Dear Zaixu,

That's not something I have observed before unfortunately. This is puzzling. Maybe something that could explain that somehow would be multiple pipeline managers being active in the same logs folder. If you don't mind sharing the code I can have a look. Also you may want to have a look at the option opt_psom.nb_resub. This sets the number of times a failed job will be resubmitted before being declared failed for good. This shouldn't happened with a "finished" job, but if you can change the behaviour with that parameter that would be a lead to what is going wrong. 

Best,

Pierre