Thanks for addressing all the open issues. All the unit tests pass on fresh checkouts. I'm still seeing the prenominate failure in one dev environment, but cannot discern its cause. I doubt that it will be a problem in general.
I changed to use ant copy plus token filter. should be more elegant than sed.
This is a sound improvement. The current strategy of testing for open ports should stop Hudson from failing.
I explored quite a few time on that. Unfortunately I don't think ant can continue run build if some part fails. Unless we want to return success even if it fails.
I modified a Chukwa unit test to fail, and it looks like both projects have the same issue. I filed
HADOOP-5434 to address this.
Other than changing the port rolling to advance mod the MAX_AVAILABLE_PORT (so if it happens to start at or near that point, a single failure to bind won't fail the test), I think the current patch is fine and could go in as-is.