[slurm-users] About specifying core distribution and overwriting when submitting a job

Hiromi Hayashi hiromi at hpc-sol.co.jp
Wed Jul 7 23:59:18 UTC 2021


Hi, all

I'd like to to change and use the default core distribution.
Environment: CentOS: 7.6 + Slurm: 20.11.3.
Cgroups are used to manage the core.

After changing the default core distribution, the -m option when 
submitting a job does not take effect.

before:
<slutmd.conf>
SelectType=select/cons_tres
SelectTypeParameters=CR_CORE_MEMORY

<result>
$ salloc -n 4
$ cat /sys/fs/cgroup/cpuset/slurm/uid_XXXX/job_xxx/cpuset.cpus
0,3,6,9
$ salloc -n 4 -m block:block
$ cat /sys/fs/cgroup/cpuset/slurm/uid_XXXX/job_xxx/cpuset.cpus
0-3
This is fine.

after:
<slutmd.conf>
SelectType=select/cons_tres
SelectTypeParameters=CR_CORE_MEMORY,CR_CORE_DEFAULT_DIST_BLOCK

<result>
$ salloc -n 4
$ cat /sys/fs/cgroup/cpuset/slurm/uid_XXXX/job_xxx/cpuset.cpus
0-3
$ salloc -n 4 -m block:cyclic
$ cat /sys/fs/cgroup/cpuset/slurm/uid_XXXX/job_xxx/cpuset.cpus
0-3  <-This is unexpected

I changed only add "CR_CORE_DEFAULT_DIST_BLOCK".
-m block: cyclic doesn't seem to work with CR_CORE_DEDAULT_DIST.

Is there any solution?

Regards,

Hiromi





More information about the slurm-users mailing list