Bug 16047

These are a holdover from when bool was not standardized.

While there is a change from pack32() to packbool() here, this is happening
going from the slurmd to the slurmstepd during startup and should not pose a
problem. The only time this change could be an issue is if the slurmd and
slurmstepd versions are mismatched between rc1 and the next 23.02 release.

Bug 16038

Signed-off-by: Tim Wickberg <tim@schedmd.com>

Bug 14557

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

Bug 16043

This was only needed to get the gres_name from the gres_context (thus
the lock) and we already have that as part of the gres_state_t.

Bug 16043

$SLURM_JOBID and $SLURM_STEPID were being interpolated by the batch env
and leading to "scontrol show step <jobid>." because SLURM_STEPID
doesn't exist in the batch env. The variables needed to be escaped so
that the step's env would interpolate correctly.

Bug 16033

Avoid erroring out on request when old_name field is present but empty.

Bug 15407

Continuation of cf34100b.

Bug 15730

In this case we found job_ptr->details->ntasks_per_node was 0 which
would trigger a change in tasks even though tasks weren't changed ever.

Bug 16028

Bug 16028

Bug 15690

explicitly requested.

i.e.
sbatch -N2 --ntasks-per-node=1 --wrap="srun -N1 -v env | grep SLURM_STEP"

Before this patch the srun would run 2 tasks instead of just 1.

Bug 15690

Bug 15690

Required by route/topology, and to fix test7.12. These need to be visible
so that non-libslurmfull-linked processes expose the real variables,
otherwise the plugin is seeing the local (uninitialized and incorrect)
definitions rather than the weak linked symbol.

node_record_count
active_node_record_count
node_record_table_ptr
switch_record_cnt
switch_levels
switch_record_table

Bug 14116

Since commit e652133c requesting gpus is mandatory for --cpus-per-cpu
like it is for --mem-per-gpu, or job is rejected.

Bug 15945

If you did something like

salloc --begin=tomorrow
salloc: Pending job allocation 10975

But then

scontrol update job=10975 taskspernode=4 numnodes=4 start=now

Before you would only get 12 tasks instead of the 16.

Bug 15690

The patch was not good enough to catch all cases. This catches all.

A simple reproducer was

salloc --ntasks-per-node=4

would not allocate more than 1 task per node.

Bug 15690

e.g. when not running with the slurmdbd

Bug 15989

After 07700e76 priority_p_init from priority/multifactor plugin
for configuration without slurmdbd deadlocks because of job_write_lock,
but in fact it can't happen at this point since it required job_list
to be initiated, which isn't fulfilled after the commit since we want
to initialize plugin before read_slurm_conf.

Introduce priority_g_recover to be used to synchronize priorities
for different plugins. Move logic dedicated to priority/basic from
sync_job_priorities to plugin.

Effectively move the boosting logic of priority/basic, which is
correct since we need this to happen only on slurmctld restart
(initialization of lowest_priority to TOP_PRIORITY) not on reconfigure.
Commit 8b14f388.

Bug 15989

Co-authored-by: Brian Christiansen <brian@schedmd.com>

Bug 15989.

This was also a compiler warning in clang.

Bug 16006

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

These were also compiler errors in clang.

Bug 16006

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

This reverts commit 942ddb79.

Use memory value from AllocTRES instead of ReqTRES to compute memory
efficiency.

Bug 15569

seff will now use the memory value in AllocTRES instead of ReqTRES.

This matches the behavior of: sacct -o reqtres

Bug 15569

CFLAGS wasn't being set back to normal.

Continuation of 2e7e3735

Continuation of 2e7e3735

We still have to send it to <= 22.05 since cray aries needs it in the
switch plug.

Bug 15973

Bug 15973

Bug 15973