Change how Slurm determines the NUMA count of a node. Ignore KNL NUMA
    that only include memory.
bug 2745

Need thread_id to distinguish between multiple threads with the same
name.

and the slurmctld doesn't wait long enough for the response it would give
up leaving the connection open and create a situation where the next message
sent could receive the response of the first one.

Bug 2739

Correct logic that calculates a step's cpus_per_task allocation
on a heterogenous job allocation. Mixing a KNL with a Xeon resulted
in a count that was between the CPU count on the two node types
and invalid on the node with smaller CPU count (e.g. 272 CPUs on
KNL, 8 on Xeon, and 2 tasks, cpus_per_task = 140).

Bug #2713.

Bug #2713.

Update slurm.spec file to have seff depend on slurm-perlapi.

when in use.

The problem here is the polling threads in the various acct_gather codes
were detached and could possibly still be polling after the plugin had
been unloaded making a seg fault with a backtrace like this...

#0  0x00007fe7af008c00 in ?? ()
#1  0x00007fe7b1138479 in __nptl_deallocate_tsd () at pthread_create.c:175
#2  0x00007fe7b11398b0 in __nptl_deallocate_tsd () at pthread_create.c:326
#3  start_thread (arg=0x7fe7b1f12700) at pthread_create.c:346
#4  0x00007fe7b0e6fb5d in clone ()
    at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109

The fix was to make the threads non-detached and join them before calling
a dlclose.

Whenever possible, avoid allocating nodes that require a reboot.
Previous logic failed to re-sort the job set table based upon
the need for rebooting to achieve the desired features (e.g. KNL
MCDRAM or CACHE mode).
bug 2726

trying to verify the cluster name (which may try to /create/ files or
directories) *before* dropping privs results in a fatal error as
slurmctld tries to create items which ultimately fail.  Moving
this process until after the privs and uid have changed allows
the process to succeed.

Reported by Jon Nelson <jdnelson@dyn.com>

Bug 2728

Reject invalid step at submit time rather than leaving it queued.
Bug 2722 describes one of the use cases triggering the bug.

This partially restores commit 03b2cfb5
Logic was not closing file descriptor, which left the file locked and
leaked an open file descriptor.

tasks-per-node/nodes != tasks print warning and ignore
ntasks-per-node.

Bug 2520

make it to the slurmctld when using message aggregation.

    for better robustness.  This cray/select plugin code has
    been modified to remove a possible timing window where two
    aeld pthreads could exist, interfering with each other through
    the global aeld_running variable.

    An additional validity check has been added to the data provided
    to aeld through an alpsc_ev_set_application_info() call.
    If an error is returned from that call, only certain errors
    need the current socket connection closed to aeld and a new
    connection established.  Other error returns will log an
    error message and keep the current session established with
    aeld.

slurm.conf instead of all.  If looking for specific addresses use
TopologyParam options No*InAddrAny.

This was broken in 15.08 with the advent of the referenced TopologyParams
the commits 9378f195 and c5312f52 are no longer needed.

Bug 2696

at the same time.

Bug 2683

Turns out making a variable static in a function will make it not safe
when dealing with threads.

If node_feature/knl_cray plugin is configured and a GresType of "hbm"
is not defined, then add it the the GRES tables. Without this, references
to a GRES of "hbm" (either by a user or Slurm's internal logic) will
generate error messages.
bug 2708

With slurm-15.08.10, we're seeing occasional segfaults in slurmstepd. The logs point to the following line: slurm-15.08.10/src/slurmd/slurmstepd/mgr.c:2612

On that line, _get_primary_group() is accessing the results of getpwnam_r():
    *gid = pwd0->pw_gid;

If getpwnam_r() cannot find a matching password record, it will set the result (pwd0) to NULL, but still return 0. When the pointer is accessed, it will cause a segfault.

Checking the result variable (pwd0) to determine success should fix the issue.

I would like to mention a problem which seems to be a calculation bug of
used_cores in slurm version 15.08.7

If a node is divided into 2 partitions using MaxCPUsPerNode like this
slurm.conf configuration

    NodeName=n1 CPUs=20
    PartitionName=cpu NodeName=n1    MaxCPUsPerNode=16
    PartitionName=gpu NodeName=n1    MaxCPUsPerNode=4

I run into a strange scheduling situation.
The situation occurs after a fresh restart of the slurmctld daemon.

I start jobs one by one:

case 1
    systemctl restart slurmctld.service
    sbatch -n 16 -p cpu cpu.sh
    sbatch -n 1  -p gpu gpu.sh
    sbatch -n 1  -p gpu gpu.sh
    sbatch -n 1  -p gpu gpu.sh
    sbatch -n 1  -p gpu gpu.sh

    => Problem now: The gpu jobs are kept in PENDING state.

This picture changes if I start the jobs this way

case 2
    systemctl restart slurmctld.service
    sbatch -n 1  -p gpu gpu.sh
    scancel <gpu job_id>
    sbatch -n 16 -p cpu cpu.sh
    sbatch -n 1  -p gpu gpu.sh
    sbatch -n 1  -p gpu gpu.sh
    sbatch -n 1  -p gpu gpu.sh
    sbatch -n 1  -p gpu gpu.sh

and all jobs are running fine.

By looking into the code I figured out a wrong calculation of 'used_cores' in
function _allocate_sc()

plugins/select/cons_res/job_test.c

_allocate_sc(...)
...
         for (c = core_begin; c < core_end; c++) {
                 i = (uint16_t) (c - core_begin) / cores_per_socket;

                 if (bit_test(core_map, c)) {
                         free_cores[i]++;
                         free_core_count++;
                 } else {
                         used_cores[i]++;
                 }
                 if (part_core_map && bit_test(part_core_map, c))
                         used_cpu_array[i]++;

This part of code seems to work only if the part_core_map exists for a
partition or on a completly free node. But in case 1 there is no
part_core_map for gpu created yet. Starting a gpu  the core_map contains
4 cores left from the cpu job. Now all non free cores of the cpu partion
are counted as used cores in the gpu partition and this condition will
match in the next code parts

    free_cpu_count + used_cpu_count >  job_ptr->part_ptr->max_cpus_per_node

what is definitely wrong.

As soon as a part_core_map appears, means a gpu job was started on a free
node (case 2) then there is no problem at all.

To get case 1 work I changed the above code to the following and all works
fine:

         for (c = core_begin; c < core_end; c++) {
                 i = (uint16_t) (c - core_begin) / cores_per_socket;

                if (bit_test(core_map, c)) {
                         free_cores[i]++;
                         free_core_count++;
                 } else {
                     if (part_core_map && bit_test(part_core_map, c)){
                         used_cpu_array[i]++;
                         used_cores[i]++;
                     }
                 }

I am not sure this code change is really good, but it fixes my problem.

Do not attempt to power down a node which has never responded if the
    slurmctld daemon restarts without state.
bug 2698

they are in a step.

1) step_ptr->step_layout has already been dereferenced plenty of times.

2) Can't possible have rpc_version >= MIN_PROTOCOL_VERSION and < 8,
   this code is dead.

Issue the "node_reinit" command on all nodes identified in a single
   call to capmc. Only if that fails will individual nodes be
   restarted using multiple pthreads. This improves efficiency
   while retaining the ability to operate on individual nodes
   when some failure occurs.
bug 2659