In 0.8, Kafka broker can't be run w/o ZK. In 0.7, you can run a broker without ZK. However, you won't able to run the high level consumer w/o ZK.
Thanks, Jun On Sat, Jan 11, 2014 at 6:52 PM, Ran RanUser <ranuse...@gmail.com> wrote: > I have two questions: > > 1. > > We plan to initially run Kafka without partitioning or replication, simply > having two single node Kafka (broker) servers in two geographically diverse > locations. Using periodic backups/snapshots across locations for disaster > recovery. > > Our producer / consumer applications will process to/from each location. > > For such configurations, is it possible to run the Kafka broker without > Zookeeper? If not, any plans for such a simplified configuration? We'd > like to remove the extra app from the stack (Zookeeper) since it's not > being utilized. > > 2. > > If we need to restore a Kafka node (simplified design above), I was hoping > Kafka and Zookeeper could bootstrap itself as long as we can restore a copy > of the /tmp/kafka-logs data. This dir contains the topics, log data, and > high watermark. > > Unfortunately, it seems without the Zookeeper data, we cannot restore the > node. > > Is there a way to restore a Kafka node if the Zookeeper data cannot be > restored? What are the specific requirements to restore a Kafka broker > from backups, if the /tmp/zookeeper must also be restored, are there any > inconsistent state issues (i.e. backup of /tmp/kafka-logs and > /tmp/zookeeper are not synchronized)? > > > Thank you! >