Hello Libo, When ZK resumes from a soft failure, like a GC, it will mark the ephemeral nodes as session timed out, and the brokers will try to re-register upon receiving the session timeout. You can re-produce this issue by signal pause the ZK process.
Guozhang On Fri, Feb 14, 2014 at 12:15 PM, Yu, Libo <libo...@citi.com> wrote: > Hi team, > > We have three brokers on our production cluster. I noticed two of them > somehow > got offline and then re-registered with zookeeper and got back online. It > seems the > issue was caused by some zookeeper issue. So I want to know what may be > the possible > cases of the issue. If I want to reproduce the issue, is there any way to > do it? Thanks. > > Regards, > > Libo > > -- -- Guozhang
