This is what I would recommend. Identify the leader and restart that one last. Zookeeper should be able to remain functional and server requests during the restart of a single node. Of course you always take a risk if that node does not come back for some reason and then you will be left with and ensemble that cannot achieve quorum. After each restart verify the state of the node.
echo stat | nc IP_ADDRESS_OF_NODE 2181 Will provide that information – Whether it is the leader or follower. We run zookeeper with Solr and do not see an impact with rolling restarts of zookeeper. The only time I see problems is if we are adding new nodes to a live system or there are problems with your ensemble. If for some reason there is a zookeeper node that is not in sync with the others and that node becomes the leader, there is a potential for the data stored in zookeeper to be incomplete or wiped out. Daniel S Washko Solutions Architect [email protected] <http://www.gannett.com/> On 7/11/17, 7:06 PM, "upendar devu" <[email protected]> wrote: We have 3 aws instances of Kafka n zookeeper each of them.can we do rolling restart of zookeeper one instance at a time. Are there any impacts doing this? Please let me know. Thanks
