The current JobBarrier implementation returns immediately if one of the monitored jobs fails. This behavior should change to always wait for all jobs to complete. Because jobs cannot be cancelled there might occur strange side effects if some jobs are still running even though the main processing line moves forward. This is especially true for jobs copying or moving around files aka everything IO related.
There's also a bug in the JobBarrier. The INSTANTIATED status is not handled.
Merged into 1.4.x with rev 15613.