Hi folks! I've got 4/5 of the subtasks under HBASE-20331 "clean up shaded packaging for 2.1" wrapped up and ready if folks could take the time to review.
The current set of commits all build on each other, there are 5 in total: * HBASE-20332 shaded mapreduce module shouldn't include hadoop * HBASE-20333 Provide a shaded client that allows downstream to provide Hadoop needs. * HBASE-19735 Create a client-tarball assembly (a change from Josh that I have +1ed and will push with these) * HBASE-20615 emphasize use of shaded client jars when they're present in an install * HBASE-20334 add a test that verifies basic client and MR integration In total, these patches make it so that out of the box folks will drop the number of dependency jars they see at runtime from ~430 to ~280. If folks turn off our default "also add everything from 'hadoop classpath' feature, then the number of jars goes from 159 to 10. (personally I think that hadoop classpath thing is a misfeature, but I'm trying to keep the scope from creeping here.) The changes also include new nightly tests that make use of the jars to interact with a standalone cluster. On Hadoop 3 it verifies doing this using Hadoop's client-facing shaded artifacts which further improves the lives of our downstream users. If folks would prefer to see the sum of the patches instead of working through applying them, there's up on a feature branch for the umbrella HBASE-20331: https://git-wip-us.apache.org/repos/asf?p=hbase.git;a=shortlog;h=refs/heads/HBASE-20331 Nightly currently has an example run of the test: https://builds.apache.org/blue/organizations/jenkins/HBase%20Nightly/detail/HBASE-20331/45/pipeline/100 https://builds.apache.org/job/HBase%20Nightly/job/HBASE-20331/45/artifact/output-integration/ Please take a look. -busbey