[slurm-users] Issue with my hello mpi toy program

Mahmood Naderan mahmood.nt at gmail.com
Thu Oct 17 07:01:14 UTC 2019


Hi,
I used to run a hello mpi for testing purposes. Now, I see that it doesn't
work. While the log file shows memory allocation problem, squeue shows that
job is in R state endlessly.

[mahmood at hpc ~]$ cat slurm_script1.sh
#!/bin/bash
#SBATCH --job-name=hello_mpi
#SBATCH --output=hellompi.log
#SBATCH --ntasks=4
#SBATCH --time=10:00
#SBATCH --partition=SEA
#SBATCH --account=fish
#SBATCH --mem=100M
mpirun ./mpihello
[mahmood at hpc ~]$ sbatch slurm_script1.sh
Submitted batch job 10
[mahmood at hpc ~]$ cat hellompi.log
[hpc.safaar.com:18059] create_and_attach: unable to create shared memory
BTL coordinating structure :: size 134217728
--------------------------------------------------------------------------
A system call failed during shared memory initialization that should
not have.  It is likely that your MPI job will now either abort or
experience performance degradation.

  Local host:  hpc.safaar.com
  System call: mmap(2)
  Error:       Cannot allocate memory (errno 12)
--------------------------------------------------------------------------
[mahmood at hpc ~]$ squeue
             JOBID PARTITION     NAME     USER ST       TIME  NODES
NODELIST(REASON)
                10       SEA hello_mp  mahmood  R       0:36      1 hpc
[mahmood at hpc ~]$ squeue
             JOBID PARTITION     NAME     USER ST       TIME  NODES
NODELIST(REASON)
                10       SEA hello_mp  mahmood  R       0:43      1 hpc
[mahmood at hpc ~]$ sacctmgr list association
format=partition,account,user,grptres
 Partition    Account       User       GrpTRES
---------- ---------- ---------- -------------
                 root
                 root       root
                 fish
       sea       fish    mahmood cpu=10,mem=8G
                local    mahmood



However, the binary file works fine outside of slurm.

Regards,
Mahmood
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20191017/94e0791f/attachment.htm>


More information about the slurm-users mailing list