When there is a connection error during a loader job run, it dumps a stack trace for every group. This has a tendency to fill up a filesystem if there are a large number of groups involved.

What if there is an option to simply give up the job and try again at a different time? Preferably this would be set per job and not globally.