Merge branch 'master' of https://github.com/monash-merc/ansible_cluster_in_a_box into HEAD

roles/nfs-client/tasks/mountFileSystem.yml

@@ -4,7 +4,7 @@
   sudo: true
 
 - name: "Mounting NFS mounts"
-  mount: name={{ item.name }} src=" {{ item.ipv4 }}:{{ item.src }} " fstype={{ item.fstype }} opts={{ item.opts }} state=mounted
+  mount: name={{ item.name }} src="{{ item.ipv4 }}:{{ item.src }} " fstype={{ item.fstype }} opts={{ item.opts }} state=mounted
   with_items: nfsMounts
   notify: "restart rpcbind"
   notify: "restart idmap"

roles/slurm-from-source/tasks/installSlurmFromSource.yml

 - name: get slurm
-  shell: wget http://www.schedmd.com/download/archive/slurm-{{ slurm_version }}.tar.bz2
+  shell: wget https://cvl.massive.org.au/slurm-{{ slurm_version }}.tar.bz2
+#  shell: wget http://www.schedmd.com/download/archive/slurm-{{ slurm_version }}.tar.bz2
   args:
     chdir: /tmp
     creates: /tmp/slurm-{{ slurm_version }}.tar.bz2

roles/slurm-from-source/tasks/main.yml

@@ -43,6 +43,10 @@
   sudo: true
   when: slurmlogdir is defined 
 
+- name: create greps directory
+  file: path={{ slurm_dir }}/etc/gres state=directory owner=slurm group=slurm mode=755
+  sudo: true
+
 - name: install deps
   yum: name={{ item }} state=latest
   with_items:
@@ -107,11 +111,27 @@
 
 - include: installSlurmFromSource.yml
 
+- name: check slurm generic resource
+  shell: "{{ slurm_gres_check }}"
+  register: slurm_generic_resource
+  ignore_errors: true
+  when: slurm_gres_check is defined
+
+- name: install gres config file 
+  template: src=gres.conf.j2 dest={{ slurm_dir }}/etc/gres.conf mode=644
+  sudo: true
+  when: slurm_generic_resource.stdout 
+
+- name: install gres sub config file 
+  template: src=gres_sub.conf.j2 dest={{ slurm_dir }}/etc/gres/gres.conf mode=644
+  sudo: true
+  when: slurm_gres_list is defined
+
 - name: install slurm.conf
   template: src=slurm.conf.j2 dest={{ slurm_dir }}/etc/slurm.conf
   sudo: true
   notify: restart slurm
-  when: slurm_use_vpn==False
+  when: slurm_use_vpn==False and slurm_gres_list is defined
 
 - name: install slurm.conf
   template: src=slurm-vpn.conf.j2 dest={{ slurm_dir }}/etc/slurm.conf

roles/slurm-from-source/templates/slurm.conf.j2

@@ -141,9 +141,9 @@ MpiParams=ports=12000-12999
 {% endfor %}
 {% endfor %}
 {% for node in nodelist|unique %}
-NodeName={{ node }} Procs={{ hostvars[node]['ansible_processor_vcpus'] }} RealMemory={{ hostvars[node].ansible_memory_mb.real.total }} Sockets={{ hostvars[node]['ansible_processor_vcpus'] }} CoresPerSocket=1 ThreadsPerCore={{ hostvars[node].ansible_processor_threads_per_core }} {% if hostvars[node].ansible_hostname.find('vis') != -1 %}Gres=gpu{% if hostvars[node].ansible_hostname.find('k1') > 0 %}:k1m{% endif %}{% if hostvars[node].ansible_hostname.find('k2') > 0 %}:k2m{% endif %}:1{% endif %} {% if hostvars[node]['ansible_processor_vcpus'] == 1 %}Weight=1{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 1 and hostvars[node]['ansible_processor_vcpus'] <= 16 %}Weight=3{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 16 and hostvars[node]['ansible_processor_vcpus'] <= 20 %}Weight=5{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 20 and hostvars[node]['ansible_processor_vcpus'] <= 40 %}Weight=7{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 40 and hostvars[node]['ansible_processor_vcpus'] <= 64 %}Weight=8{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 64 and hostvars[node]['ansible_processor_vcpus'] <= 128 %}Weight=9{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 128 %}Weight=10{% endif %} Feature=stage1 State=UNKNOWN  
+NodeName={{ node }} Procs={{ hostvars[node]['ansible_processor_vcpus'] }} RealMemory={{ hostvars[node].ansible_memory_mb.real.total }} Sockets={{ hostvars[node]['ansible_processor_vcpus'] }} CoresPerSocket=1 ThreadsPerCore={{ hostvars[node].ansible_processor_threads_per_core }} {% if hostvars[node].ansible_hostname.find('vis') != -1 %}Gres=gpu{% if hostvars[node].ansible_hostname.find('k1') > 0 %}:k1{% endif %}{% if hostvars[node].ansible_hostname.find('k2') > 0 %}:k2{% endif %}:1{% endif %} {% if hostvars[node]['ansible_processor_vcpus'] == 1 %}Weight=1{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 1 and hostvars[node]['ansible_processor_vcpus'] <= 16 %}Weight=3{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 16 and hostvars[node]['ansible_processor_vcpus'] <= 20 %}Weight=5{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 20 and hostvars[node]['ansible_processor_vcpus'] <= 40 %}Weight=7{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 40 and hostvars[node]['ansible_processor_vcpus'] <= 64 %}Weight=8{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 64 and hostvars[node]['ansible_processor_vcpus'] <= 128 %}Weight=9{% endif %}{% if hostvars[node]['ansible_processor_vcpus'] > 128 %}Weight=10{% endif %} Feature=stage1 State=UNKNOWN  
 {% endfor %}
 
 {% for queue in slurmqueues %}
-PartitionName={{ queue.name }} {% if queue.default %}Default=yes{% endif %} Nodes={{ groups[queue.group]|join(',') }} DefMemPerCPU=1024 DefaultTime=24:00:00 State=UP 
+PartitionName={{ queue.name }} {% if queue.default %}Default=yes{% endif %} Nodes={{ groups[queue.group]|join(',') }} DefaultTime=24:00:00 State=UP 
 {% endfor %}