Fix login-only nodes in configless mode

README.md

@@ -31,10 +31,12 @@ package in the image.
 
 `openhpc_slurmdbd_host`: Optional. Where to deploy slurmdbd if are using this role to deploy slurmdbd, otherwise where an existing slurmdbd is running. This should be the name of a host in your inventory. Set this to `none` to prevent the role from managing slurmdbd. Defaults to `openhpc_slurm_control_host`.
 
-`openhpc_slurm_configless`: Optional, default false. If True then slurm's ["configless" mode](https://slurm.schedmd.com/configless_slurm.html) is used. **NB: Requires Centos8/OpenHPC v2.**
+`openhpc_slurm_configless`: Optional, default false. If true then slurm's ["configless" mode](https://slurm.schedmd.com/configless_slurm.html) is used. **NB: Requires Centos8/OpenHPC v2.**
 
 `openhpc_munge_key_path`: Optional, default ''. Define a path for a local file containing a munge key to use, otherwise one will be generated on the slurm control node.
 
+`openhpc_login_only_nodes`: Optional. If using "configless" mode specify the name of an ansible group containing nodes which are login-only nodes (i.e. not also control nodes), if required. These nodes will run `slurmd` to contact the control node for config.
+
 ### slurm.conf
 
 `openhpc_slurm_partitions`: list of one or more slurm partitions.  Each partition may contain the following values:

defaults/main.yml

@@ -50,3 +50,4 @@ ohpc_release_repos:
   "8": "http://repos.openhpc.community/OpenHPC/2/CentOS_8/x86_64/ohpc-release-2-1.el8.x86_64.rpm" # ohpc v2 for Centos 8
 openhpc_slurm_configless: false
 openhpc_munge_key: ''
+openhpc_login_only_nodes: ''

molecule/README.md

@@ -15,6 +15,8 @@ test4  | 1            | N                       | 2x compute node, accounting en
 test5  | 1            | N                       | As for #1 but configless
 test6  | 1            | N                       | 0x compute nodes, configless
 test7  | 1            | N                       | 1x compute node, no login node, configless
+test8  | 1            | N                       | 2x compute node, 2x login-only nodes, configless
+test9  | 1            | N                       | As test8 but uses `--limit=testohpc-control,testohpc-compute-0` and checks login nodes still end up in slurm.conf
 
 # Local Installation & Running
 

molecule/test8/INSTALL.rst

@@ -0,0 +1,22 @@
+*******
+Docker driver installation guide
+*******
+
+Requirements
+============
+
+* Docker Engine
+
+Install
+=======
+
+Please refer to the `Virtual environment`_ documentation for installation best
+practices. If not using a virtual environment, please consider passing the
+widely recommended `'--user' flag`_ when invoking ``pip``.
+
+.. _Virtual environment: https://virtualenv.pypa.io/en/latest/
+.. _'--user' flag: https://packaging.python.org/tutorials/installing-packages/#installing-to-the-user-site
+
+.. code-block:: bash
+
+    $ python3 -m pip install 'molecule[docker]'

molecule/test8/converge.yml

@@ -0,0 +1,19 @@
+---
+- name: Converge
+  hosts: all
+  tasks:
+    - name: "Include ansible-role-openhpc"
+      include_role:
+        name: "ansible-role-openhpc/"
+      vars:
+        openhpc_enable:
+          control: "{{ inventory_hostname in groups['testohpc_control'] }}"
+          batch: "{{ inventory_hostname in groups['testohpc_compute'] }}"
+          runtime: true
+        openhpc_slurm_control_host: "{{ groups['testohpc_control'] | first }}"
+        openhpc_slurm_partitions:
+          - name: "compute"
+        openhpc_cluster_name: testohpc
+        openhpc_slurm_configless: true
+        openhpc_login_only_nodes: 'testohpc_login'
+

molecule/test8/molecule.yml

@@ -0,0 +1,79 @@
+---
+name: single partition, group is partition
+driver:
+  name: docker
+platforms:
+  - name: testohpc-control
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_control
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+
+  - name: testohpc-login-0
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_login
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+
+  - name: testohpc-login-1
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_login
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+
+  - name: testohpc-compute-0
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_compute
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+  - name: testohpc-compute-1
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_compute
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+provisioner:
+  name: ansible
+  # ansible_args:
+  #   - --limit=testohpc-control,testohpc-compute-0
+verifier:
+  name: ansible

molecule/test8/verify.yml

@@ -0,0 +1,12 @@
+---
+
+- name: Check slurm hostlist
+  hosts: testohpc_login # NB for this test this is 2x non-control nodes, so tests they can contact slurmctld too
+  tasks:
+  - name: Get slurm partition info
+    command: sinfo --noheader --format="%P,%a,%l,%D,%t,%N" # using --format ensures we control whitespace
+    register: sinfo
+  - name: 
+    assert:                        # PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
+      that: "sinfo.stdout_lines == ['compute*,up,60-00:00:00,2,idle,testohpc-compute-[0-1]']"
+      fail_msg: "FAILED - actual value: {{ sinfo.stdout_lines }}"

molecule/test9/INSTALL.rst

@@ -0,0 +1,22 @@
+*******
+Docker driver installation guide
+*******
+
+Requirements
+============
+
+* Docker Engine
+
+Install
+=======
+
+Please refer to the `Virtual environment`_ documentation for installation best
+practices. If not using a virtual environment, please consider passing the
+widely recommended `'--user' flag`_ when invoking ``pip``.
+
+.. _Virtual environment: https://virtualenv.pypa.io/en/latest/
+.. _'--user' flag: https://packaging.python.org/tutorials/installing-packages/#installing-to-the-user-site
+
+.. code-block:: bash
+
+    $ python3 -m pip install 'molecule[docker]'

molecule/test9/converge.yml

@@ -0,0 +1,19 @@
+---
+- name: Converge
+  hosts: all
+  tasks:
+    - name: "Include ansible-role-openhpc"
+      include_role:
+        name: "ansible-role-openhpc/"
+      vars:
+        openhpc_enable:
+          control: "{{ inventory_hostname in groups['testohpc_control'] }}"
+          batch: "{{ inventory_hostname in groups['testohpc_compute'] }}"
+          runtime: true
+        openhpc_slurm_control_host: "{{ groups['testohpc_control'] | first }}"
+        openhpc_slurm_partitions:
+          - name: "compute"
+        openhpc_cluster_name: testohpc
+        openhpc_slurm_configless: true
+        openhpc_login_only_nodes: 'testohpc_login'
+

molecule/test9/molecule.yml

@@ -0,0 +1,79 @@
+---
+name: single partition, group is partition
+driver:
+  name: docker
+platforms:
+  - name: testohpc-control
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_control
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+
+  - name: testohpc-login-0
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_login
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+
+  - name: testohpc-login-1
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_login
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+
+  - name: testohpc-compute-0
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_compute
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+  - name: testohpc-compute-1
+    image: ${MOLECULE_IMAGE}
+    pre_build_image: true
+    groups:
+      - testohpc_compute
+    command: /sbin/init
+    tmpfs:
+      - /run
+      - /tmp
+    volumes:
+      - /sys/fs/cgroup:/sys/fs/cgroup:ro
+    networks:
+      - name: net1
+provisioner:
+  name: ansible
+  ansible_args:
+    - --limit=testohpc-control,testohpc-compute-0
+verifier:
+  name: ansible

molecule/test9/verify.yml

@@ -0,0 +1,13 @@
+---
+
+- hosts: testohpc_control
+  tasks:
+  - name: Check both compute nodes are listed and compute-0 is up
+    shell: 'sinfo --noheader --Node --format="%P,%a,%l,%D,%t,%N"' # using --format ensures we control whitespace
+    register: sinfo
+  - assert:                        # PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
+      that: "sinfo.stdout_lines == ['compute*,up,60-00:00:00,1,idle,testohpc-compute-0','compute*,up,60-00:00:00,1,unk*,testohpc-compute-1']" # NB: -1 goes 'down' after a while!
+      fail_msg: "FAILED - actual value: {{ sinfo.stdout_lines }}"
+  - name: Check login nodes in config
+    command: "grep NodeName={{ item }} /etc/slurm/slurm.conf"
+    loop: "{{ groups['testohpc_login'] }}"

tasks/main.yml

@@ -7,6 +7,11 @@
   when: "openhpc_enable.get(item, false)"
   tags: always
 
+- name: Set slurmd as service for openhpc_login_only_nodes
+  set_fact:
+    openhpc_slurm_service: "slurmd"
+  when: openhpc_login_only_nodes and (openhpc_login_only_nodes in group_names)
+
 - name: Install packages
   block:
     - include_tasks: install.yml

tasks/runtime.yml

@@ -75,7 +75,7 @@
     group: root
     mode: 0644
   when:
-    - openhpc_enable.batch | default(false) | bool
+    - openhpc_slurm_service == 'slurmd'
     - openhpc_slurm_configless
   notify:
     - Reload SLURM service

templates/slurm.conf.j2

@@ -137,6 +137,12 @@ PartitionName={{part.name}} \
         {% endfor %}{# group #}
 {% endfor %}{# partitions #}
 
+
+# Define slurmd nodes not in partitions for configless login-only nodes:
+{%if openhpc_login_only_nodes %}{% for node in groups[openhpc_login_only_nodes] %}
+NodeName={{ node }}
+{% endfor %}{% endif %}
+
 # Want nodes that drop out of SLURM's configuration to be automatically
 # returned to service when they come back.
 ReturnToService=2