Merge pull request #2 from NREL/vtest-v1.157-cleanup

Removed duplicated config from vtest + fixed state directory being defined for all nodes

Author: sjpb

environments/devel/terraform/inventory.tpl

@@ -6,6 +6,9 @@ ansible_ssh_common_args='-o ProxyCommand="ssh rocky@${proxy_fip} -W %h:%p"'
 [control]
 ${control.name} ansible_host=${[for n in control.network: n.fixed_ip_v4 if n.access_network][0]} server_networks='${jsonencode({for net in control.network: net.name => [ net.fixed_ip_v4 ] })}'
 
+[control:vars]
+appliances_state_dir=/var/lib/state
+
 [admin]
 ${cluster_name}-devadmin
 

environments/lab/terraform/inventory.tpl

@@ -6,6 +6,9 @@ ansible_ssh_common_args='-o ProxyCommand="ssh rocky@${proxy_fip} -W %h:%p"'
 [control]
 ${control.name} ansible_host=${[for n in control.network: n.fixed_ip_v4 if n.access_network][0]} server_networks='${jsonencode({for net in control.network: net.name => [ net.fixed_ip_v4 ] })}'
 
+[control:vars]
+appliances_state_dir=/var/lib/state
+
 [login]
 %{ for login in logins ~}
 ${login.name} ansible_host=${[for n in login.network: n.fixed_ip_v4 if n.access_network][0]} server_networks='${jsonencode({for net in login.network: net.name => [ net.fixed_ip_v4 ] })}'

environments/nrel/hooks/pre.yml

@@ -1,12 +1,88 @@
 ---
 
+- name: "Add a stack admin user"
+  hosts: all
+  # tags: pre_tasks_all
+  become: true
+  gather_facts: true
+  tasks:
+    - name: Does localhome/stack dir exist
+      ansible.builtin.stat:
+        path: /localhome/stack
+      register: localstack
+
+    # - name: Does localhome/rocky dir exist
+    #   ansible.builtin.stat:
+    #     path: /localhome/rocky
+    #   register: localrockyuserdir
+
+    - name: Add stack user
+      ansible.builtin.include_role:
+        name: vs.core.stack_user
+      when: not localstack.stat.exists
+
+    # - name: Add rocky user role
+    #   ansible.builtin.include_role:
+    #     name: vshpc.rocky_user
+    #   when: not localrockyuserdir.stat.exists
+    #   tags:
+    #     - add_stack_user
+  tags:
+    - add_stack_user
+
+- name: A cp user home to /var/lib/USER
+  hosts: all
+  # tags: pre_tasks_all
+  become: true
+  gather_facts: false
+
+  # tasks:
+  #   - name: Does home/USER dir exist
+  #     ansible.builtin.stat:
+  #       path: "/home/{{ appliances_local_users_ansible_user_name }}"
+  #     register: homerocky
+
+  #   - name: Does var/lib/rocky dir exist
+  #     ansible.builtin.stat:
+  #       path: "/var/lib/{{ appliances_local_users_ansible_user_name }}"
+  #     register: varlibrocky
+
+  #   - name: Cp homerocky to var/lib/ dir
+  #     ansible.builtin.command: cp -a /home/{{ appliances_local_users_ansible_user_name }} /var/lib/{{ appliances_local_users_ansible_user_name }}
+  #     when:
+  #       - homerocky.stat.exists
+  #     register: cph
+  #     changed_when: cph.rc != 0
+
+  #   # - name: cp auth_keys
+  #   #   shell: |
+  #   #     cp -a /home/rocky/.ssh /var/lib/rocky/.ssh
+  #   #   when:
+  #   #     - varlibrocky.stat.exists
+  #   #     - homerocky.stat.exists
+
+  #   - name: Hack passwd file for localhome/rocky
+  #     ansible.builtin.lineinfile:
+  #       path: /etc/passwd
+  #       regexp: '^rocky.*1000.*'
+  #       line: "rocky:x:1000:1000:vsRockyUser:/var/lib/{{ appliances_local_users_ansible_user_name }}:/bin/bash"
+  # tags:
+  #   - rocky_localhome
+
 - name: "Do the preliminary node setups"
   hosts: all
   # tags: pre_tasks_all
   become: true
   tasks:
-    - include_role:
+    - name: Does vs_pre_complete.txt exist
+      ansible.builtin.stat:
+        path: /root/vs_pre_complete.txt
+      register: vs_pre_complete
+
+    - name: Pre tasks now
+      ansible.builtin.include_role:
         name: vshpc.prov.pre-tasks
+      when: not vs_pre_complete.stat.exists
 
 # TODO: KBENDL - check compatibility with new playbook
 - name: "NREL pre - Mount cephfs volumes"
@@ -16,80 +92,17 @@
     - aco.core.cephfs
   become: true
   tasks:
-    - include_role:
+    - name: Does ceph exist?
+      ansible.builtin.stat:
+        path: /etc/ceph
+      register: vs_does_cephdir
+    - name: CephFS now
+      ansible.builtin.include_role:
         name: aco.core.cephfs
       tags:
         - aco.core.cephfs
-
-    # - name: Set timezone to Americas/Denver
-    #   timezone:
-    #     name: America/Denver
-    #   tags:
-    #     - pre_set_time_on_servers
-
-    # - include_role:
-    #     name: aco.core.ntp
-    #   tags:
-    #     - pre_set_time_on_servers
-
-# - name: "NREL PRE - dnf packages"
-#   hosts: all
-#   tags: dnf_initial_update
-#   become: true
-#   tasks:
-#     - lineinfile:
-#         path: /etc/yum.repos.d/cuda-rhel8.repo
-#         regexp: '^enabled'
-#         line: enabled=0
-
-#     - name: dnf fix challenging dependency issues
-#       dnf:
-#         name: "*"
-#         nobest: true
-#         allowerasing: true
-#         skip_broken: true
-
-# - name: "NREL pre - set cluster root password"
-#   hosts: all
-#   tags: root_pass_cli
-#   become: true
-#   tasks:
-#     - name: include root if called
-#       include_role:
-#         name: root_pass_cli
-#       when: new_root_pass is defined
-
-# - name: "NREL pre - set admin hostnames to hpc.nrel.gov domain"
-#   hosts: control,login
-#   tags: set_hpc_hostname
-#   become: true
-#   tasks:
-#     - shell: "hostnamectl set-hostname {{ ansible_hostname }}.hpc.nrel.gov"
-
-# This may be fixed now
-# # ref: Error: cannot setup namespace using "/usr/bin/newuidmap": should have setuid or have filecaps setuid:
-# - name: "podman uid-gid workaround"
-#   hosts:
-#     - cluster
-#   become: true
-#   tags:
-#     - vs_pre
-#     - uid_gid_hack
-#   tasks:
-#     - name: "set newuidmap/newgidmap perms so containers will start"
-#       file:
-#         path: "{{ item }}"
-#         mode: '4755'
-#       loop:
-#         - /usr/bin/newgidmap
-#         - /usr/bin/newuidmap
-
-# TODO - get latest playbook
-# - name: Ensure yum repos are local
-#   hosts: all
-#   tags: ansible_yum_config
-#   become: true
-#   tasks:
-#     - include_role:
-#         name: ansible_yum_config
-# [email protected]:acoinf/vshpc.prov.pre-tasks.git
+      when: not vs_does_cephdir.stat.exists
+    - name: Tmp umount /home
+      ansible.builtin.shell: systemctl stop home.mount
+      when: not vs_does_cephdir.stat.exists
+      become: true

environments/nrel/inventory/group_vars/all/defaults.yml

@@ -44,7 +44,6 @@ openhpc_generic_packages:
   - munge
   - mariadb-connector-c # only actually needed on slurmdbd
   - hwloc-libs          # only actuall needed on slurmd
-
   # below added here to get pingpong and pingmatrix hpctests working:
   - mpitests-openmpi