Some frontend PlannerTests rely on HBase tables being split into specific regions with those regions assigned to specific region servers. Right now, the HBase tables are created via the HBase shell with a single region, then they are populated via Hive DMLs, then there is a Java program that splits the tables into appropriate regions and assigns those regions to region servers. Once that is done, nothing is maintaining the assignments. The Java code for doing the splitting is hard to maintain and flaky. The assignments can sometimes drift due to rebalancing.
We should convert this to specify the splits at HBase table creation time. We should have the frontend PlannerTest do assignments at setup time to avoid flakiness due to rebalancing. This should move some flakiness out of dataload.