letaoj commented on PR #138: URL: https://github.com/apache/flink-agents/pull/138#issuecomment-3315108162
@xintongsong > When calling put() with an ActionState whose generatedActionTask is not null (meaning the action is not finished), the action state will not be flushed to the external Kafka and will be kept in memory, until another put() that changes the generatedActionTask to null (meaning the action is finished). Am I understanding correctly? We will flush the data to the backend store in every put request. Kafka's compaction logic should ideally compact the same key with two different value when later the action is completed and present the latest value when read it. During recovery, if the action is not finished, then we will go ahead and execute the action directly instead of looking it up from the store. > When we say get() will ignore states after a not-found sequence number, does it include new states written after the not-founding get()? E.g., we put s1, s2 into the store, and then try to get s1 with a mismatched action or seq-num, then its expected that both s1, s2 will not be used. But what if we then put a s3 into the store? Yes, it will ignore it. Since the newly written s3 will not and also should not be used for recovery. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
