Hi team, I was going through flink agent and kafka source implementations. • I see we can have process order by key (sequential processing per key). • Do we have discussion on how Kafka Queue semantics will be helpful for out of order records processing/inferencing, as much as parallel as possible to ensure high throughput ?

Hi @Shekhar Rajak, I have learned about Kafka Queue, and I'm afraid it's not a good match for Flink Agents. Flink's stateful processing requires all data with the same key to be processed on by the same parallel sub-task, so that the state of that key is only accessed from one place, and so is Flink Agents. Flink also has stateless use cases where Kafka Queue might be applicable, but Flink Agents are always stateful. And for those stateless use cases, the computation overhead is usually not heavy, so it's probably not very necessary to increase the processing parallelism with Kafka Queue.