fix nfs and make openhpc fully-capable in compute-init

Author: sjpb

ansible/roles/compute_init/README.md

@@ -35,9 +35,7 @@ The following roles are currently fully functional:
 - `resolv_conf`: all functionality
 - `etc_hosts`: all functionality
 - `nfs`: client functionality only
-- `stackhpc.openhpc`: all functionality, except that the control server name
-  must be the control node's `inventory_hostname`; `openhpc_slurm_control_host`
-  and `openhpc_slurm_control_host_address` are ignored.
+- `stackhpc.openhpc`: all functionality
 
 # Development/debugging
 
@@ -96,8 +94,8 @@ as in step 3.
      support certain subsets of role functionality or variables
      Examples: resolv_conf, stackhpc.openhpc
 
-- Some hostvars are tempalted from hostvars from other nodes, which aren't
-  available in the current approach:
+- Some variables are defined using hostvars from other nodes, which aren't
+  available v the current approach:
 
     ```
     [root@rl9-compute-0 rocky]# grep hostvars /mnt/cluster/hostvars/rl9-compute-0/hostvars.yml 
@@ -116,8 +114,12 @@ as in step 3.
     More generally, there is nothing to stop any group var depending on a
     "{{ hostvars[] }}" interpolation ...
 
-    Currently, this has been worked around for the following cases:
-    - The inventory hostname for the control node, indirected via `.api_address`
-      in the above hostvars. This is needed for the default nfs configuration
-      and the slurmctld namne. For compute-init this has been Defined using
-      "{{ groups['control'] | first }}" as the hostvars do include the groups.
+    Only `nfs_server_default` and `openhpc_slurm_control_host` are of concern
+    for compute nodes - both of these indirect via `api_address` to
+    `inventory_hostname`. This has been worked around by replacing this with
+    "{{ groups['control'] | first }}" which does result in the control node
+    inventory hostname when templating.
+
+    Note that although `groups` is defined in the templated hostvars, when
+    the hostvars are loaded using `include_vars:` is is ignored as it is a
+    "magic variable" determined by ansible itself and cannot be set.

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

@@ -9,9 +9,9 @@
     enable_compute: "{{ os_metadata.meta.enable_compute | default(false) | bool }}"
     enable_resolv_conf: "{{ os_metadata.meta.enable_resolv_conf | default(false) | bool }}"
     enable_etc_hosts: "{{ os_metadata.meta.enable_etc_hosts | default(false) | bool }}"
+    enable_nfs: "{{ os_metadata.meta.enable_nfs | 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
     resolv_conf_nameservers: []
 
     nfs_client_mnt_point: "/mnt"
@@ -20,9 +20,8 @@
     nfs_configurations:
     nfs_enable:
       clients: false
-    nfs_enable:
-      server: false
-      clients: false
+
+    # openhpc: no defaults required
 
   tasks:
     - block:
@@ -106,8 +105,10 @@
 
     # TODO: - name: NFS client mount
     - name: If nfs-clients is present
-      include_tasks: nfs-clients.yml
-      when: nfs_enable.clients | bool or ('nfs_enable' in item and item.nfs_enable.clients | bool)
+      include_tasks: ../tasks/nfs-clients.yml
+      when:
+        - enable_nfs
+        - nfs_enable.clients | bool or ('nfs_enable' in item and item.nfs_enable.clients | bool)
       loop: "{{ nfs_configurations }}"
 
     # TODO: - name: Manila mount
@@ -130,7 +131,7 @@
     - name: Set slurmctld location for configless operation
       lineinfile:
         path: /etc/sysconfig/slurmd
-        line: "SLURMD_OPTIONS='--conf-server {{ groups['control'] | first }}'"
+        line: "SLURMD_OPTIONS='--conf-server {{ openhpc_slurm_control_host_address | default(openhpc_slurm_control_host) }}'"
         regexp: "^SLURMD_OPTIONS="
         create: yes
         owner: root
@@ -152,3 +153,4 @@
     - name: Ensure node is resumed
       # TODO: consider if this is always safe for all job states?
       command: scontrol update state=resume nodename={{ ansible_hostname }}
+      # TODO: make safe for repeated runs

ansible/roles/compute_init/tasks/install.yml

@@ -12,7 +12,7 @@
     - files
     - library
     - filter_plugins
-    - playbooks
+    - tasks
 
 - name: Inject files from roles
   copy:
@@ -35,7 +35,7 @@
     - src: ../../basic_users/filter_plugins/filter_keys.py
       dest: filter_plugins/filter_keys.py
     - src: ../../stackhpc.nfs/tasks/nfs-clients.yml
-      dest: playbooks/nfs-clients.yml
+      dest: tasks/nfs-clients.yml
 
 - name: Add filter_plugins to ansible.cfg
   lineinfile:
@@ -52,4 +52,4 @@
     dest: /etc/ansible-init/playbooks/1-compute-init.yml
     owner: root
     group: root
-    mode: 0644
+    mode: 0644

environments/common/inventory/group_vars/all/nfs.yml

@@ -3,7 +3,7 @@
 # See: https://github.com/stackhpc/ansible-role-cluster-nfs
 # for variable definitions
 
-nfs_server_default: "{{ groups['control'] | first }}" # avoid using hostvars so
+nfs_server_default: "{{ groups['control'] | first }}" # avoid using hostvars for compute-init
 
 nfs_configurations:
   - comment: Export /exports/home from Slurm control node as /home

environments/common/inventory/group_vars/all/openhpc.yml

@@ -13,7 +13,7 @@ openhpc_slurm_accounting_storage_type: 'accounting_storage/slurmdbd'
 openhpc_slurmdbd_mysql_database: slurm_acct_db
 openhpc_slurmdbd_mysql_password: "{{ vault_mysql_slurm_password }}"
 openhpc_slurmdbd_mysql_username: slurm
-openhpc_slurm_control_host: "{{ hostvars[groups['control'].0].api_address }}"
+openhpc_slurm_control_host: "{{ groups['control'] | first }}" # avoid using hostvars for compute-init
 openhpc_slurmdbd_host: "{{ openhpc_slurm_control_host }}"
 openhpc_slurm_partitions:
   - name: "compute"