Hi Elias, You'll have to do some rolling restarts, but downtime can be limited. There's two things you have to consider at a high level: 1) How to migrate zookeeper without downtime -Starting with a quorum of 3, add two of the new servers to the quorum bringing it up to 5 -Once everything is in sync, you can kill two of the old servers and update the quorum back to 3 in the config -Then you'll do a basic replace of the remaining old server -Further discussion in detail on the ZK mailing list --> http://zookeeper-user.578899.n2.nabble.com/How-to-replace-a-zookeeper-server-td7472559.html -If you're on ZK 3.5, you're life can be simplified somewhat with dynamic configuration --> https://zookeeper.apache.org/doc/trunk/zookeeperReconfig.html#ch_reconfig_dyn
2) Update the ZK quorum for kafka -Update the zookeeper.connect string in the kafka to reflect the new 5 node quorum you create above, rolling restart kafka -Once you have the 3 node new quorum up and running, update zookeeper.connect again and rolling restart kafka -Note: any old high level consumers will be impacted by the changed zookeeper string as well, so you'll need to consider that Frankly, if you can take some downtime, it's probably easier to spin up the new zookeeper quorum on it's own, copy over the zookeeper data directories and do a hard migration. You could do that work in the background pretty easily, then bring down kafka, make sure you have everything sync'd across to the new quorum, update zookeeper.connect for kafka, start the new quorum, start kafka, and be done without any of the mess above. Hope that helps, Dustin On Wed, Jun 8, 2016 at 2:11 AM, Elias Abacioglu < elias.abacio...@deltaprojects.com> wrote: > Hi, > > Is there a proper way to change the zookeeper cluster? > I want to migrate new zookeeper cluster, preferably with little or no > downtime. > > Is it possible? >
