If jobs within a single stage share a single resource it may be necessary to tear it down after phase completion (failure/success). This could be implemented as an optional cleanup Job. It should be possible to run the cleanup Job only upon failure.