It seems that we could ignore some IOE in moveOldFiles() and ensure the end results of this call is that all files are gone (either moved, or removed).
This looks like a good idea
Since checksum exceptions should be reasonably common for partial history files, would it make sense to use RawLocalFileSystem as the src fs in moveOldFiles?
Actually, the job history log location is governed by the conf setting "hadoop.job.history.location" and this could possibly be on a FS other than LocalFS. So, I do not think we could use RawLocalFileSystem.