Add job burst_buffer_state and delay_boot fields to scontrol and squeue
    output. Also add ability to modify delay_boot from scontrol.

Add salloc, sbatch and srun option of --delay-boot=<time>, which will
    temporarily delay booting nodes into the desired state for a job in the
    hope of using nodes already

Add job AdminComment field which can only be set by a Slurm administrator.

Add event trigger on burst buffer errors (see strigger man page,
    --burst_buffer option).

Insure reported expected job start time is not in the past for pending jobs.
bug 3002

burst_buffer/cray: Requeue, but do not hold a job which fails the pre_run
    operation.
bug 3009

Only mark job_id as zero for batch step (when all job steps would be
cleared), not for individual steps which prevented successive steps
from being cancelled.

Bug 2984.

slurmstepd modified to pre-load all relevant plugins at startup to avoid
    the possibility of modified plugins later resulting in inconsistent API
    or data structures and a failure of slurmstepd.
bug 2334

bug 2655

Bug 2983.

Don't about step launch if job reaches expected end time while node is
    configuring/booting (NOTE: The job end time will be adjusted after node
    becomes ready for use).
bug 2985

Optionally lock slurmstepd in memory for performance reasons and to avoid
    possible SIGBUS if the daemon is paged out at the time of a Slurm upgrade
    (changing plugins). Controlled via new LaunchParameters options of
    slurmstepd_memlock and slurmstepd_memlock_all.
bug 2334

frequency other than AcctGatherNodeFreq.

Filesystem instead of Lustre.

Network dataset.

Locking slurmstepd in memory can result in exhausting real memory
  in some cases, resulting in failure of the slurmstepd process.
  This reverts commit 03cf4a5d,
  but the logic will be returned using a configuration parameter
  in Slurm version 17.02.
bug 2334

This should improve performance and prevent failure if a local
  group ID lookup fails.
bug 2928

Prevent slurmd abort if hwloc library fails to populate the "children"
    arrays (observed with hwloc version "dev-333-g85ea6e4").

4 billion jobs in your database you will want to update your slurmctld at
the same time as your slurmdbd to prevent roll over of this variable as
it is 32 bit previous versions of Slurm.

Bug 2955.

Make EnforcePartLimit support logic work with any ordering of partitions
    in job submit request.
Developed jointly with Alejandro Sanchez <alex@schedmd.com>
bug 2920

Calculation switched the node count in place of the cpu count, which
results in incorrect estimates.

CID 44784.

CID 44787.

Bug 2950. Also identified as CID 56684 (copy+paste error).

CID 45023 and 45024.

Fix task:CPU binding logic for some processors. This bug was introduced
in version 16.05.1 to address KNL bunding problem.
bug 2972

Needed due to part_filter_set() calls; without write lock this can race
returning inconsistent results to 'sinfo'.

Bug 2958.

Regression test fixes if SelectTypePlugin not managing memory and no node
    memory size set (defaults to 1 MB per node).

Fix race condition in the account_gather plugin that could result in job
    stuck in COMPLETING state.
bug 2973

Prior logic used to create an advanced reservation based upon a core
  count would ignore the specialized cores. Then when a job tried to
  use the reservation, it would consider the specialized cores and
  not be able to use the core count used in the reservation creation.
  This change considers specialized cores when creating the reservation.

If there are a huge number of jobs, the previous logic could keep
  the slurmctld daemon busy with locks set for several seconds.
bug 2719

If slurmstepd had been swapped out before upgrade happened it could
  easily lead to SIGBUS at any time after upgrade. Prevent that by
  mlocking it.
bug 2334