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

@@ -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