This can fail due to race conditions now: e.g. a cleanup node to rm a failed .bam write gets killed, then you have a broken product (truncated .bam) if you kill before resubmitting. If you don't kill, then you might submit and have a .bam cleaned up between the new node being marked completed due to the truncated/broken .bam existing, thus making the new workflow fail as the .bam is rmed out from under it.