The underlying problem is in the sched plugin logic in SLURM v2.4

but not node count the node count is correctly figured out.

Without this fix, gang scheduling mode could start without creating
a list resulting in an assert when jobs are submitted.

This change permits a user to get a zero size allocation by specifying
a task count of zero with no node count specification.

packing the step layout structure.

we must use a small block instead of a shared midplane block.

still messages we find when we poll but haven't given it back to the real
time yet.

I was doing some checking to find out why the the 2.4 branch and
master branch of schedmd was not allowing held jobs to be modified,
when attempting to do so,  scontrol would return:

slurm_update error: Requested partition configuration not available now

I did some debugging and found that it was caused by code added to the tail end
of job_limits_check() in job_mgr.c. It  had this addition:

        } else if (job_ptr->priority == 0) {   /* user or administrator hold */
                fail_reason = WAIT_HELD;
        }

It is causes all modifications done by scontrol on held jobs, to fail.

I'd like to propose quieting down the job_mgr a tad.  This is a refinement to:
https://github.com/SchedMD/slurm/commit/30a986f4c600291876f4ec3e3949934512f2cba5

a job kill timeout aren't always reported to the system.  This is now
handled by the runjob_mux plugin.

I'd like to add the following disclaimer to the documentation of the --mem option to the salloc/sbatch/srun commands. There is currently similar wording in the slurm.conf file, but I've received a bug report in which the memory limits were exceeded (until the next accounting poll).

NOTE: Enforcement of memory limits currently requires enabling of accounting,
which samples memory use on a periodic basis (data need not be stored,  just  collected).
A task may exceed the memory limit until the next periodic accounting sample.

Rod Schultz, Bull

sub-blocks.

to make a larger small block and are running with sub-blocks.

function didn't always work correctly.

rerun autogen.sh

the next step in the allocation only uses part of the allocation it gets
the correct cnodes.

In etc/init.d/slurm move check for scontrol after sourcing
/etc/sysconfig/slurm. Patch from Andy Wettstein, University of Chicago.

According to man slurm.conf, the default for NodeAddr is NodeName:

  "By  default, the NodeAddr will be identical in value to NodeName."

However, it seems the default is NodeHostname (when that differs from
NodeName): With the following in slurmnodes.conf:

Nodename=c0-0 NodeHostname=compute-0-0 ...

I get

NodeName=c0-0 Arch=x86_64 CoresPerSocket=2
   CPUAlloc=0 CPUErr=0 CPUTot=4 Features=intel,rack0,hugemem
   Gres=(null)
***
   NodeAddr=compute-0-0 NodeHostName=compute-0-0
***
   OS=Linux RealMemory=3949 Sockets=2
   State=IDLE ThreadsPerCore=1 TmpDisk=10000 Weight=1027
   BootTime=2012-05-08T15:07:08 SlurmdStartTime=2012-05-25T10:30:10

(This is with 2.4.0-0.pre4.)

(We are planning to use cx-y instead of compute-x-y (the rocks default)
on our next cluster, to save some typing.)

--
Regards,
Bjørn-Helge Mevik, dr. scient,
Research Computing Services, University of Oslo

This change makes the code consistent with the documentation.
Note that "bf_res=" will continue to be recognized for now.
Patch from Rod Schultz, Bull.