Sorry for late reply. I don't think, that `flush()` or `close()` would throw -- during flush/close we if sends go wrong, the corresponding `Callback()` would be triggered and we should throw from `recordSendError()` right?
I would not remove key/value from the log statement. Note, that for regular write into repartition and output topics, both `key` and `value` are actual object and human readable. Only if `RecordCollector.send()` is called to write into a changelog topic, the data is serialized outside. Thus, it might actually be b cleaner fix, to not use the regular record collector when writing to change log streams and address the issue there. Compare `StoreChangeLogger#logChange` -- we might want to log error messages there? [ Full content available at: https://github.com/apache/kafka/pull/5224 ] This message was relayed via gitbox.apache.org for [email protected]
