1) It means the consumer has not call poll() for quite some time and hence be kicked out by the consumer coordinator (the new consumer use heartbeat as the failure detection protocol, you can read more on the wiki docs about its design). If you have high variance of record processing latency, e.g. sometimes a single record would take very long time to process, or if your applications sometimes have a GC it will trigger this situation. In newer versions of Streams it can be auto-handled and would not be thrown to the users. Which Kafka version are you using?
2) this WARN log entry is normal, and can be caused by multiple reasons. For example, brokers would actively close connections that have been idle for some time, and the client will just reconnect to broker when encountering this issue. On Tue, Apr 10, 2018 at 7:34 AM, 杰 杨 <funk...@live.com> wrote: > Hi： > when I run kafka streams applications for a period of time > and will be throw ERROR > > .1) questions > > org.apache.kafka.clients.consumer.internals.ConsumerCoordinator.error.301|[Consumer > clientId=filter-StreamThread-1-consumer, groupId=filter] Offset commit > failed on partition sample-1 at offset 68067604: The coordinator is not > aware of this member. > I check broker logs and found some warnnings > > Member filter-StreamThread-3-consumer-7ce502f2-51d6-4042-9ddb-ffdb0ed37894 > in group filter has failed, removing it from the group > (kafka.coordinator.group.GroupCoordinator) > > the all broker is health. > why sometimes will be throw this error? > > .2) questions > > brokers sometimes throw the WARNNing: > > java.io.IOException: Connection to 1 was disconnected before the response > was read > > is it mens that the broker which broker.id is 1 is't health ? > but I check topic describe found all broker is well > > 发送自 Windows 10 版邮件<https://go.microsoft.com/fwlink/?LinkId=550986>应用 > > -- -- Guozhang
