When more than 1 ping cycle is spawned simultaneously (for instance
REQUEST_PING + REQUEST_NODE_REGISTRATION_STATUS for the selected nodes),
we do not track a separate ping_start time for each cycle. When ping_begin()
is called, the information about the previous ping cycle is lost. Then when
ping_end() is called for the first of the two cycles, we set ping_start=0,
which is incorrectly used to see if the last cycle ran for more than
PING_TIMEOUT seconds (100s), thus incorrectly triggering the:

 error("Node ping apparently hung, many nodes may be DOWN or configured "
       "SlurmdTimeout should be increased");

Bug 3914

Bug 3941.

Get rid of any race conditions and call anything that was in
_pre_task_privileged from the parent instead of the child.

NOTE: This should be safe as we don't execute the task until after
_exec_wait_child_wait_for_parent is signaled which happens after all this is
long over.

Bug 3865

Fix regression in commit e5c05549 that would put the stepd pid into the memory cgroup instead of the task's pid.

Beforehand this would put the result of getpid() into the cgroup.  Before
e5c05549 this was done in the child of the fork which would get you
the task's pid, but moving it to run in the parent broke this logic.

What this patch does is adds pid to the input parameters of
task_g_pre_launch_priv making it so we could use the correct pid.

Pretty much fix the entire purpose of this max_agent_queue.

Bug 3159

Bug 3956

Bug 3967

step wasn't always gathered correctly.

Bug 3531

Fix for possible slurmctld abort with use of salloc/sbatch/srun
    --gres-flags=enforce-binding option.
bug 4008

Update from commit b40bd8d3

Clarify --immediate option.

By removing the real locks we can get into a race condition where the prolog
starts and finishes before we get here and then we end up waiting forever.

Making the mutex a static seemed to help in many cases, but didn't
completely close the window.  Changing slurm_cond_wait to
slurm_cond_timedwait fixed the scenario where we would hit the window, but
not degrade performance the original commit provides.

There were also spots where if the job or step didn't exist it wouldn't
signal the conditional also providing a spot this could get stuck not
starting the job.

Fix regression from commit 52ce3ff0

Bug 3977

Code provided by Ole Nielsen <Ole.H.Nielsen@fysik.dtu.dk>

Bug 3985

This is a regression from commit fec995e0.

It turns out using tok here was erroneous for situations where the gres had
a type and name and potentially a count (i.e. network:gigabit:1)

_get_gres_req_cnt() would alter the incoming char *config which is what tok
was.  So when we print it back to the requested string it would only have
what was there to the first ':'.  As we didn't need to \0 out the first char
as we skip over it anyway I just kept track of what the replaced \0 was for
the number portion and put it back when we are done copying it.

Related to bug 3521

bug 3979

Bug 3967

Bug 3979 and 3989

This reverts commit d49081df.

Bug 3979 and 3989

will have a time displayed when truncating time.

Bug 3940.

Otherwise we can end up printing Start times greater than End times,
leading to confusion when reading sacct output. 0 is displayed as Unknown.
Cosmetic change.

Bug 3940.

This behavior was introduced in bug 2504, commit 7fb0c981 and bug 2643
commit 988edf12 respectively.

The reasoning is that sysadmins who see nodes with Reason "Not Responding"
but they can manually ping/access the node end up confused. That reason
should only be set if the node is trully not responding, but not if the
HealthCheckProgram execution failed or returned non-zero exit code. For
that case, the program itself would take the appropiate actions, such
as draining the node and setting an appropiate Reason.

Bug 3931

Bug 3963.

Bug 3938.