The GRES data structures now support per-job, per-node, per-socket,
per-task parameters, plus cpus-per-gres and mem-per-gres. These
fields are not currently used, but the infrastructure is in place.
In addition there is a new cons_tres plugin, which is partly
functional, but needs a lot more work.
There are also some new tests, but slurm_unit and expect.

Add configuration paramerers SlurmctldAddr for use with virtual IP to manage
    backup slurmctld daemons.
bug 4768

Add configuration paramerers SlurmctldPrimaryOnProg and
    SlurmctldPrimaryOffProg, which define programs to execute when a slurmctld
    daemon becomes the primary server or goes from primary to backup mode.
bug 4768

Large results from this will cause scheduler performance problems,
usually due to running inside a VM without the Linux vDSO module.

Bug 4961.

The included lightweight corefile description is no longer valid,
but misleading at best.

Bug introduced in commit 11a75ff4

When any of these --exclusive modes couldn't be satisfied, Slurm was
returning an incorrect ESLURM_NODE_NOT_AVAIL, having as a consequence
scheduling problems as described in the bug. The fix makes it so the
error code is properly set to ESLURM_NODES_BUSY, fixing also the
scheduling problems and working over the correct share_node_bitmap.

Continuation of commits from bug 4932:
e2a14b8d
fc4e5ac9

Bug 5047.

Fix an issue in the bit manipulation log introduced in commit 892ffa89.

Bug 4997.

And related KillOnBadExit setting in slurm.conf.

These only affect an individual job step, not the entire job.

Bug 5023.

Replace select_p_select_jobinfo_sprint() with the same NO-OP
that the other plugins (except alps and bluegene) have implemented.

Bug 5077.

Bug 5049

Time limit was incorrectly changed when swapping qos. Now we check
qos/assoc/partition all together and deny the chang if not consistent.
Fixed how we check coordinator permissions. Rearranged the function to update
QOS before updating partition in order to enforce AllowQOS and DenyQOS options.
One memory leak and some comments fixed too.

Bug 4685

1. Identifies nodes which are unavailable to a specific job,
adding a call to filter_by_node_owner() in select_nodes()
 where the node list is generated.
2. Removes the "unavail_node_str" argument to select_nodes()
as it is no longer useful. This string originally was originally
generated once at the start of the job scheduling logic for all jobs,
but since each job can have a different set of
 unavailable nodes (dedicated to user, group, etc.)
so the same string for all jobs can be misleading.

Bug 4932.

Prevent from wrongly returning, ESLURM_NODE_NOT_AVAIL from _pick_best_nodes when some jobs are using "--exclusive=user"

Bug 4932.

Improve performance of 'squeue -u' when PrivateData=jobs is
enabled by moving the UID filter code ahead of the more expensive
PrivateData=job checks.

Bug 5056.

See commit 0dabf4e7.
Bug 4932.

regression from ef1f3e73.
Bug 4885.

If scaling_available_frequencies file is not available, then derive
values from scaling_min_freq and scaling_max_freq values. If
cpuinfo_cur_freq file is not available then try to use scaling_cur_freq.

bug 4910

Bug 4921

This should never happen, but it does harden the code

This is a refinement to commit 95ab36b3
and also applies to node feature weighting

bug 4968