Also I think that completing jobs should be dumped to disk when saving
system state. Or else, after slurmctld restart a completing node with
tasks hanged may be wrongly made idle(the job reported on the node is
taken orphaned) and allocated to new jobs.
在 2010-09-16四的 08:46 -0700,Jette, Moe写道: