Re: [slurm-users] [EXT] Re: How to find core count per job per node

[Ole Holm Nielsen]

Hi Tom,

I think that "pestat -j jobid" gives you the information you are asking 
for.  If not, please copy your exact output and explain why this isn't 
what you need.

Thanks,
Ole


On 21-10-2019 21:14, Tom Wurgler wrote:
Well, not really what I needed after all.

I have 24 core nodes. I submit a 36 way job then do a pestat :

pestat -j job1

Shows node1:  24
             node2:  12

Now submit another 36 way job.  It uses the other half of node2 and then 
also a node3

so pestat -j job1
             node1: 24
             node2: 24

and pestat -j job2
             node2: 24
             node3: 24

I'd like it to say:

pestat -j job1
             node1: 24
             node2: 12

and pestat -j job2
             node2: 12
             node3: 24

Does that make sense?

Thanks for any info.

tom

------------------------------------------------------------------------
*From:* slurm-users <slurm-users-boun...@lists.schedmd.com> on behalf of 
Ole Holm Nielsen <ole.h.niel...@fysik.dtu.dk>
*Sent:* Friday, October 18, 2019 2:15 PM
*To:* slurm-users@lists.schedmd.com <slurm-users@lists.schedmd.com>
*Subject:* [EXT] Re: [slurm-users] How to find core count per job per node
WARNING: This is an EXTERNAL email. Please think before RESPONDING or 
CLICKING on links/attachments.



On 18-10-2019 19:56, Tom Wurgler wrote:
I need to know how many cores a given job is using per node.
Say my nodes have 24 cores each and I run a 36 way job.
It take a node and a half.
scontrol show job id
shows me 36 cores, and the 2 nodes it is running on.
But I want to know how it split the job up between the nodes.

The "pestat" tool can tell you the CPUload of nodes belonging to a job:

pestat -j jobid

Get pestat from
https://nam05.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FOleHolmNielsen%2FSlurm_tools%2Ftree%2Fmaster%2Fpestat&amp;data=01%7C01%7Ctwurgl%40goodyear.com%7C331258e9a6114731131c08d753f75e7b%7C939e896692854a9a9f040887efe8aae0%7C0&amp;sdata=P%2BZUZsyZrjyGSQq52IzYZQL6g4JSJ8FAF1vnc8gHgQI%3D&amp;reserved=0

The "psjob" tool prints the processes on nodes of a given job when
executed on the control node:

psjob jobid

get psjob and other tools from
https://nam05.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2FOleHolmNielsen%2FSlurm_tools%2Ftree%2Fmaster%2Fjobs&amp;data=01%7C01%7Ctwurgl%40goodyear.com%7C331258e9a6114731131c08d753f75e7b%7C939e896692854a9a9f040887efe8aae0%7C0&amp;sdata=1ejEnGWdiLUY9csk%2FljtAoGJl3KkKNKnz%2BoSVqKkQ3c%3D&amp;reserved=0