Comment 4 for bug 1731456

Right, the jobs would fail anyway, as they got killed by timeout. But even when this happened, we used to have a safe margin reserved for log collection -- we got logs even from time-outed jobs. This no longer seems to work well in the cases linked above. Maybe the safe margin we had got somehow shrunk by accident. The jobs reported 3 hr 4 min run length, while previously our jobs that timed out reported run lengths around 2 hr 50 min.