I'm working on https://issues.cloudera.org/browse/IMPALA-4277, which entails building against Hive 2 and Hadoop 3 and HBase 3. This entails making some small incompatible changes to the fe Java code to use updated APIs. My first plan was to try and create compatibility shims, but this looks like it will be more trouble than it's worth.
So I'm proposing to create a new branch called hadoop-next that will include any incompatible changes so that we can collaborate on this work. I think we should keep this branch compilable but initially not worry about getting tests to pass. I'll also create a gerrit branch for it. I anticipate that there will only be a small diff between master and this branch, and I'll periodically merge from master to hadoop-next. Eventually once we're ready to bump our dependency versions, we can cherry-pick the changes back to master. I believe the policy for this is lazy consensus, so I'm going to go ahead and do it :) - Tim