From 2aee70cbfd4209cd79ef90c5a6634dc967f5007e Mon Sep 17 00:00:00 2001
From: Chris Hines <chris.hines@monash.edu>
Date: Thu, 23 Feb 2017 12:41:01 +1100
Subject: [PATCH] better template for gres.conf allowing for Grid K1 GPUS and
different CPU counts
---
roles/slurm-common/files/scripts/nvidia-probe.py | 2 +-
roles/slurm-common/tasks/main.yml | 5 +++++
roles/slurm-common/templates/gres.conf.j2 | 2 +-
3 files changed, 7 insertions(+), 2 deletions(-)
diff --git a/roles/slurm-common/files/scripts/nvidia-probe.py b/roles/slurm-common/files/scripts/nvidia-probe.py
index 4b3e93e1..ba23982e 100755
--- a/roles/slurm-common/files/scripts/nvidia-probe.py
+++ b/roles/slurm-common/files/scripts/nvidia-probe.py
@@ -35,7 +35,7 @@ try:
if not line :
break
#print "Line is ",line
- pe=re.compile('GPU\s*(\d*).*Tesla\s*(\S*)')
+ pe=re.compile('GPU\s+(\d*):\s+\S+\s+(\S*)')
m=pe.search(line)
if not m:
#print "No match found"
diff --git a/roles/slurm-common/tasks/main.yml b/roles/slurm-common/tasks/main.yml
index 2e414650..d99cb2ff 100644
--- a/roles/slurm-common/tasks/main.yml
+++ b/roles/slurm-common/tasks/main.yml
@@ -90,6 +90,11 @@
register: probeOutput
check_mode: no
+- name: get cpu count
+ shell: 'lscpu | grep "On-line CPU" | cut -f 2 -d ":" | sed "s/\ *//g"'
+ register: cpucount
+ check_mode: no
+
- name: "set nvidiaprobe slurm_gres_list"
set_fact: "slurm_gres_list={{ probeOutput.stdout }}"
diff --git a/roles/slurm-common/templates/gres.conf.j2 b/roles/slurm-common/templates/gres.conf.j2
index 24001d0d..9d13ec66 100644
--- a/roles/slurm-common/templates/gres.conf.j2
+++ b/roles/slurm-common/templates/gres.conf.j2
@@ -1,5 +1,5 @@
#slurm gres file for {{ ansible_hostname }}
#No Of Devices={{ slurm_gres_list | length }}
{% for gr in slurm_gres_list %}
-Name={{ gr.name }} Type={{ gr.type }} File={{ gr.file }} CPUs=0-23
+Name={{ gr.name }} Type={{ gr.type }} File={{ gr.file }} CPUs={{ cpucount.stdout }}
{% endfor %}
--
GitLab