Why ray worker out of memory due to job failed instead of the worker being restarted ？

Ray Libraries (Data, Train, Tune, Serve)
1

image
image3454×1108 413 KB

In other woker out of memory case, the job did not failed, It was still running.

How can ifigure it out ？

2

does it go through if you allocate memory on the worker node?

3

I use ray.data.map api ,the fn parameter should be actor not worker.Then If worker node failed, the job will not be killed