When using ThreadsPerCore > 1, it appears that DefMemPerCPU is being
scaled by slurmctld, but not by slurmd/slurmstepd.

For example, we set ThreadsPerCore=2 and DefMemPerCPU=100. Running a
single core job, we would expect two threads to be allocated and AllocMem
on the assigned node to increase by 200MB. scontrol reports that AllocMem
increased by 200MB, but the task/cgroup plugin only sees 100M of RAM.

It looks like the problem may lie in common/slurm_cred.c:format_core_allocs().
The function counts the job/step cores and multiplies the mem_limit's,
but it does not scale the CPU count like in slurmd/slurmd/req.c:_check_job_credential().
See bug 309

This is the correction of the slurm client which otherwise fails the
PMI2_Init() step. This hidden bug was introduced in the standalone
client made from BULL and David's modification made it occur : a wrong
size (by 1 too short) was passed to strncpy for keys, then David
replaced strncpy by MPICH's MPIU_Strncpy forcing a terminal '\0'.

Bug 359

messed up output.  Use the parsable options if you want this kind of
view.

Effects jobs with --exclusive and --cpus-per-task options
bug 355

A simple one-line fix to the "_adjust_cpus_nppcu" function that I had
added.  I had added this function as part of the NPPCU functionality;
however it wasn't a problem until that squeue patch.  That was because
then squeue had been updated to use this function and in this one case
the default value for the internal variable "ntasks_per_core" wound up
not being the 0xffff (65535) that I previously had coded for (as in
"select/cons_res") but instead was 0.  Therefore, in that adjustment
function of mine, I simply added a second clause to the if-statment
where I check for the sentinel value that also checks whether it is 0.
This resolved the problem.  Because we do not usually use
"select/serial", I did not notice this.

This can happen if something outside of Slurm opens the srun socket
and writes to it, since the data will not be of a form that Slurm
can decode.
Bug 354

Rather than a job

This removes logic added three years ago that would automatically set a job's cpus_per_task value in order to reset a job's mem_per_cpu value and scale the cpus_per_task by the same value. Equivalent logic did not exist in the step allocation logic. Just return an error instead. This change will be made in Slurm version 2.6, but this batch is made for version 2.5. The original patch introducing the problem is in commit: cc00cc70b9c90816afc511e0261e449857176332bug 352

enum.

Bug 351

This is extends the logic of commit
ba58d59c
to the following RPC types:
job complete
batch script complete and
job step complete

In acct_gather_energy/rapl initialization, if the fopen of /proc/cpuinfo fails we should treat this as a fatal condition rather than continue.  Patch is attached.

Problem found by Coverity tool, CID 20186
bug 331

This applies the same logic as added for job signal and batch job
submit as in commit ba58d59c