on BGQ systems.

and SLURM_STEP_ID env vars are set.

Based upon a study of the preemption logic, putting smaller jobs
at the top of the queue (lower priority) will improve algorithm
efficiency in terms of not releasing more resources than needed.

merged together.

Conflicts:
	src/sacct/print.c

4TB of memory on them.

sacct.

The previous logic would get an authentication credential while
waiting for a message to forward. This change will re-read an
authtentication credential if the wait is too long (greater than
or equal to the message timeout).

Replace white space with tabs

Add support for the new ScheduleParameters of preempt_strict_order
and preempt_reorder_count

Add SchedulerParameters options of preempt_reorder_count and
preempt_strict_order.

This is for consistency. All of the other backfill parameters
start with "bf_".

Conflicts:
	NEWS
	src/slurmd/slurmd/req.c

Add support for SchedulerParameters value of bf_max_job_start that limits
the total number of jobs that can be started in a single iteration of the
backfill scheduler.
bug 607

A test for bf_max_job_user was > rather than >=

the correct lib.

Since the QOS priority is 32-bits, base the preemption priority
upon that 32-bit field and ignore the job size.
Variant of commit  a3e74ef2

completes

This avoids some misleading error message

The job preemption order was based upon the lowest 16 bits of a
QOS priority, but the field holds 32 bits. This uses a value of
0xffff if the QOS priority is higher.

If the slurm.conf changes to remove nodes from a job step in the
process of running Node Health Check, the step record would be
purged. Then when the NHC completes, it's step record pointer
would be invalid.
The change here will prevent such a job step record from being
purged.
This was found while investigating bug 612, but would be a very
unusual condition and does not fix this bug.

always be set to SLURM_SUCCESS since it isn't used before.