[ https://issues.apache.org/jira/browse/KAFKA-9301?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Tomoyuki Saito updated KAFKA-9301: ---------------------------------- Description: h2. ProducerBatch split makes ProducerBatch.produceFuture completed KAFKA-3995 introduced ProducerBatch split and resend when RecordTooLargeException happens on broker side. When ProducerBatch split happens, ProducerBatch.produceFuture becomes completed, even though records in a batch will be resent to a broker. h2. KafkaProducer#flush implementation With the current implementation, KafkaProducer#flush blocks until accumulated ProducerBatches get completed. As described above, that does not ensure all the sent records get completed. This issue is also mentioned in: https://github.com/apache/kafka/pull/6469 was: h2. ProducerBatch split makes ProducerBatch.produceFuture completed KAFKA-3995 introduced ProducerBatch split and resend when RecordTooLargeException happens on broker side. When ProducerBatch split happens, ProducerBatch.produceFuture becomes completed, even though records in a batch will be resent to a broker. h2. KafkaProducer#flush implementation With the current implementation, KafkaProducer#flush blocks until accumulated ProducerBatches to be completed. As described above, that does not ensure all the sent records get completed. This issue is also mentioned in: https://github.com/apache/kafka/pull/6469 > KafkaProducer#flush should block until all the sent records get completed > ------------------------------------------------------------------------- > > Key: KAFKA-9301 > URL: https://issues.apache.org/jira/browse/KAFKA-9301 > Project: Kafka > Issue Type: Bug > Components: producer > Affects Versions: 0.11.0.0 > Reporter: Tomoyuki Saito > Priority: Major > > h2. ProducerBatch split makes ProducerBatch.produceFuture completed > KAFKA-3995 introduced ProducerBatch split and resend when > RecordTooLargeException happens on broker side. When ProducerBatch split > happens, ProducerBatch.produceFuture becomes completed, even though records > in a batch will be resent to a broker. > h2. KafkaProducer#flush implementation > With the current implementation, KafkaProducer#flush blocks until accumulated > ProducerBatches get completed. As described above, that does not ensure all > the sent records get completed. > This issue is also mentioned in: https://github.com/apache/kafka/pull/6469 -- This message was sent by Atlassian Jira (v8.3.4#803005)