Ticket 5690

Summary: cgroups with cpusets fails
Product: Slurm Reporter: Anthony DelSorbo <anthony.delsorbo>
Component: ConfigurationAssignee: Tim Wickberg <tim>
Status: RESOLVED INFOGIVEN QA Contact:
Severity: 3 - Medium Impact    
Priority: ---    
Version: 18.08.0   
Hardware: Linux   
OS: Linux   
Site: NOAA Alineos Sites: ---
Atos/Eviden Sites: --- Confidential Site: ---
Coreweave sites: --- Cray Sites: ---
DS9 clusters: --- HPCnow Sites: ---
HPE Sites: --- IBM Sites: ---
NOAA SIte: NESCC OCF Sites: ---
Recursion Pharma Sites: --- SFW Sites: ---
SNIC sites: --- Linux Distro: ---
Machine Name: Selene CLE Version:
Version Fixed: Target Release: ---
DevPrio: --- Emory-Cloud Sites: ---

Description Anthony DelSorbo 2018-09-09 13:02:24 MDT 
During a recent training session we reorganized our configuration files to arrive at a slurm configuration with the following setting in slurm.conf:

TaskPlugin=task/affinity,task/cgroup
TaskPluginParam=task/affinity,task/cgroup

this resulted in:

scontrol: error: Bad TaskPluginParam: task/affinity
scontrol: fatal: Unable to process configuration file

Assuming the configuration settings were in error, I changed this to:

TaskPlugin=task/affinity,task/cgroup
TaskPluginParam=Cores,Cpusets

While this cleared up the configuration error, I next ran into the following error on the compute node when running a simple job:

Sep  9 14:56:28 s0014 slurmstepd[81325]: Munge cryptographic signature plugin loaded
Sep  9 14:56:28 s0014 slurmstepd[81325]: error: slurm_build_cpuset: mkdir(/dev/cpuset/slurm2131): No such file or directory
Sep  9 14:56:28 s0014 slurmstepd[81325]: error: task_p_pre_setuid: slurm_build_cpuset() failed
Sep  9 14:56:28 s0014 slurmstepd[81325]: error: _spawn_job_container: Failed to invoke task plugins: one of task_p_pre_setuid functions returned error
Sep  9 14:56:28 s0014 slurmd[80884]: Launching batch job 2131 for UID 1209
Sep  9 14:56:28 s0014 slurmstepd[81331]: task affinity plugin loaded with CPU mask 0000000000...0ffffffffffff
Sep  9 14:56:28 s0014 slurmstepd[81331]: Munge cryptographic signature plugin loaded
Sep  9 14:56:28 s0014 slurmstepd[81331]: error: slurm_build_cpuset: mkdir(/dev/cpuset/slurm2131): No such file or directory
Sep  9 14:56:28 s0014 slurmstepd[81331]: error: task_p_pre_setuid: slurm_build_cpuset() failed
Sep  9 14:56:28 s0014 slurmstepd[81331]: error: Failed to invoke task plugins: one of task_p_pre_setuid functions returned error
Sep  9 14:56:28 s0014 slurmstepd[81331]: error: job_manager exiting abnormally, rc = 4020
Sep  9 14:56:28 s0014 slurmstepd[81331]: job 2131 completed with slurm_rc = 4020, job_rc = 0
Sep  9 14:56:28 s0014 slurmstepd[81331]: sending REQUEST_COMPLETE_BATCH_SCRIPT, error:4020 status 0
Sep  9 14:56:28 s0014 slurmd[80884]: error: task_p_slurmd_release_resources: rmdir(/dev/cpuset/slurm2131) failed No such file or directory

This, of course, results in the node being drained due to the error.

What is strange about this error is, since the compute node is a RHEL 7 system, there is no /dev/cpuset.  

Removing the "cpusets" setting from the configuration to:

TaskPluginParam=Cores

permits the job to succeed.

It's not clear to me how to proceed to resolve this issue.  Please advise.
Comment 1 Tim Wickberg 2018-09-09 13:07:21 MDT 
(In reply to Anthony DelSorbo from comment #0)
> During a recent training session we reorganized our configuration files to
> arrive at a slurm configuration with the following setting in slurm.conf:
> 
> TaskPlugin=task/affinity,task/cgroup
> TaskPluginParam=task/affinity,task/cgroup

The config I sent over to you had removed TaskPluginParam as those settings are not needed with TaskPlugin=task/cgroup added in now.

I'm not sure where you got that setting from?

> It's not clear to me how to proceed to resolve this issue.  Please advise.

Please delete the TaskPluginParam config line entirely.
Comment 2 Tim Wickberg 2018-09-10 16:35:43 MDT 
Updating to resolved/infogiven, please reopen if you have any further questions.

- Tim