If an update for a job that is being requeued[hold] happens, we don't want
to modify the start_time of the old record. Pending + Completing is
equivalent to Requeue.

Bug 4855.

Missed from 500e23b5

Bug 4988

job_ptr was still being referenced after releasing the read locks. Also
prolog_running_decr() needs to have a job write lock and a fed read
lock.

Bug 4947

Bug 4990.

While restarting slurmctld with -R flag, features were emptied.
This patch just does nothing when recover=2 (-R) which means active and
available features are preserved from the state files. Needed for
backup slurmctld controller.

Bug 4927

Bug 4882

If block_map_size is zero (as seen on FreeBSD when not using HWLOC),
xmalloc(0) will return NULL. Writing to str[0] then is invalid.

Bug 4845.

Bug 4971

-X doesn't show cumulative statistics for each job but statistics relevant to
the job allocation itself.

Bug 4967

It fixes a scontrol segfault when updating a reservation with empty users or
accounts and the parameters were specified more than once.

Bug 4945

Bug 4778

Bug 4949

Initialize cluster_name to NULL, allowing it to be safely
g_free'd later on.

Bug 4898.

Bug 3318

Use unpackstr_xmalloc_chooser as part of safe_unpackstr_xmalloc macro.
This ensures that the slurmdbd is always working on escaped strings.

Fixes CVE-2018-7033.

Bug 4792.

Reset the original time_limit value afterwards, otherwise the job
time may get squashed down to as low as 1 minute.

Bug 4796.

Bug 4881

Bug 4933

Bug 4878

into contribs directory.

This should expose O_CLOEXEC on older glibc versions,
addressing build issues on CLE 5.2.

Bug 4843.

MsgType=6500 wasn't being expanded:

"
slurm_persist_conn_open: Something happened with the
receiving/processing of the persistent connection init message to
XXX.XXX.XXX.XXX:6819: Failed to unpack MsgType=6500 message
"

Bug 4924

(slurmdbd was down when slurmctld was started).

Bug 4864

Bug 4888

Set scontrol exit code to 1 if attempting to update a node to
DRAIN or DOWN without specifying a reason.

Bug 4920.

when slurmctld was started) you could loose QOS usage information.

Bug 4865

6bc5dc30
6d4518fe

Looks like a regression accidentally introduced in d8770a66.

Bug 4769.

Bug 4549

Could happen if SuspendTimeout was shorter than ResumeTimeout.

Bug 4863

Continuation of 7d246784

Also throws spurious errors of:
"slurmd: error: Domain socket directory /var/spool/slurmd: No such file or directory"
if you SlurmdSpoolDir is located elsewhere.

Bug 4289.

Add additional protection in slurmctld as well.

Bug 4826.

job_limits_check() uses the job desc to call _valid_pn_min_mem(). This
second function might adjust the following values (up to date):

cpus_per_task
pn_min_memory
min_cpus
max_cpus
pn_min_cpus

If the function returns success, these adjusted members need to be copied
back to the job_record. It turns out pn_min_cpus wasn't copied back,
thus the logs claimed to automatically increase pn_min_cpus but actually
the job record wasn't modified and the select plugin tried to allocate
wrong amount of resources.

Bug 4823.

No longer needed, and will cause errors on FreeBSD systems
build with WITHOUT_KERBEROS.

Bug 4805.