batch job submission. That srun should be long gone when the allocation
actually occurs.

Failure to do so seems to cause slurmctld exit() memory corruption and
seg fault.

  /etc/init.d/slurm

inducing invalid memory reference.

prolog and epilog. Needed to synchronize slurmctld with user job execution:
Don't start script until partition is ready and clear all jobs upon
termination (offloading work from slurmctld).

kill jobs, etc.

perform job startup and termination,

Modify pthread shutdown logic.