KIP-19 should address this issue. https://cwiki.apache.org/confluence/display/KAFKA/KIP-19+-+Add+a+request+ti meout+to+NetworkClient
Thanks, Jiangjie (Becket) Qin On 6/8/15, 10:44 AM, "Christofer Hedbrandh" <christo...@knewton.com> wrote: >I think this question might relate to the very recently posted "callback >handler is not getting called if cluster is down" topic from "ankit >tyagi". > >I am using the 0.8.2.1 new producer send(ProducerRecord<K,V> record, >Callback callback) with a Callback and never calling .get() on the >Future<RecordMetadata>. I have noticed that it takes ~10-15 minutes before >Callback.onCompletion() is called with an Exception when I take down the >Kafka cluster or block connections. > >From the documentation it is unclear what the expected producer behavior >should be. I had assumed that the configuration "retries" and "timeout.ms" >should control how long it would take for the Callback to be called with >an >Exception, but that does not seem to be the case. > >How do I know how long it will take for the Callback to be called with an >Exception. And can I somehow configure this to be much shorter than ~10-15 >minutes. > >relevant broker configs: >zookeeper.session.timeout.ms: 6000 >zookeeper.connection.timeout.m: 1000000 > >relevant producer configs: >retries: 0 >linger.ms: 0 >timeout.ms: 30000 >metadata.fetch.timeout.ms: 10000 >metadata.max.age.ms: 10000 >reconnect.backoff.ms: 10 >retry.backoff.ms: 100 > > >Thanks, >Christofer
