Improve CXConnect's "self-recovery" capability when it fails. The fact that the job can run for 10 hours, generating over 40+GB (representing 22 hours) of datafiles, then fail at the last moment with only one or 2 hours to go, and delete all those data files, and start all the way back at square one, just doesn't seem optimized.

When the job fails, it should be able to evaluate its failure point (at least to the nearest hour) from the datafiles and logs, and then pick back up from that recovery point.