Automatic offset commit led to the loss of some messages about confluent-kafka-go

[Mashiro707]

以下内容是由ChatGPT翻译-->以下内容是由Google翻译

[milindl]

Unfortunately, there is no in-built way to manage this situation. It has to be resolved on the application end, for example, by maintaining a map of offsets to booleans, and moving the stored offsets forward only if all the prior offsets are processed already. Consumption could block in case the map exceeds a specific size to prevent the map from growing too much.

In the future, there is a proposed KIP which will handle it on the broker-end, but that might take a lot of time if it's implemented. So I'd suggest not waiting for that.