Merge branch 'ansible_check' into 'master'

Ansible check

See merge request !118

Former-commit-id: 1dadbbd1

roles/allow_stale_nfs/tasks/main.yml

@@ -8,6 +8,15 @@
   become: true
   become_user: root
 
+- name: remove old line
+  lineinfile:
+  args:
+    dest: "/etc/profile"
+    regexp: "^PATH=/usr/local/bin:/bin:/usr/bin$"
+    state: absent
+  become: true
+  become_user: root
+
 - name: remove /usr/local/ from the PATH in /etc/profile
   lineinfile:
   args:

roles/etcHosts/tasks/main.yml

@@ -2,18 +2,34 @@
   copy: src=files/etcHosts dest=/etc/hosts owner=root mode=644
   sudo: true
 
+- name: get hostname by sysctl
+  shell: sysctl kernel.hostname | cut -f 3 -d " "
+  register: sysctl_hostname
+  check_mode: no
+  become: true
+  become_user: root
+
 - name: set hostname by sysctl
   shell: sysctl kernel.hostname="{{ inventory_hostname }}"
   sudo: true
+  when: not "{{ sysctl_hostname.stdout }}" == "{{ inventory_hostname }}"
+
+- name: get domainname by sysctl
+  shell: sysctl kernel.domainname | cut -f 3 -d " "
+  register: sysctl_domainname
+  check_mode: no
+  become: true
+  become_user: root
 
 - name: set domainname by sysctl
   shell: sysctl kernel.domainname="{{ domain }}"
   sudo: true
+  when: not "{{ sysctl_domainname.stdout }}" == "{{ domain }}"
 
 - name: set /etc/sysconfig/network on CentOS 6
   lineinfile: dest=/etc/sysconfig/network line='HOSTNAME={{ inventory_hostname }}' regexp='^HOSTNAME'
   sudo: true
-  when: ansible_distribution == "CentOS"
+  when: ansible_distribution == "CentOS" and ansible_distribution_major_version == "6"
 
 - name: set /etc/sysctl.conf on Debian 8
   lineinfile: dest=/etc/sysctl.conf line='kernel.domainname = {{ domain }}' regexp='^#kernel.domainname'

roles/gluster_volcreate/tasks/main.yml

@@ -11,15 +11,6 @@
   become: true
   become_user: root
 
-- name: peer status
-  shell: "gluster peer status"
-  become: true
-  become_user: root
-  register: peer_status
-
-- name: debug peer status
-  debug: var=peer_status
-
 - name: create volume
   gluster_volume: 
     name: "{{ volname }}"

roles/gpu/tasks/main.yml

@@ -78,17 +78,17 @@
 
 - name: set uninstall default
   set_fact: 
-    install_driver: false
+    uninstall_driver: false
 
 - name: set install
   set_fact: 
     install_driver: true
-  when: not nvidia_driver.stat.exists or not installed_driver_version == nvidia_version
+  when: not nvidia_driver.stat.exists or not installed_driver_version.stdout == nvidia_version
 
 - name: set uninstall
   set_fact: 
     uninstall_driver: true
-  when: nvidia_driver.stat.exists and not installed_driver_version == nvidia_version
+  when: nvidia_driver.stat.exists and not installed_driver_version.stdout == nvidia_version
 
 - name: stop the persistence daemon
   service: name=nvidia-persistenced state=stopped
@@ -96,6 +96,12 @@
   become_user: root
   when: uninstall_driver
 
+- name: kill any X processes
+  shell:  ps ax | grep "X :0" | grep -v grep | cut -f 1 -d " " | xargs -I{} kill -9 {}
+  become: true
+  become_user: root
+  when: uninstall_driver
+
 - name: get nvidia driver 
   get_url: url=http://consistency0/src/NVIDIA-Linux-x86_64-{{ nvidia_version }}.run dest=/tmp/NVIDIA-Linux-x86_64-{{ nvidia_version }}.run
   become: true

roles/hpcsystems/tasks/main.yml

@@ -5,6 +5,7 @@
     - openssl-devel
     - openldap-devel
     - python-pip
+    - git
   become: true
   become_user: root
   when: ansible_os_family == "RedHat"
@@ -38,6 +39,7 @@
   git: 
     repo: git@gitlab.erc.monash.edu.au:hpc-team/hpcsystem.git
     dest: /usr/local/hpcsystem
+    accept_hostkey: True
 
 - name: install hpcsystem_config
   git: 

roles/ldapclient/tasks/configLdapClient.yml

@@ -25,12 +25,12 @@
   when: ldapCaCertContents is defined
 
 - name: "Copy system auth"
-  template: src=system-auth.j2 dest=/etc/pam.d/system-auth
+  template: src=system-auth.j2 dest=/etc/pam.d/system-auth-ac
   become: true
   become_user: root
 
 - name: "Copy password auth"
-  template: src=password-auth.j2 dest=/etc/pam.d/password-auth
+  template: src=password-auth.j2 dest=/etc/pam.d/password-auth-ac
   become: true
   become_user: root
 

roles/move_homedir/tasks/main.yml

@@ -2,11 +2,16 @@
   file: path=/local_home owner=root group=root state=directory
   sudo: true 
 
+- name: stat the local_home path
+  stat: path=/local_home/{{ ansible_user }}
+  register: local_home_path
+
 - name: copy the {{ ansible_user }} home
   shell: cp -ar /home/{{ ansible_user }} /local_home
   ignore_errors: true
   sudo: true
   register: home_copied
+  when: not local_home_path.stat.exists
 
 
 - name: edit passwd file

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"

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 }}"
 

roles/slurm-common/templates/gres.conf.j2

 #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 %}

roles/slurm-start/tasks/main.yml

@@ -19,6 +19,7 @@
   template: dest=/etc/systemd/system/slurmdbd.service src=slurmdbd.service.j2 mode=644
   sudo: true
   when: use_systemd is defined and start_slurmdbd is defined
+  register: slurmdbd_service_installed
 
 - name: copy slurm init script
   template: dest=/etc/init.d/slurm src=slurm.initd.j2  mode=755
@@ -29,16 +30,28 @@
   template: dest=/etc/systemd/system/slurmd.service src=slurmd.service.j2 mode=644
   sudo: true
   when: use_systemd is defined and start_slurmd is defined
+  register: slurmd_service_installed
 
 - name: slurmctld.service
   template: dest=/etc/systemd/system/slurmctld.service src=slurmctld.service.j2 mode=644
   sudo: true
   when: use_systemd is defined and start_slurmctld is defined
+  register: slurmctld_service_installed
 
 - name: reload systemd
   shell: systemctl daemon-reload
   sudo: true
-  when: use_systemd is defined
+  when: use_systemd is defined and start_slurmd is defined and slurmd_service_installed | changed
+
+- name: reload systemd
+  shell: systemctl daemon-reload
+  sudo: true
+  when: use_systemd is defined and start_slurmctld is defined and slurmctld_service_installed | changed
+
+- name: reload systemd
+  shell: systemctl daemon-reload
+  sudo: true
+  when: use_systemd is defined and start_slurmdbd is defined and slurmdbd_service_installed | changed
 
 - name: start munge
   service: name=munge state=started enabled=yes