Bug 7265

We want to do all the dependency tests locally even if globals.local
isn't setup for federation. If globals.local is setup for federation
then we already test AND/OR remote dependencies.

Bug 6068

Bug 6068

Signed-off-by: Marshall Garey <marshall@schedmd.com>

Bug 6068

Signed-off-by: Marshall Garey <marshall@schedmd.com>

Bug 6068

Bug 6068

Bug 8675

Signed-off-by: Brian Christiansen <brian@schedmd.com>

Bug 8772

Signed-off-by: Nate Rini <nate@schedmd.com>

Replacing send_user by log_*.
Using print_success/failure to end the test.
Breaking long lines.

Bug 8772

Signed-off-by: Nate Rini <nate@schedmd.com>

Bug 8772

No functional change

Bug 8318

Use functions defined in globals instead of send_user for printing
information.

Bug 8318

Running the test required more than 4GB of memory. This is quite a lot
and is not needed to properly test the functionality.

Bug 8318

If KillOnBadExit is enabled we should expect the failed step to be
CANCELLED not FAILED.

Bug 8318

Removing trailing spaces.
Breaking long lines.
Renaming funny words.

Bug 7211.

Simply make sure that the second env var is a derivative of the first
env var, so they can never be the same.

Bug 7211

Bug 7022

Signed-off-by: Michael Hinton <hinton@schedmd.com>

Run the srun job in the background initially to get pid.
Directly pass in the pid of the srun job instead of grabbing the first
srun pid found on the system.
Check that bg and fg actually succeed.

Bug 7022

As of commit 5975f124 and 6dcb598c, --gpus-per-task needs -n to be
known beforehand. This isn't possible when there is a node range
(--nodes=X-Y). So these two test cases were no longer were relevant.
Rework them to have a fixed node count, so ntasks can be statically
determined before allocation.

Bug 8560

Replace 10-second sleep with sacct job poll until job shows up.
Poll on both `-j <job_id>` and `--name=<job_name>`, since job data
propagates to both queries at different times.

There was a 10-second sleep for job data to propagate to the DB and
show up in sacct. However, sometimes this sleep wasn't long enough.
Instead of hoping jobs show up within 10 seconds, explicitly wait for
them.

Bug 6460

Bug 7031

Bug 7031

Test 39.12 was not skipped if default partition was non homogeneous
and only 1 node had craynetwork.
Now it uses the new get_craynetwork_count based on get_gres_count.

Warning messages adapted.

Bug 8357

Signed-off-by: Michael Hinton <hinton@schedmd.com>

Code simplification reusing existing code reused in several places.

Bug 8357

Signed-off-by: Michael Hinton <hinton@schedmd.com>

Bug 7094

Bug 7094

Each step has its own cgroup, which applies to GRES GPU devices if
devices are constrained. So a repeat index in CUDA_VISIBLE_DEVICES is
expected, since that index is a local ID to the devices in step's
cgroup. Update the test to reflect this.

Bug 7094

Bug 7265.

No need to call exit() in main()

Bug 7265.

Fix style of if(var == SLURM_SUCCESS).

Bug 7265.

is assigned to job.

Factor --gpus-per-task(gres_per_task) in --gpus (gres_per_job) count

Logic of _eval_nodes has appropriate handling for per job gpus, which
should be used also in case of GRES specified by --gpus-per-task. This
can be achieved by properly initializing gres_per_job if by
gres_per_task * num_tasks.

Bug 8133.

Bug 7621

Make slurmrestd compile by default if http-parser libraries are
available, or break the build if explicitly enabled and http-parser
is missing.

If there are any environmental varibales that start with SLURMD, then
this test have a false failure.

Bug 7665.

Bug 7416

I also added a cluster name there to make it only generate this on the
cluster you are using otherwise you will generate this account on
other clusters in the dbd making a potential impossible to remove account
scenario since your user might only have an account on your cluster and
not the others.