On a SLURM 2.2.7 setup with the standard FIFO scheduler and no
priorities or accounts/limits, I had a user encounter the following
scenario:
On the same partition, the user (call them "A") queued 4 jobs, and a
second user ("B") queued a job, so 5 total jobs queued. Let's say they
had sequential job IDs from 0 - 4.
Job 0 began to run, and then A placed a hold on job 1. Once job 0
finished, and job 2 began to run, A released the hold on job 1.
However, instead of executing after job 2, it effectively "went to the
end of the line" and did not execute until after job 4 finished. Is
this the expected behavior?
I would have expected, given the FIFO scheduler, that job 1 would have
held its relative position at the head of the queue in a non-runnable
state, and upon release of the hold, it would then regain its runnable
status and be selected for execution before jobs 3, 4, ...
Thanks,
V. Ram
--
http://www.fastmail.fm - Email service worth paying for. Try it for free
