in Gunther's patch, Hive Server 2 will use the session handle for the handle identifier.
So there isn't an issue, right?
The use of finally has bad properties with respect to debugability because any exceptions thrown out of the finally block (including ones like OOM or NPE) will hide the original exception.
The -f case is certainly a bug, still an improvement over the current trunk (or Hive 0.10) behavior.
My inclination is to wait for the test cases that I launched this afternoon to finish and roll the rc with Gunther's version of the patch tomorrow morning. We can file a follow up jira with a refactoring of the CliDriver.run that doesn't have so many return points. smile
Does that sounds reasonable?