Added topology aware scheduled for compute VMs

Author: wtripp180901

ansible/.gitignore

@@ -96,3 +96,5 @@ roles/*
 !roles/nhc/**
 !roles/eessi/
 !roles/eessi/**
+!roles/topology/
+!roles/topology/**

ansible/roles/topology/defaults/main.yml

@@ -0,0 +1,2 @@
+# If set to non-empty string, will override topology.conf file auto-detected from OpenStack project
+topology_topology_override: ""

ansible/roles/topology/library/map_hosts.py

@@ -0,0 +1,98 @@
+#!/usr/bin/python
+
+# Copyright: (c) 2025, StackHPC
+# Apache 2 License
+
+from ansible.module_utils.basic import AnsibleModule
+import openstack
+
+DOCUMENTATION = """
+---
+module: map_hosts
+short_description: Creates map of OpenStack VM network topology
+description:
+    - Creates map representing the network topology tree of an OpenStack project with a heirarchy
+      of: Availability Zone -> Hypervisors/Baremetal nodes -> VMs/Baremetal instances
+options:
+    compute_vms:
+        description:
+            - List of VM names within the target OpenStack project to include in the tree
+        required: true
+        type: str
+author:
+    - Steve Brasier, William Tripp, StackHPC
+"""
+
+RETURN = """
+topology:
+    description:
+      Map representing tree of project topology. Top level keys are AZ names, their values
+      are maps of shortened unique identifiers of hosts UUIDs to lists of VM names
+    returned: success
+    type: dict[str, dict[str,list[str]]]
+    sample:
+      "nova-az":
+        "afe9":
+          - "mycluster-compute-0"
+          - "mycluster-compute-1"
+        "00f9":
+          - "mycluster-compute-vm-on-other-hypervisor"
+"""
+
+EXAMPLES = """
+- name: Get topology map
+  map_hosts:
+    compute_vms:
+      - mycluster-compute-0
+      - mycluster-compute-1
+"""
+
+def min_prefix(uuids, start=4):
+    """ Take a list of uuids and return the smallest length >= start which keeps them unique """
+    for length in range(start, len(uuids[0])):
+        prefixes = set(uuid[:length] for uuid in uuids)
+        if len(prefixes) == len(uuids):
+            return length
+
+def run_module():
+    module_args = dict(
+        compute_vms=dict(type='list', elements='str', required=True)
+    )
+    module = AnsibleModule(argument_spec=module_args, supports_check_mode=True)
+
+    conn = openstack.connection.from_config()
+
+    servers = [s for s in conn.compute.servers() if s["name"] in module.params["compute_vms"]]
+
+    topo = {}
+    all_host_ids = []
+    for s in servers:
+        az = s['availability_zone']
+        host_id = s['host_id']
+        if host_id != '':
+            all_host_ids.append(host_id)
+            if az not in topo:
+                topo[az] = {}
+            if host_id not in topo[az]:
+                topo[az][host_id] = []
+            topo[az][host_id].append(s['name'])
+
+    uuid_len = min_prefix(list(set(all_host_ids)))
+
+    for az in topo:
+        topo[az] = dict((k[:uuid_len], v) for (k, v) in topo[az].items())
+
+    result = {
+        "changed": True, 
+        "topology": topo,
+    }
+    
+    module.exit_json(**result)
+
+
+def main():
+    run_module()
+
+
+if __name__ == "__main__":
+    main()

ansible/roles/topology/tasks/main.yml

@@ -0,0 +1,16 @@
+- name: Map instances to hosts
+  become: false
+  map_hosts:
+    compute_vms: "{{ groups['compute'] }}"
+  register: _topology
+  delegate_to: localhost
+  run_once: true
+
+- name: Template topology.conf
+  become: true
+  ansible.builtin.template:
+    src: templates/topology.conf.j2
+    dest: /etc/slurm/topology.conf
+    owner: root
+    group: root
+    mode: 0644

ansible/roles/topology/templates/topology.conf.j2

@@ -0,0 +1,13 @@
+# topology.conf
+# Switch Configuration
+{% if topology_topology_override != '' %}
+{{ topology_topology_override }}
+{% else %}
+{% for az in _topology.topology.keys() %}
+{% for instance_host in _topology.topology[az].keys() %}
+SwitchName={{ instance_host }} Nodes={{ _topology.topology[az][instance_host] | join(",") }}
+{% endfor %}
+SwitchName={{ az }} Switches={{ _topology.topology[az].keys() | join(",") }}
+{% endfor %}
+SwitchName=master Switches={{ _topology.topology.keys() | join(",") }}
+{% endif %}

ansible/slurm.yml

@@ -61,6 +61,9 @@
   tags:
     - openhpc
   tasks:
+    - include_role:
+        name: topology
+      when: appliances_mode == 'configure'
     - include_role:
         name: stackhpc.openhpc
         tasks_from: "{{ 'runtime.yml' if appliances_mode == 'configure' else 'main.yml' }}"

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

@@ -57,6 +57,7 @@ openhpc_config_default:
     - enable_configless
   TaskPlugin: task/cgroup,task/affinity
   ReturnToService: 2 # workaround for templating bug TODO: Remove once on stackhpc.openhpc v1.2.0
+  TopologyPlugin: topology/tree
 
 # default additional slurm.conf parameters when "rebuild" enabled:
 openhpc_config_rebuild: