Code calls list_find_first to search in resv_list whether the requested
name for the new reservation already exists. If it exists, resv_ptr is
set with the pointer to the existing reservation. Then the code goto
bad_parse label and xfreed that resv_ptr, thus corrupting the list data
by freeing the existing reservation. This is fixed by only freeing memory on the
new local resv_ptr instead of always freeing memory.  xfree is also not
sufficient for freeing the memory, we needed to call _del_resv_rec() or we would
leak the memory we had transferred from the resv_desc_ptr.  This also involved
NULLing out the other variables freed after bad_parse, or you would get
double frees.

Bug 3558.

assoc_mgr_make_tres_str_from_array.

array task id.

This is combination with the previous commit.

Bug 3563

I also made seff work on a non-standard perl install.

I would had loved to do this in multiple commits, but it wasn't that possible.

moment the dbd is assigning a db_index.

Bug 3512

This is a definite culprit of creating runaway jobs.

because the dbd is setting a db_index.

in the middle of setting db_indexes.

multifactor.  Almost impossible to make happen, but still a bug.

The stepd->protocol_version field is not set at this point,
if not handled here then the following call to stepd_get_uid
will fail as a protocol_version of zero is rejected.

This appeared to work in 16.05 and older as the stepd_get_uid()
API call used to ignore the invalid protocol_version and
return the correct result anyways, but that was fixed in c177ff95.

Bug .

correctly with Slurm tools.

What this does is replace the incoming string with the gres actually chosen.

Continuation of last commit.

Bug 3521

the database.

Symbol required for read_slurm_cgroup_conf() call to work.

Bug 3550.

burst_buffer/cray - Fix parsing for discontinuous allocated nodes. A job
    allocation of "20,22" must be expressed as "20\n22" when passed to
    the dw_wlm_cli program.
bugs 3540 and 3544

capmc_resume (Cray resume node script) - Do not disable changing a node's
    active features if SyscfgPath is configured in the knl.conf file.
bug 3533

If a job is cancelled by the user while it's allocated nodes are being
    reconfigured (i.e. the capmc_resume program is rebooting nodes for the job)
    and the node reconfiguration fails (i.e. the reboot fails), then don't
    requeue the job but leave it in a cancelled state. Note the JOB_RECONFIG_FAIL
    state flag is currently only used by capmc_resume, but could be used for
    other programs responsible for node reboots.
bug 3392

bug 3538

This is a partial reversion of commit 69684648
NOTE: sbatch does not support --cpu_bind (although the documentation does list
  the option) and the --mem_bind options set SBATCH_* environment variables
  that nothing every looks at. In other words, it needs some work.
Bugs 3519 and 3188

bug 3525

Copy NEWS item updates from v16.05 applied since v17.02.0 tag to
  to NEWS for v17.02.1

Convert a slurmctd power management data structure from array to list in
    order to eliminate the possibility of zombie child suspend/resume
    processes.
bug 3516

This now matches the behavior documented in sbatch.

This resolves a problem where the maximum cpu frequency would be set
to the minimum available on the node by the batch step. This is due
to the batch step leaving cpu_freq_{min,max,gov} uninitialized to zero,
which is then translated to a request to set the frequency to the lowest
available in the node. This did not impact 16.05 or earlier, as a request
for a zero frequency was ignored by a quirk of _cpu_freq_freqspec_num.
This quirk was removed by commit f40e1c01 before 17.02.0-rc1.

Bug 3510.

bug 3516

from 10 to 100.
bug 3516

information about a job to scontrol.

cause potential deadlock when/if TRES changed in the database and the
slurmctld wasn't made aware of the change.  This would be very rare.

The lock was originally there to keep new jobs from grabbing the assoc
information.  If the lock was done afterwards the worst case is we get the new
information.

It was determined we didn't need the write locks on the job and no locks were
needed on the node either.

Doing these different locked beforehand would create a window where you could
get a config write lock

This will be triggered after either a burst buffer job_begin function
  or select plugin job_begin function fails. Without this change, the
  "squeue -i" and "scontrol show job" commands can report old job
  state information.
bug 3504