distributed Transaction Processing: The XA Specification http://pubs.opengroup.org/onlinepubs/009680699/toc.pdf  RocketMQ Benchmark