From f2198631a157aec0667258d4da271fdb15beadc4 Mon Sep 17 00:00:00 2001
From: Simon Michnowicz <simon.michnowicz@monash.edu>
Date: Tue, 1 Aug 2017 14:59:21 +1000
Subject: [PATCH] change the way a script is called to determine number of GPUs
Unfortunately this new way also produced weird Python errors, which could
only be fixed by running as root (which is probably problem with original
way)
---
roles/slurm-common/tasks/main.yml | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/roles/slurm-common/tasks/main.yml b/roles/slurm-common/tasks/main.yml
index d99cb2ff..1f3c7e56 100644
--- a/roles/slurm-common/tasks/main.yml
+++ b/roles/slurm-common/tasks/main.yml
@@ -85,10 +85,15 @@
when: slurm_gres_check is defined
check_mode: no
+#as we change home directory earlier gather_facts causes the 'script' role to fail. so we do the probe this way
+- name: Copy Gres - Test script to /tmp
+ copy: src="scripts/nvidia-probe.py" dest="/tmp" mode="u=rwx,g=rx,o=rx"
+
- name: Gres - Test for Nvidia devices
- script: scripts/nvidia-probe.py
+ shell: /tmp/nvidia-probe.py
register: probeOutput
check_mode: no
+ sudo: true
- name: get cpu count
shell: 'lscpu | grep "On-line CPU" | cut -f 2 -d ":" | sed "s/\ *//g"'
@@ -98,6 +103,9 @@
- name: "set nvidiaprobe slurm_gres_list"
set_fact: "slurm_gres_list={{ probeOutput.stdout }}"
+- name: simon debug of slurm_gres_list
+ debug: msg="List of slurm_gres_list is {{ slurm_gres_list }}"
+
- name: template gres.conf file
template: src="gres.conf.j2" dest={{ slurm_dir }}/etc/gres.conf mode=644
sudo: true
--
GitLab