step_mgr was incrementing step_id twice for each job step started.
srun only notes that a job is complete when it makes the allocation,
otherwise it only notes that the job step has completed and the
allocation remains active.

step_mgr was incrementing step_id twice for each job step started.

Srun uses existing allocation if SLURM_JOBID environment variable
is set.

slurm_confirm_allocation.

available data).

(block or cyclic).

set, nodes will change state from DOWN to ALLOCATED or IDLE upon registration.

Add new srun options: --kill-off (no kill on node failure), --share, and
--time=minutes.

   needed.

   here, just fall through after calling qsw_setcap()

No processing beyond that in place yet.

Get i/o header size from function call to be kept in sync with un/pack