for solution A , actually including two parts: - zookeeper : manual operation to migrate zookeeper, so far I think it needs at least 6 steps. this part will be tricky, we will have 3 zookeeper exhibitor , and 5 new zookeeper without e - Kafka : this should be easier , try launch new one to join the brokers cluster, and manual migration / reassign partitions.
By the way, I also have a concern about `Solution A`, you know we have 5 production cluster right now, the manual effort will be big and dangerous. :), that's why I ask your guys experience firstly here. On Wed, Jan 3, 2018 at 8:39 PM, Tony Liu <jiangtao....@zuora.com> wrote: > Hi All, > > This post here is aimed to ask experience about what did you do migration > `Kafka/zookeeper` ? :) > > All of Kafka/zookeeper are running on AWS, because of some reason, we have > to replace all the existed server (you can simply think we will terminate > the old server and create new server to replace it), which is running with > Kafka/zookeeper. > > the Kafka infrastructure : > > - 3 zookeeper with exhibitor running together. > - 6 brokers. > > requirements: > > - terminate all of the 3 zookeeper with exhibitor, and also 6 brokers. > - launch 5 new zookeepers without exhibitor. > - launch 6 new brokers. > - 0 or minimum downtime, try our best no impact for the Kafka client. > > > so far, I have two prepared solution, but any of them has pros & cons, the > solution A is too much manual operations needed, the solution B is some > development effort required and also need to migration all of the Kafka > client to use the self-developed Kafka client with `fail-over` feature. > > *Solution A*: simply speaking, this solution is mainly working on manual > operation to add znode / broker to the existed cluster, when all of the > thing ready, I will step by step to terminate the old one. > > > > > *Solution B:* simply speaking, the solution will work like fail-over > way, a new cluster will be launched, and control Kafka client's traffic to > the new cluster, and terminate the old cluster. > > > > > >
