Hi, Asked in the IRC, no responses, so trying here.
I'm thinking if we can implement end to end exact once semantics based on Kafka's concept of client side offset tracking. In the local RocksDB, store the kafka offset for every value. Then when updating the value always check if the current offset of the current IncomingMessageEnvelope is smaller than the offset in the key value store. If so, discard the message. The down side is it would increase the storage cost. Do you think this works? Any issues? Thanks, Gavin Li
