made

Mention the ability to use glob(7) patterns for Epilog and Prolog
in the slurm.conf(5) man page.

When prolog and epilog scripts start to diverge on multiple clusters
within the same center it begins to get very painful to keep these
scripts up-to-date, and their implementation usually ends up being
quite convoluted.

One solution is to have prolog and epilog simply run a series of
scripts from a directory, e.g. the epilog might do

 for script in /etc/slurm/epilog.d/*; do
   $script || exit 1;
 done

Then an optional bit of prolog/epilog logic can be implemented
as a little scriptlet and only installed on the needed clusters.

However, it might be nice if this globbing functionality was standard in
SLURM, because it would be more easily available and standardized
for sysadmins.

This is especially nice for plugins that use the job_control_env()
environment to affect behavior of the epilog and prolog. With

 Prolog=/etc/slurm/prolog.d/*
 Epilog=/etc/slurm/epilog.d/*

in slurm.conf, a plugin could be a self-contained package by dropping
a setup script into /etc/slurm/prolog.d/, a cleanup script into
/etc/slurm/epilog.d/ and the actual plugin which implements the
srun/sbatch/salloc option could be installed into /etc/slurm/lua.d/
or /etc/slurm/plugstack.conf.d/

This patch implements a glob syntax for the run_script logic
in slurmd. It should not change behavior of existing configs,
since they should not have any special glob characters in
the Prolog and Epilog config settings.

I'm just sending this out as a request for comments at this time.

for this yet though.

Update daemon core file location documenation and modify slurmdbd and
slurmctld code to use original working directory if -D (foreground)
option used. Based upon patch from Don Albert, Bull.

There is a delay in slurmstepd that prevents jobs with many nodes
from sending step completion messages at the same time to the srun
command which is not needed on front end systems (bluegene or cray).
This has been removed and should improve performance.

Note this is an old bug. The new code keeps slurmstepd alive and it
keeps trying to send step completion message to slurmctld.

This make the same patch to select/linear as Carles Fenoy's patch to
select/cons_res plugin.

Added gres_cpus test. Without this test it could lead to the error "fatal: cons_res: sync loop not progressing" With this patch a job will be rejected if asking for unavailable configuration.

This prevents the problem where a daemon does chdir() then setuid()
resulting in a working directory that it can not write a core file to.

Print an error rather than setting CUDA_VISIBLE_DEVICES environment
variable to  "NoDevFiles" if no device files defined.

Patch 4f68cde5bd6b4fcf839f6694457373c81d9548ba from chaos/slurm by Don Lipari, LLNL

Patch 4f68cde5bd6b4fcf839f6694457373c81d9548ba from chaos/slurm by Don Lipari, LLNL

The change in function call order of commit e60abe43
resulted in slurmd daemons on front-end systems not registering with the
proper node name.

put in a clean state when a block is freed.