support nfs for compute-init

Author: sjpb

ansible/roles/compute_init/README.md

@@ -32,9 +32,12 @@ to configure the node before the services on the control node are available
 (which requires running the site.yml playbook).
 
 The following roles are currently fully functional:
-- `resolv_conf`
-- `etc_hosts`
-- `stackhpc.openhpc`
+- `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.
 
 # Development/debugging
 
@@ -113,7 +116,8 @@ as in step 3.
     More generally, there is nothing to stop any group var depending on a
     "{{ hostvars[] }}" interpolation ...
 
-    Currently, the only functionality this has been problematic for is setting
-    the control node address for the slurmd node, which has been done using
-    the (metadata-provided) IP, given this is needed to do the NFS mount anyway
-    in the absence of working internal DNS.
+    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.

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

@@ -13,7 +13,17 @@
     # 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"
+    nfs_client_mnt_options:
+    nfs_client_mnt_state: mounted
+    nfs_configurations:
+    nfs_enable:
+      clients: false
+    nfs_enable:
+      server: false
+      clients: false
+
   tasks:
     - block:
         - name: Report skipping initialization if not compute node
@@ -95,6 +105,10 @@
       when: enable_etc_hosts
 
     # 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)
+      loop: "{{ nfs_configurations }}"
 
     # TODO: - name: Manila mount
 
@@ -116,7 +130,7 @@
     - name: Set slurmctld location for configless operation
       lineinfile:
         path: /etc/sysconfig/slurmd
-        line: "SLURMD_OPTIONS='--conf-server {{ server_node_ip }}'"
+        line: "SLURMD_OPTIONS='--conf-server {{ groups['control'] | first }}'"
         regexp: "^SLURMD_OPTIONS="
         create: yes
         owner: root

ansible/roles/compute_init/tasks/install.yml

@@ -12,6 +12,7 @@
     - files
     - library
     - filter_plugins
+    - playbooks
 
 - name: Inject files from roles
   copy:
@@ -33,7 +34,9 @@
       dest: library/os_manila_share.py
     - 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
+
 - name: Add filter_plugins to ansible.cfg
   lineinfile:
     path: /etc/ansible-init/ansible.cfg

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: "{{ hostvars[groups['control'] | first ].internal_address }}"
+nfs_server_default: "{{ groups['control'] | first }}" # avoid using hostvars so
 
 nfs_configurations:
   - comment: Export /exports/home from Slurm control node as /home