Dear Chaoyi,

I realize this update probably comes too late to be relevant anymore, but I am posting this as a future reference. I just released PSOM 1.2.0 which should be stable on *nix systems.
https://github.com/SIMEXP/psom/releases/...

One of the new features is the detection of inactive jobs, which become marked as failed. You should therefore not have jobs "running" forever. You can use opt.nb_resub to automatically resubmit jobs that fail a number of times before giving up. So if some of your jobs randomly crash with an out of memory, setting opt.nb_resub to 1 or 2 may be enough to complete the pipeline fully automatically despite the failures. 

I hope this helps. Best,

Pierre