0.8 is a backwards incompatible release, so this behaviour is intended by design. We will write about the deployment plan in more detail, but at a high level, here is what you would want to do -
1. Use a separate zookeeper namespace on your existing zk cluster, or use a new zk cluster with the same namespace as your 0.7 Kafka deployment. 2. Set up a 0.8 Kafka cluster per 0.7 Kafka cluster and run the migration tool to copy data from 0.7 to 0.8. 3. Move all consumers to 0.8 Kafka cluster 4. Move all producers to 0.8 Kafka cluster In general, for a topic, you want to ensure that all consumers have moved to 0.8 cluster before the producers of the same topic. Thanks, Neha On Mon, Nov 26, 2012 at 2:21 PM, Jason Rosenberg <j...@squareup.com> wrote: > Hmmm.....this sounds like a bug. If I want to upgrade my kafka version, > without changing my zookeeper deployment, will I have issues? Will I have > to manually go and reset all of the zookeeper data directories? > > Jason > > On Mon, Nov 26, 2012 at 1:45 PM, Chris Curtin <curtin.ch...@gmail.com>wrote: >> >> One note for others: you need to remove your zookeeper configuration >> (/opt/zookeeper/version-2 or /tmp/zookeeper/version-2) otherwise you get >> tons of errors starting the brokers. Looks like the data stored in >> zookeeper for 0.7.2 isn't compatible with 0.8.0. >> >>
