compute nodes must now be marked as available for topo-aware scheduling

Author: wtripp180901

ansible/slurm.yml

@@ -63,7 +63,9 @@
   tasks:
     - include_role:
         name: topology
-      when: appliances_mode == 'configure'
+      # Gated on topology group having compute nodes but role also
+      # needs to run on control and login nodes
+      when: appliances_mode == 'configure' and (groups['topology'] | length) > 0
     - include_role:
         name: stackhpc.openhpc
         tasks_from: "{{ 'runtime.yml' if appliances_mode == 'configure' else 'main.yml' }}"

environments/common/inventory/groups

@@ -24,6 +24,12 @@ openhpc
 [builder]
 # Do not add hosts here manually - used as part of Packer image build pipeline. See packer/README.md.
 
+[topology]
+# Compute nodes to be included in the Slurm topology plugin's topology tree
+# Should be set to `compute` if enabled
+# Note that this feature currently assumes all compute nodes are VMs, enabling
+# when the cluster contains baremetal compute nodes may lead to unexpected scheduling behaviour
+
 [podman:children]
 # Hosts running containers for below services:
 opensearch