Bug 5983

In case of the error code paths (like collective timeout) it is possible
that a callback provided by PMIx will be called twice leading to a
segmentation fault.
This commit fixes it by properly accounting callback invocations.

Bug 5983

Bug 5935

This could have caused core dumps if communication failed for one
reason or another.

Signed-off-by: Boris Karasev <karasev.b@gmail.com>

Bug 5935

This patch does 2 things:
1. When a step fails on some node, then mark it as complete on those
   nodes. This is needed so that when the step ends on the other
   nodes, slurmctld recognized the step as completely done.
2. If the step does not have the --no-kill option set, then when some
   node fails, send a request to terminate the step on ALL of its nodes.

Bug 5805

when env is overwritten by the command line.

Bug 5977

Bug 6016

Bug 6092

configuration.

Bug 5872

Bug 6015.

Was attributed to the wrong author by mistake.

This reverts commit 8b5acf90.

Bug 5977

started on a node.

Bug 5865

Bug 6061

bug 6016

Bug 6015

Bug 6015

The srun man page states that --prolog=none or --epilog=none would
prevent the SrunProlog or SrunEpilog scripts defined in slurm.conf from
running. However, with SLURM_TEST_EXEC set in the environment, srun
would look for a script named "none" and run it if found. If not found,
then srun would print an error message, but the SrunProlog or
SrunEpilog would (correctly) not run.

* Prevent srun from running a script named "none" that is in the path.
* This also prevents the error message if it is not found.

Bug 5948.

Moved to https://github.com/SchedMD/slurm-gcp

Left README with link to new location. Will remove gcp dir in
master/19.05.

Bug 5842

Bug 5985

scheduler.

Main scheduler detects if a job asked for more than one partition and splits
one scheduling tasks for each partition. If the job asked for a reservation and
the intersection between the nodes in the reservation and the partition does
not suffice to satisfy the job, then skip scheduling this partition.

Otherwise this may cause useless scheduling effort and possibly overwriting
incorrectly the PD reason of a job.

bug 5895

when read in from the slurm.conf file.

Bug 5994

Bug 5704

Caused bad license counts. Comes from a8961fd7.

Bug 5704

Bug 5686

Regression introduced when support for up to 9 backup controllers was
added in c61ffa1a.

Bug 6001.

Bug 5826

Bug 5793

when trying to register.

If slurmd never connects to slurmctld and then slurmd is shutdown,
there is a small memory leak.
This is not an important memory leak; however, this will fix things
for valgrind when testing a slurmd by itself.

Bug 5989

Set errno to return value of pthread_* functions during error.
All pthread macros now handle errors in the same way.

Do not set errno immediately, because in threaded programs errno is a
macro that expands into a function call. We don't want to incur that
performance cost unless it's actually an error.
See The Linux Programming Interface, 3.5.2 (pg.53) and 29.2 (pg 621).

Bug 5901

Bug 5976

Bug 5917

Bug 5865

to not require -Wno-format-truncation CFLAG.

Bug 4310
Bug 5655
Bug 3482

Fix memory leak by freeing iterator.
Simplify logic to be more intuitive.
Separate out _check_exec() into two logically different portions.
Get rid of unnecessary goto's.

Bug 5848

and distributed.

Bug 5774

previously accrued for priority.

Bug 5881

Brian approved.

have the same id (for example, if a job was requeued) by also checking the
submit time.

Bug 5197

Bug 5197