I have been looking for differences in the jhist files between the initial kill, and the second one, and I cannot find much as far as the content is concerned. So I think it might have something to do with the order of the events.
In the second jhist file there are two AMStarted events. I assume that is to allow for the history server to show how many have been started.
The mapFinishTime of all the new MAP_ATTEMPT_FINISH_EVENTS is 0. The finishTime for these events are all showing the same finish time which I think is when the attempt was recovered, not the original event finish time. The state of the event changed from "map" (which seems like a bug) to "SUCCEEDED" which looks more correct to me. And finally the clockSplits counters are all 0 in the new one as well.
Similarly the TASK_FINISH_EVENTS have the finish time of the recovery, not the actual finish time.
I could not find anything else that is significantly different.