Jobs on scavenged nodes never get preempted, only get canceled

Viewed 13

I sometimes run jobs on the scavenge partition on my HPC system. However, I've never had a job that was preempted, which should happen when a higher priority user requests to use a particular scavenge node.

Instead, my jobs regularly get canceled. I get something similar to following message in my all of my canceled slurm out files:

...
   3540500   2969.8072      0.66356586     0.19427035     1.1890431      1.7702431      14.373949      14.373949      14.373949    
   3541000   2988.6985      0.63768049     0.16779792     1.1237651      1.6317543      14.404362      14.404362      14.404362    
   3541500   3000.7733      0.65101783     0.1871139      1.1630755      1.7166775      14.423735      14.423735      14.423735    
slurmstepd: error: *** STEP 2943114.0 ON p08r03n30 CANCELLED AT 2022-09-16T00:48:13 ***
srun: error: p08r03n30: tasks 0-3: Terminated
srun: Force Terminated StepId=2943114.0
Fri Sep 16 00:48:13 EDT 2022
Completed in 0:01:04:57.589

This makes it difficult because I need to go in and manually requeue these jobs. Since I may only catch them after my first set of jobs have already finished, this occasionally doubles the time needed to run any set of jobs.

I asked my system administrators about this, but the last time I heard from them they weren't quite sure why jobs were getting canceled.

Is there any documentation or setting for why a job would be for terminated rather than preempted (to allow for requeueing in the future)?

0 Answers
Related