We'd like to upgrade our production ZK quorums to 3.4. Apart from https://issues.apache.org/jira/browse/ZOOKEEPER-1149, what issues are there doing a rolling upgrade of ZK nodes from 3.3 to 3.4? Can a mixed cluster work OK? I presume to reduce reelections we should upgrade followers first, then upgrade the leader? Has anyone done this?
If this isn't recommended, our only other option is to build a new quorum and shutdown/switch each service to the new quorum, which is pretty painful (having done that before). --Dave
