I've noticed that parameter JobCheckpointDir has a (from my point of view) inconsistent behavior.

* in sbatch executions, it is exported as CWD
* in srun it is also exported as CWD,
* except when it is manually set with "--checkpoint-dir=dir". If so, that value is exported.
* value defined in slurm.conf is, as far as I know, never read.

I have created this small patch to correct that behaviour. Now it is exported with the value configured on slurm.conf. If nothing is set, returned value is the defined on common/read_config.h,

The "scontrol hold/release" commands accept either "name=" or "jobname=". I've modified the documentation to only show "jobname" for consistentcy with the "scontrol update" command.

I have also modified the "scancel" command to accept "--jobname=" in addition to the existing "--name=" and "-n".

If a sleep was interrupted or ran a bit long, the backfill scheduler
run times could be significantly wrong as the sleep time was based
upon the calculation of sleep_count x desired_sleep_time. This new
logic captures and uses the actual sleep time for good accuracy.
bug 1939

Conflicts:
	NEWS

bug 1947

Fix teardown race condition that can result in infinite loop.
bug 1947

This prevents a step from being launched if the job is killed
while the prolog is running. Reproducing the original failure
requires use of srun to trigger the prolog and using scancel
while that prolog is running.
bug 1755

anomaly when only asking for 1 (task_id was never set to INFINITE).

Plug memory leak associated with successful replacement of failing node.

on allocation failure, prevent skipping over xfree of strings.

And add missing documenation.
Bug 1921

GRES were not being properly tracks for multiple simultaneous steps.
A step which could have run later could be rejected as never being
able to run.
Replacement for commit dd842d79, which was reverted in commit 6f73812875c
bug 1925

That commit would address a limited subset of problems and introduce
other bugs rather than fixing the root of the problem.

and you use all the GRES up instead of reporting the configuration isn't
available you hold the requesting step until the GRES is available.