Merge branch 'main' into feat/compute-volume-type

Author: sjpb

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

@@ -51,31 +51,36 @@ openhpc_state_save_location: "{{ appliances_state_dir + '/slurmctld' if applianc
 # NB: override in environments/site/inventory/group_vars/all/openhpc.yml, not here:
 openhpc_config_extra: {}
 
-# default additional slurm.conf parameters for the appliance:
+# additional default slurm.conf parameters for the appliance:
 openhpc_config_default:
   TaskPlugin: task/cgroup,task/affinity
   ReturnToService: 2 # workaround for templating bug TODO: Remove once on stackhpc.openhpc v1.2.0
-  TopologyPlugin: "topology/{{ 'tree' if (topology_nodes | length) > 0 else 'flat' }}"
 
-# default additional slurm.conf parameters when "rebuild" enabled:
+# additional default slurm.conf parameters when "rebuild" enabled:
 openhpc_config_rebuild:
   RebootProgram: /opt/slurm-tools/bin/slurm-openstack-rebuild
   SlurmctldParameters:
     - reboot_from_controller
   ResumeTimeout: 300
 
-# default additional slurm.conf parameters when "nhc" enabled:
+# additional default slurm.conf parameters when "nhc" enabled:
 openhpc_config_nhc:
   HealthCheckProgram: /usr/sbin/nhc
   HealthCheckInterval: 300
   HealthCheckNodeState: NONDRAINED_IDLE
 
+# additional default slurm.conf parameters when "topology" enabled:
+openhpc_config_topology:
+  TopologyPlugin: topology/tree
+
 # indirection to allow automatic construction of slurm.conf parameters:
 openhpc_config_groups:
   - enabled: "{{ groups['rebuild'] | length > 0 }}"
     config: "{{ openhpc_config_rebuild }}"
   - enabled: "{{ groups['nhc'] | length > 0 }}"
     config: "{{ openhpc_config_nhc }}"
+  - enabled: "{{ groups['topology'] | length > 0 }}"
+    config: "{{ openhpc_config_topology }}"
   - enabled: true
     config: "{{ openhpc_config_extra }}"
 

environments/common/inventory/groups

@@ -17,9 +17,14 @@ compute
 # Login group to use for running mpi-based testing.
 login
 
+[additional]
+# Additional nodes to include in "cluster" group
+# Automatically populated from OpenTofu variable additional_nodegroups
+
 [cluster:children]
 # All nodes in the appliance - add e.g. service nodes not running Slurm here.
 openhpc
+additional
 
 [builder]
 # Do not add hosts here manually - used as part of Packer image build pipeline. See packer/README.md.

environments/site/tofu/login.tf

@@ -40,7 +40,7 @@ module "login" {
   # not using openstack_compute_instance_v2.control.access_ip_v4 to avoid
   # updates to node metadata on deletion/recreation of the control node:
   control_address = openstack_networking_port_v2.control[var.cluster_networks[0].network].all_fixed_ips[0]
-  security_group_ids = [for o in data.openstack_networking_secgroup_v2.login: o.id]
+  security_group_ids = lookup(each.value, "security_group_ids", [for o in data.openstack_networking_secgroup_v2.login: o.id])
   baremetal_nodes = data.external.baremetal_nodes.result
 
   # input dict validation:
@@ -63,5 +63,6 @@ module "login" {
     "ip_addresses",
     "gateway_ip",
     "nodename_template",
+    "security_group_ids"
   ]
 }