[ https://issues.apache.org/jira/browse/KAFKA-6326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16283053#comment-16283053 ]
HongLiang edited comment on KAFKA-6326 at 12/8/17 4:51 AM: ----------------------------------------------------------- [~huxi_2b] zookeeper.session.timeout.ms is 120 sec. but the zk session timeout not problem in this case. the problem is broker has down. the controller still send broker .wait 30 sec. I think controller not wait. because controller know the broker has down. was (Author: hongliang): [~huxi_2b] zookeeper.session.timeout.ms is 120 sec. but the zk session timeout not problem in this case. the problem is broker has down(zookeeper has know in session.timeout). the controller still send broker .wait 30 sec. I think controller not wait. because controller know the broker has down. > when broker is unavailable(such as broker's machine is down), controller will > wait 30 sec timeout > -------------------------------------------------------------------------------------------------- > > Key: KAFKA-6326 > URL: https://issues.apache.org/jira/browse/KAFKA-6326 > Project: Kafka > Issue Type: Bug > Components: controller > Affects Versions: 1.0.0 > Reporter: HongLiang > Attachments: _f7d1d2b4-39ae-4e02-8519-99bcba849668.png, > _f7d1d2b4-39ae-4e02-8519-99bcba849668.png, fast-recver-shutdownbroker.diff > > > when broker is unavailable(such as broker's machine is down), controller will > wait 30 sec timeout by dedault. it seems to be that the timeout waiting is > not necessary. It will be increase the MTTR of dead broker . -- This message was sent by Atlassian JIRA (v6.4.14#64029)