get resolv_conf, etc_hosts and stackhpc.openhpc working

Author: sjpb

ansible/roles/compute_init/files/compute-init.yml

@@ -6,7 +6,9 @@
   vars:
     os_metadata: "{{ lookup('url', 'http://169.254.169.254/openstack/latest/meta_data.json') | from_json }}"
     server_node_ip: "{{ os_metadata.meta.k3s_server }}"
-    compute_groups: "{{ os_metadata.meta.compute_groups | default([]) }}"
+    enable_slurmd: "{{ os_metadata.meta.enable_slurmd | default(false) | bool }}"
+    enable_resolv_conf: "{{ os_metadata.meta.enable_senable_resolv_conf | default(false) | bool }}"
+    enable_etc_hosts: "{{ os_metadata.meta.enable_etc_hosts | default(false) | bool }}"
 
     # TODO: "= role defaults" - could be moved to a vars_file: on play with similar precedence effects
     # this is a good example: common environment actually defines this (non-functional w/o compute groups), but role default is empty
@@ -17,10 +19,10 @@
         - name: Report skipping initialization if not compute node
           # meta: end_play produces no output
           debug:
-            msg: "Skipping compute initialization as metadata compute_groups is empty"
+            msg: "Skipping compute initialization: Metadata enable_slurmd is not true"
 
         - meta: end_play
-      when: compute_groups | length == 0 
+      when: not enable_slurmd
 
     - name: Ensure the mount directory exists
       file:
@@ -37,16 +39,15 @@
         fstype: nfs
         opts: ro,sync
         state: mounted
-      register: nfs_mount_result
-      ignore_errors: true
       register: _mount_mnt_cluster
+      ignore_errors: true
       # TODO: add some retries here?
 
     - block:
         - name: Report skipping initialization if cannot mount nfs
           # meta: end_play produces no output
           debug:
-            msg: "Skipping compute initialization as cannot mount exports/cluster share"
+            msg: "Skipping compute initialization: Failed to mount /exports/cluster from control node {{ server_node_ip }}"
 
         - meta: end_play
       when: _mount_mnt_cluster.failed
@@ -56,6 +57,84 @@
       include_vars:
         file: "/mnt/cluster/hostvars/{{ ansible_hostname }}/hostvars.yml" # can't use inventory_hostname
 
-    - name: Demonstrate hostvars have loaded
-      debug:
-        var: prometheus_version
+      # TODO: should /mnt/cluster now be UNMOUNTED to avoid future hang-ups?
+
+    - name: Configure resolve.conf
+      block:
+        - name: Set nameservers in /etc/resolv.conf
+          ansible.builtin.template:
+            src: /etc/ansible-init/templates/resolv.conf.j2
+            dest: /etc/resolv.conf
+            owner: root
+            group: root
+            mode: u=rw,og=r
+
+        - name: Disable NetworkManager control of resolv.conf
+          ansible.builtin.copy:
+            src: /etc/ansible-init/files/NetworkManager-dns-none.conf
+            dest: /etc/NetworkManager/conf.d/90-dns-none.conf
+            owner: root
+            group: root
+            mode: u=rw,og=r
+          register: _copy_nm_config
+
+        - name: Reload NetworkManager
+          ansible.builtin.systemd:
+            name: NetworkManager
+            state: reloaded
+          when: _copy_nm_config.changed | default(false)
+      when: enable_resolv_conf
+
+    - name: Copy cluster /etc/hosts
+      copy:
+        src: /mnt/cluster/hosts
+        dest: /etc/hosts
+        owner: root
+        group: root
+        mode: 0644
+      when: enable_etc_hosts
+
+    # TODO: - name: NFS client mount
+
+    # TODO: - name: Manila mount
+
+    # TODO: - name: Basic users setup
+
+    # TODO: - name: Configure EESSI
+
+    # TODO: - name: Configure openhpc
+    # NB: don't need conditional block on enable_slurmd as have already exited
+    # if not the case
+    - name: Write Munge key
+      copy:
+        content: "{{ openhpc_munge_key }}"
+        dest: "/etc/munge/munge.key"
+        owner: munge
+        group: munge
+        mode: 0400
+
+    - name: Set slurmctld location for configless operation
+      lineinfile:
+        path: /etc/sysconfig/slurmd
+        line: "SLURMD_OPTIONS='--conf-server {{ server_node_ip }}'"
+        regexp: "^SLURMD_OPTIONS="
+        create: yes
+        owner: root
+        group: root
+        mode: 0644
+
+    - name: Ensure Munge service state
+      service:
+        name: munge
+        enabled: true
+        state: started
+
+    - name: Ensure slurmd service state
+      service:
+        name: slurmd
+        enabled: true
+        state: started
+
+    - name: Ensure node is resumed
+      # TODO: consider if this is always safe for all job states?
+      command: scontrol update state=resume nodename={{ ansible_hostname }}