Merge pull request #201 from stackhpc/fix/hpl-partitions

Fix hpctests HPL where cluster has more more than one partition

Author: sjpb

.github/workflows/stackhpc.yml

@@ -69,7 +69,9 @@ jobs:
           . venv/bin/activate
           . environments/${{ matrix.cloud }}/activate
           cd $APPLIANCES_ENVIRONMENT_ROOT/terraform
-          echo "::set-output name=messages::$(../../skeleton/\{\{cookiecutter.environment\}\}/terraform/getfaults.py  $PWD)"
+          TF_FAIL_MSGS="$(../../skeleton/\{\{cookiecutter.environment\}\}/terraform/getfaults.py  $PWD)"
+          echo $TF_FAIL_MSGS
+          echo "::set-output name=messages::${TF_FAIL_MSGS}"
         env:
           OS_CLOUD: openstack
           TF_VAR_cluster_name: ci${{ github.run_id }}

ansible/roles/hpctests/library/slurm_node_info.py

@@ -47,17 +47,16 @@ def run_module():
     if module.check_mode:
         module.exit_json(**result)
 
-    node_spec = ','.join(module.params['nodes'])
-    _, stdout,_ = module.run_command("sinfo --Format All --node %s" % node_spec, check_rc=True)
+    _, stdout,_ = module.run_command("sinfo --Format All --Node", check_rc=True) # `--nodes` doesn't filter enough, other partitions are still shown
     lines = stdout.splitlines()
-    # if len(lines) > 2:
-    #     raise ValueError('Info requested for nodes which are not homogenous: %s' % lines)
     info = {}
     params = [v.strip() for v in lines[0].split('|')]
     values = [line.split('|') for line in lines[1:]]
+    nodelist_ix = params.index('NODELIST')
     print(values)
     for ix, param in enumerate(params):
-        info[param] = [nodeinfo[ix].strip() for nodeinfo in values]
+        info[param] = [nodeinfo[ix].strip() for nodeinfo in values if nodeinfo[nodelist_ix].strip() in module.params['nodes']]
+        # info[param] = [nodeinfo[nodelist_ix] for nodeinfo in values]
     result['info'] = info
 
     module.exit_json(**result)

ansible/roles/hpctests/tasks/hpl-solo.yml

@@ -16,7 +16,7 @@
 - name: Check nodes are homogenous
   assert:
     that: "{{ hpctests_nodeinfo.info[item] | unique | length == 1 }}"
-    fail_msg: "Selected nodes are not homogenous: {{ item }} = {{ hpctests_nodeinfo.info[item] }}"
+    fail_msg: "Selected nodes are not homogenous: {{ item }} ({{ hpctests_nodeinfo.info['NODELIST'] }}) = {{ hpctests_nodeinfo.info[item] }}"
   loop:
     - SOCKETS
     - CORES

ansible/roles/hpctests/templates/hpl-build.sh.j2

@@ -4,7 +4,7 @@
 #SBATCH --output=%x.%a.out
 #SBATCH --error=%x.%a.out
 #SBATCH --exclusive
-{%if hpctests_nodes is defined %}#SBATCH --nodelist={{ hpctests_nodes }}{% endif %}
+{%if hpctests_nodes is defined %}#SBATCH --nodelist={{ hpctests_computes.stdout_lines[0] }}{% endif %}
 
 echo HPL arch: {{ hpctests_hpl_arch }}