Typically "preferred leader election” would fail if/when one or more brokers still did not come back online after being down for some time. Is that your scenario?
-Zakee > On Aug 11, 2016, at 12:42 AM, Sudev A C <sudev...@goibibo.com> wrote: > > Hi, > > With *auto.leader.rebalance.enable=true* Kafka runs preferred election tool > in regular intervals rebalancing topic-partition leaders. Now if preferred > election fails to elect the preferred leader for a topic partition it falls > back to any other leader in ISR. > > But for us in production when this happens, the newly elected leader fails > to serve/produce messages anymore. If you describe the topic the latest > offset shown is *0* but a leader is shown. > > My current fix is to restart the controller broker. > 1. New controller comes up..fresh preferred election happens . > 2. During this time around offsets comes back to a number from zero. > > Log - http://pastebin.com/raw/qeBnNPA6# > Kafka Version - 0.9.1 > > Should I turn off auto.leader.rebalance or can I work around this problem. > Any suggestions ? > Why would offset go to zero ? > > -- > Thanks > Sudev Ambadi ____________________________________________________________ Affordable Wireless Plans Set up is easy. Get online in minutes. Starting at only $14.95 per month! www.netzero.net?refcd=nzmem0216
