In 0.7, it's not very easy to decommission a broker using ZK based producers. It's possible to do that with a vip (then you can't do partitioning). In 0.8 (probably 0.8.1), you can use a tool to move all partitions off a broker first and then decommission it.
Thanks, Jun On Sun, Feb 17, 2013 at 2:19 AM, 王国栋 <wangg...@gmail.com> wrote: > Hi Jun, > > If we use high level producer based on zookeeper, how can we decommission a > broker without message loss? > > Since we want to partition the log with IP, if all the brokers use the > same vip, we can not use the customized partition strategy. > > Thanks. > > > On Mon, Jan 7, 2013 at 12:52 AM, Jun Rao <jun...@gmail.com> wrote: > > > In 0.7, one way to do this is to use a vip. All producers send data to > the > > vip. To decommission a broker, you first take the broker out of vip so > no > > new data will be produced to it. Then you let the consumer drain the data > > (you can use ConsumerOffsetChecker to check if all data has been > consumed). > > Finally, you can shut down the broker. > > > > This will be much easier in 0.8 because of replication. > > > > Thanks, > > > > Jun > > > > On Sat, Jan 5, 2013 at 11:34 PM, Bae, Jae Hyeon <metac...@gmail.com> > > wrote: > > > > > Hi > > > > > > If I want to terminate kafka broker gracefully. Before termination, it > > > should stop receiving the traffic from producers and wait until all > > > data will be consumed. > > > > > > I don't think that kafka 0.7.x is supporting this feature. If I want > > > to implement this feature for myself, could you give me a brief sketch > > > of implementation? > > > > > > Thank you > > > Best, Jae > > > > > > > > > -- > Guodong Wang > 王国栋 >
